Merge pull request #488 from router-for-me/gemini

Unify the Gemini executor style

.gitignore

@@ -15,6 +15,7 @@ pgstore/*
 gitstore/*
 objectstore/*
 static/*
+refs/*
 
 # Authentication data
 auths/*
@@ -30,3 +31,7 @@ GEMINI.md
 .vscode/*
 .claude/*
 .serena/*
+
+# macOS
+.DS_Store
+._*

.goreleaser.yml

@@ -1,5 +1,7 @@
 builds:
   - id: "cli-proxy-api"
+    env:
+      - CGO_ENABLED=0
     goos:
       - linux
       - windows
@@ -34,4 +36,4 @@ changelog:
   filters:
     exclude:
       - '^docs:'
-      - '^test:'
+      - '^test:'

README.md

@@ -25,6 +25,7 @@ Get 10% OFF GLM CODING PLAN：https://z.ai/subscribe?ic=8JVLJQFSKB
 - Claude Code support via OAuth login
 - Qwen Code support via OAuth login
 - iFlow support via OAuth login
+- Amp CLI and IDE extensions support with provider routing
 - Streaming and non-streaming responses
 - Function calling/tools support
 - Multimodal input support (text and images)
@@ -48,6 +49,18 @@ CLIProxyAPI Guides: [https://help.router-for.me/](https://help.router-for.me/)
 
 see [MANAGEMENT_API.md](https://help.router-for.me/management/api)
 
+## Amp CLI Support
+
+CLIProxyAPI includes integrated support for [Amp CLI](https://ampcode.com) and Amp IDE extensions, enabling you to use your Google/ChatGPT/Claude OAuth subscriptions with Amp's coding tools:
+
+- Provider route aliases for Amp's API patterns (`/api/provider/{provider}/v1...`)
+- Management proxy for OAuth authentication and account features
+- Smart model fallback with automatic routing
+- **Model mapping** to route unavailable models to alternatives (e.g., `claude-opus-4.5` → `claude-sonnet-4`)
+- Security-first design with localhost-only management endpoints
+
+**→ [Complete Amp CLI Integration Guide](docs/amp-cli-integration.md)**
+
 ## SDK Docs
 
 - Usage: [docs/sdk-usage.md](docs/sdk-usage.md)
@@ -78,6 +91,14 @@ Native macOS menu bar app to use your Claude Code & ChatGPT subscriptions with A
 
 Browser-based tool to translate SRT subtitles using your Gemini subscription via CLIProxyAPI with automatic validation/error correction - no API keys needed
 
+### [CCS (Claude Code Switch)](https://github.com/kaitranntt/ccs)
+
+CLI wrapper for instant switching between multiple Claude accounts and alternative models (Gemini, Codex, Antigravity) via CLIProxyAPI OAuth - no API keys needed
+
+### [ProxyPal](https://github.com/heyhuynhgiabuu/proxypal)
+
+Native macOS GUI for managing CLIProxyAPI: configure providers, model mappings, and endpoints via OAuth - no API keys needed.
+
 > [!NOTE]  
 > If you developed a project based on CLIProxyAPI, please open a PR to add it to this list.
 

README_CN.md

@@ -48,6 +48,17 @@ CLIProxyAPI 用户手册： [https://help.router-for.me/](https://help.router-fo
 
 请参见 [MANAGEMENT_API_CN.md](https://help.router-for.me/cn/management/api)
 
+## Amp CLI 支持
+
+CLIProxyAPI 已内置对 [Amp CLI](https://ampcode.com) 和 Amp IDE 扩展的支持，可让你使用自己的 Google/ChatGPT/Claude OAuth 订阅来配合 Amp 编码工具：
+
+- 提供商路由别名，兼容 Amp 的 API 路径模式（`/api/provider/{provider}/v1...`）
+- 管理代理，处理 OAuth 认证和账号功能
+- 智能模型回退与自动路由
+- 以安全为先的设计，管理端点仅限 localhost
+
+**→ [Amp CLI 完整集成指南](docs/amp-cli-integration_CN.md)**
+
 ## SDK 文档
 
 - 使用文档：[docs/sdk-usage_CN.md](docs/sdk-usage_CN.md)
@@ -78,6 +89,14 @@ CLIProxyAPI 用户手册： [https://help.router-for.me/](https://help.router-fo
 
 一款基于浏览器的 SRT 字幕翻译工具，可通过 CLI 代理 API 使用您的 Gemini 订阅。内置自动验证与错误修正功能，无需 API 密钥。
 
+### [CCS (Claude Code Switch)](https://github.com/kaitranntt/ccs)
+
+CLI 封装器，用于通过 CLIProxyAPI OAuth 即时切换多个 Claude 账户和替代模型（Gemini, Codex, Antigravity），无需 API 密钥。
+
+### [ProxyPal](https://github.com/heyhuynhgiabuu/proxypal)
+
+基于 macOS 平台的原生 CLIProxyAPI GUI：配置供应商、模型映射以及OAuth端点，无需 API 密钥。
+
 > [!NOTE]  
 > 如果你开发了基于 CLIProxyAPI 的项目，请提交一个 PR（拉取请求）将其添加到此列表中。
 

cmd/server/main.go

@@ -59,7 +59,9 @@ func main() {
 	var claudeLogin bool
 	var qwenLogin bool
 	var iflowLogin bool
+	var iflowCookie bool
 	var noBrowser bool
+	var antigravityLogin bool
 	var projectID string
 	var vertexImport string
 	var configPath string
@@ -71,7 +73,9 @@ func main() {
 	flag.BoolVar(&claudeLogin, "claude-login", false, "Login to Claude using OAuth")
 	flag.BoolVar(&qwenLogin, "qwen-login", false, "Login to Qwen using OAuth")
 	flag.BoolVar(&iflowLogin, "iflow-login", false, "Login to iFlow using OAuth")
+	flag.BoolVar(&iflowCookie, "iflow-cookie", false, "Login to iFlow using Cookie")
 	flag.BoolVar(&noBrowser, "no-browser", false, "Don't open browser automatically for OAuth")
+	flag.BoolVar(&antigravityLogin, "antigravity-login", false, "Login to Antigravity using OAuth")
 	flag.StringVar(&projectID, "project_id", "", "Project ID (Gemini only, not required)")
 	flag.StringVar(&configPath, "config", DefaultConfigPath, "Configure File Path")
 	flag.StringVar(&vertexImport, "vertex-import", "", "Import Vertex service account key JSON file")
@@ -135,7 +139,8 @@ func main() {
 
 	wd, err := os.Getwd()
 	if err != nil {
-		log.Fatalf("failed to get working directory: %v", err)
+		log.Errorf("failed to get working directory: %v", err)
+		return
 	}
 
 	// Load environment variables from .env if present.
@@ -229,13 +234,15 @@ func main() {
 		})
 		cancel()
 		if err != nil {
-			log.Fatalf("failed to initialize postgres token store: %v", err)
+			log.Errorf("failed to initialize postgres token store: %v", err)
+			return
 		}
 		examplePath := filepath.Join(wd, "config.example.yaml")
 		ctx, cancel = context.WithTimeout(context.Background(), 30*time.Second)
 		if errBootstrap := pgStoreInst.Bootstrap(ctx, examplePath); errBootstrap != nil {
 			cancel()
-			log.Fatalf("failed to bootstrap postgres-backed config: %v", errBootstrap)
+			log.Errorf("failed to bootstrap postgres-backed config: %v", errBootstrap)
+			return
 		}
 		cancel()
 		configFilePath = pgStoreInst.ConfigPath()
@@ -258,7 +265,8 @@ func main() {
 		if strings.Contains(resolvedEndpoint, "://") {
 			parsed, errParse := url.Parse(resolvedEndpoint)
 			if errParse != nil {
-				log.Fatalf("failed to parse object store endpoint %q: %v", objectStoreEndpoint, errParse)
+				log.Errorf("failed to parse object store endpoint %q: %v", objectStoreEndpoint, errParse)
+				return
 			}
 			switch strings.ToLower(parsed.Scheme) {
 			case "http":
@@ -266,10 +274,12 @@ func main() {
 			case "https":
 				useSSL = true
 			default:
-				log.Fatalf("unsupported object store scheme %q (only http and https are allowed)", parsed.Scheme)
+				log.Errorf("unsupported object store scheme %q (only http and https are allowed)", parsed.Scheme)
+				return
 			}
 			if parsed.Host == "" {
-				log.Fatalf("object store endpoint %q is missing host information", objectStoreEndpoint)
+				log.Errorf("object store endpoint %q is missing host information", objectStoreEndpoint)
+				return
 			}
 			resolvedEndpoint = parsed.Host
 			if parsed.Path != "" && parsed.Path != "/" {
@@ -288,13 +298,15 @@ func main() {
 		}
 		objectStoreInst, err = store.NewObjectTokenStore(objCfg)
 		if err != nil {
-			log.Fatalf("failed to initialize object token store: %v", err)
+			log.Errorf("failed to initialize object token store: %v", err)
+			return
 		}
 		examplePath := filepath.Join(wd, "config.example.yaml")
 		ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
 		if errBootstrap := objectStoreInst.Bootstrap(ctx, examplePath); errBootstrap != nil {
 			cancel()
-			log.Fatalf("failed to bootstrap object-backed config: %v", errBootstrap)
+			log.Errorf("failed to bootstrap object-backed config: %v", errBootstrap)
+			return
 		}
 		cancel()
 		configFilePath = objectStoreInst.ConfigPath()
@@ -319,7 +331,8 @@ func main() {
 		gitStoreInst = store.NewGitTokenStore(gitStoreRemoteURL, gitStoreUser, gitStorePassword)
 		gitStoreInst.SetBaseDir(authDir)
 		if errRepo := gitStoreInst.EnsureRepository(); errRepo != nil {
-			log.Fatalf("failed to prepare git token store: %v", errRepo)
+			log.Errorf("failed to prepare git token store: %v", errRepo)
+			return
 		}
 		configFilePath = gitStoreInst.ConfigPath()
 		if configFilePath == "" {
@@ -328,17 +341,21 @@ func main() {
 		if _, statErr := os.Stat(configFilePath); errors.Is(statErr, fs.ErrNotExist) {
 			examplePath := filepath.Join(wd, "config.example.yaml")
 			if _, errExample := os.Stat(examplePath); errExample != nil {
-				log.Fatalf("failed to find template config file: %v", errExample)
+				log.Errorf("failed to find template config file: %v", errExample)
+				return
 			}
 			if errCopy := misc.CopyConfigTemplate(examplePath, configFilePath); errCopy != nil {
-				log.Fatalf("failed to bootstrap git-backed config: %v", errCopy)
+				log.Errorf("failed to bootstrap git-backed config: %v", errCopy)
+				return
 			}
 			if errCommit := gitStoreInst.PersistConfig(context.Background()); errCommit != nil {
-				log.Fatalf("failed to commit initial git-backed config: %v", errCommit)
+				log.Errorf("failed to commit initial git-backed config: %v", errCommit)
+				return
 			}
 			log.Infof("git-backed config initialized from template: %s", configFilePath)
 		} else if statErr != nil {
-			log.Fatalf("failed to inspect git-backed config: %v", statErr)
+			log.Errorf("failed to inspect git-backed config: %v", statErr)
+			return
 		}
 		cfg, err = config.LoadConfigOptional(configFilePath, isCloudDeploy)
 		if err == nil {
@@ -351,13 +368,15 @@ func main() {
 	} else {
 		wd, err = os.Getwd()
 		if err != nil {
-			log.Fatalf("failed to get working directory: %v", err)
+			log.Errorf("failed to get working directory: %v", err)
+			return
 		}
 		configFilePath = filepath.Join(wd, "config.yaml")
 		cfg, err = config.LoadConfigOptional(configFilePath, isCloudDeploy)
 	}
 	if err != nil {
-		log.Fatalf("failed to load config: %v", err)
+		log.Errorf("failed to load config: %v", err)
+		return
 	}
 	if cfg == nil {
 		cfg = &config.Config{}
@@ -387,7 +406,8 @@ func main() {
 	coreauth.SetQuotaCooldownDisabled(cfg.DisableCooling)
 
 	if err = logging.ConfigureLogOutput(cfg.LoggingToFile); err != nil {
-		log.Fatalf("failed to configure log output: %v", err)
+		log.Errorf("failed to configure log output: %v", err)
+		return
 	}
 
 	log.Infof("CLIProxyAPI Version: %s, Commit: %s, BuiltAt: %s", buildinfo.Version, buildinfo.Commit, buildinfo.BuildDate)
@@ -396,7 +416,8 @@ func main() {
 	util.SetLogLevel(cfg)
 
 	if resolvedAuthDir, errResolveAuthDir := util.ResolveAuthDir(cfg.AuthDir); errResolveAuthDir != nil {
-		log.Fatalf("failed to resolve auth directory: %v", errResolveAuthDir)
+		log.Errorf("failed to resolve auth directory: %v", errResolveAuthDir)
+		return
 	} else {
 		cfg.AuthDir = resolvedAuthDir
 	}
@@ -429,6 +450,9 @@ func main() {
 	} else if login {
 		// Handle Google/Gemini login
 		cmd.DoLogin(cfg, projectID, options)
+	} else if antigravityLogin {
+		// Handle Antigravity login
+		cmd.DoAntigravityLogin(cfg, options)
 	} else if codexLogin {
 		// Handle Codex login
 		cmd.DoCodexLogin(cfg, options)
@@ -439,6 +463,8 @@ func main() {
 		cmd.DoQwenLogin(cfg, options)
 	} else if iflowLogin {
 		cmd.DoIFlowLogin(cfg, options)
+	} else if iflowCookie {
+		cmd.DoIFlowCookieAuth(cfg, options)
 	} else {
 		// In cloud deploy mode without config file, just wait for shutdown signals
 		if isCloudDeploy && !configFileExists {

config.example.yaml

@@ -1,6 +1,16 @@
+# Server host/interface to bind to. Default is empty ("") to bind all interfaces (IPv4 + IPv6).
+# Use "127.0.0.1" or "localhost" to restrict access to local machine only.
+host: ""
+
 # Server port
 port: 8317
 
+# TLS settings for HTTPS. When enabled, the server listens with the provided certificate and key.
+tls:
+  enable: false
+  cert: ""
+  key: ""
+
 # Management API settings
 remote-management:
   # Whether to allow remote (non-localhost) management access.
@@ -38,6 +48,9 @@ proxy-url: ""
 # Number of times to retry a request. Retries will occur if the HTTP response code is 403, 408, 500, 502, 503, or 504.
 request-retry: 3
 
+# Maximum wait time in seconds for a cooled-down credential before triggering a retry.
+max-retry-interval: 30
+
 # Quota exceeded behavior
 quota-exceeded:
   switch-project: true # Whether to automatically switch to another project when a quota is exceeded
@@ -46,69 +59,132 @@ quota-exceeded:
 # When true, enable authentication for the WebSocket API (/v1/ws).
 ws-auth: false
 
-# Gemini API keys (preferred)
-#gemini-api-key:
-#  - api-key: "AIzaSy...01"
-#    base-url: "https://generativelanguage.googleapis.com"
-#    headers:
-#      X-Custom-Header: "custom-value"
-#    proxy-url: "socks5://proxy.example.com:1080"
-#  - api-key: "AIzaSy...02"
-
-# API keys for official Generative Language API (legacy compatibility)
-#generative-language-api-key:
-#  - "AIzaSy...01"
-#  - "AIzaSy...02"
+# Gemini API keys
+# gemini-api-key:
+#   - api-key: "AIzaSy...01"
+#     base-url: "https://generativelanguage.googleapis.com"
+#     headers:
+#       X-Custom-Header: "custom-value"
+#     proxy-url: "socks5://proxy.example.com:1080"
+#     excluded-models:
+#       - "gemini-2.5-pro"     # exclude specific models from this provider (exact match)
+#       - "gemini-2.5-*"       # wildcard matching prefix (e.g. gemini-2.5-flash, gemini-2.5-pro)
+#       - "*-preview"          # wildcard matching suffix (e.g. gemini-3-pro-preview)
+#       - "*flash*"            # wildcard matching substring (e.g. gemini-2.5-flash-lite)
+#   - api-key: "AIzaSy...02"
 
 # Codex API keys
-#codex-api-key:
-#  - api-key: "sk-atSM..."
-#    base-url: "https://www.example.com" # use the custom codex API endpoint
-#    headers:
-#      X-Custom-Header: "custom-value"
-#    proxy-url: "socks5://proxy.example.com:1080" # optional: per-key proxy override
+# codex-api-key:
+#   - api-key: "sk-atSM..."
+#     base-url: "https://www.example.com" # use the custom codex API endpoint
+#     headers:
+#       X-Custom-Header: "custom-value"
+#     proxy-url: "socks5://proxy.example.com:1080" # optional: per-key proxy override
+#     excluded-models:
+#       - "gpt-5.1"         # exclude specific models (exact match)
+#       - "gpt-5-*"         # wildcard matching prefix (e.g. gpt-5-medium, gpt-5-codex)
+#       - "*-mini"          # wildcard matching suffix (e.g. gpt-5-codex-mini)
+#       - "*codex*"         # wildcard matching substring (e.g. gpt-5-codex-low)
 
 # Claude API keys
-#claude-api-key:
-#  - api-key: "sk-atSM..." # use the official claude API key, no need to set the base url
-#  - api-key: "sk-atSM..."
-#    base-url: "https://www.example.com" # use the custom claude API endpoint
-#    headers:
-#      X-Custom-Header: "custom-value"
-#    proxy-url: "socks5://proxy.example.com:1080" # optional: per-key proxy override
-#    models:
-#      - name: "claude-3-5-sonnet-20241022" # upstream model name
-#        alias: "claude-sonnet-latest" # client alias mapped to the upstream model
+# claude-api-key:
+#   - api-key: "sk-atSM..." # use the official claude API key, no need to set the base url
+#   - api-key: "sk-atSM..."
+#     base-url: "https://www.example.com" # use the custom claude API endpoint
+#     headers:
+#       X-Custom-Header: "custom-value"
+#     proxy-url: "socks5://proxy.example.com:1080" # optional: per-key proxy override
+#     models:
+#       - name: "claude-3-5-sonnet-20241022" # upstream model name
+#         alias: "claude-sonnet-latest" # client alias mapped to the upstream model
+#     excluded-models:
+#       - "claude-opus-4-5-20251101" # exclude specific models (exact match)
+#       - "claude-3-*"               # wildcard matching prefix (e.g. claude-3-7-sonnet-20250219)
+#       - "*-thinking"               # wildcard matching suffix (e.g. claude-opus-4-5-thinking)
+#       - "*haiku*"                  # wildcard matching substring (e.g. claude-3-5-haiku-20241022)
 
 # OpenAI compatibility providers
-#openai-compatibility:
-#  - name: "openrouter" # The name of the provider; it will be used in the user agent and other places.
-#    base-url: "https://openrouter.ai/api/v1" # The base URL of the provider.
-#    headers:
-#      X-Custom-Header: "custom-value"
-#    # New format with per-key proxy support (recommended):
-#    api-key-entries:
-#      - api-key: "sk-or-v1-...b780"
-#        proxy-url: "socks5://proxy.example.com:1080" # optional: per-key proxy override
-#      - api-key: "sk-or-v1-...b781" # without proxy-url
-#    # Legacy format (still supported, but cannot specify proxy per key):
-#    # api-keys:
-#    #   - "sk-or-v1-...b780"
-#    #   - "sk-or-v1-...b781"
-#    models: # The models supported by the provider.
-#      - name: "moonshotai/kimi-k2:free" # The actual model name.
-#        alias: "kimi-k2" # The alias used in the API.
+# openai-compatibility:
+#   - name: "openrouter" # The name of the provider; it will be used in the user agent and other places.
+#     base-url: "https://openrouter.ai/api/v1" # The base URL of the provider.
+#     headers:
+#       X-Custom-Header: "custom-value"
+#     api-key-entries:
+#       - api-key: "sk-or-v1-...b780"
+#         proxy-url: "socks5://proxy.example.com:1080" # optional: per-key proxy override
+#       - api-key: "sk-or-v1-...b781" # without proxy-url
+#     models: # The models supported by the provider.
+#       - name: "moonshotai/kimi-k2:free" # The actual model name.
+#         alias: "kimi-k2" # The alias used in the API.
 
-#payload: # Optional payload configuration
-#  default: # Default rules only set parameters when they are missing in the payload.
-#    - models:
-#        - name: "gemini-2.5-pro" # Supports wildcards (e.g., "gemini-*")
-#          protocol: "gemini" # restricts the rule to a specific protocol, options: openai, gemini, claude, codex
-#      params: # JSON path (gjson/sjson syntax) -> value
-#        "generationConfig.thinkingConfig.thinkingBudget": 32768
-#  override: # Override rules always set parameters, overwriting any existing values.
-#    - models:
-#        - name: "gpt-*" # Supports wildcards (e.g., "gpt-*")
-#          protocol: "codex" # restricts the rule to a specific protocol, options: openai, gemini, claude, codex
-#      params: # JSON path (gjson/sjson syntax) -> value
-#        "reasoning.effort": "high"
+# Vertex API keys (Vertex-compatible endpoints, use API key + base URL)
+# vertex-api-key:
+#   - api-key: "vk-123..."                        # x-goog-api-key header
+#     base-url: "https://example.com/api"         # e.g. https://zenmux.ai/api
+#     proxy-url: "socks5://proxy.example.com:1080" # optional per-key proxy override
+#     headers:
+#       X-Custom-Header: "custom-value"
+#     models:                                     # optional: map aliases to upstream model names
+#       - name: "gemini-2.0-flash"                # upstream model name
+#         alias: "vertex-flash"                   # client-visible alias
+#       - name: "gemini-1.5-pro"
+#         alias: "vertex-pro"
+
+# Amp Integration
+# ampcode:
+#   # Configure upstream URL for Amp CLI OAuth and management features
+#   upstream-url: "https://ampcode.com"
+#   # Optional: Override API key for Amp upstream (otherwise uses env or file)
+#   upstream-api-key: ""
+#   # Restrict Amp management routes (/api/auth, /api/user, etc.) to localhost only (recommended)
+#   restrict-management-to-localhost: true
+#   # Force model mappings to run before checking local API keys (default: false)
+#   force-model-mappings: false
+#   # Amp Model Mappings
+#   # Route unavailable Amp models to alternative models available in your local proxy.
+#   # Useful when Amp CLI requests models you don't have access to (e.g., Claude Opus 4.5)
+#   # but you have a similar model available (e.g., Claude Sonnet 4).
+#   model-mappings:
+#     - from: "claude-opus-4.5"       # Model requested by Amp CLI
+#       to: "claude-sonnet-4"         # Route to this available model instead
+#     - from: "gpt-5"
+#       to: "gemini-2.5-pro"
+#     - from: "claude-3-opus-20240229"
+#       to: "claude-3-5-sonnet-20241022"
+
+# OAuth provider excluded models
+# oauth-excluded-models:
+#   gemini-cli:
+#     - "gemini-2.5-pro"     # exclude specific models (exact match)
+#     - "gemini-2.5-*"       # wildcard matching prefix (e.g. gemini-2.5-flash, gemini-2.5-pro)
+#     - "*-preview"          # wildcard matching suffix (e.g. gemini-3-pro-preview)
+#     - "*flash*"            # wildcard matching substring (e.g. gemini-2.5-flash-lite)
+#   vertex:
+#     - "gemini-3-pro-preview"
+#   aistudio:
+#     - "gemini-3-pro-preview"
+#   antigravity:
+#     - "gemini-3-pro-preview"
+#   claude:
+#     - "claude-3-5-haiku-20241022"
+#   codex:
+#     - "gpt-5-codex-mini"
+#   qwen:
+#     - "vision-model"
+#   iflow:
+#     - "tstars2.0"
+
+# Optional payload configuration
+# payload:
+#   default: # Default rules only set parameters when they are missing in the payload.
+#     - models:
+#         - name: "gemini-2.5-pro" # Supports wildcards (e.g., "gemini-*")
+#           protocol: "gemini" # restricts the rule to a specific protocol, options: openai, gemini, claude, codex
+#       params: # JSON path (gjson/sjson syntax) -> value
+#         "generationConfig.thinkingConfig.thinkingBudget": 32768
+#   override: # Override rules always set parameters, overwriting any existing values.
+#     - models:
+#         - name: "gpt-*" # Supports wildcards (e.g., "gpt-*")
+#           protocol: "codex" # restricts the rule to a specific protocol, options: openai, gemini, claude, codex
+#       params: # JSON path (gjson/sjson syntax) -> value
+#         "reasoning.effort": "high"

docker-compose.yml

@@ -19,6 +19,7 @@ services:
       - "8085:8085"
       - "1455:1455"
       - "54545:54545"
+      - "51121:51121"
       - "11451:11451"
     volumes:
       - ./config.yaml:/CLIProxyAPI/config.yaml

docs/amp-cli-integration.md

@@ -0,0 +1,443 @@
+# Amp CLI Integration Guide
+
+This guide explains how to use CLIProxyAPI with Amp CLI and Amp IDE extensions, enabling you to use your existing Google/ChatGPT/Claude subscriptions (via OAuth) with Amp's CLI.
+
+## Table of Contents
+
+- [Overview](#overview)
+  - [Which Providers Should You Authenticate?](#which-providers-should-you-authenticate)
+- [Architecture](#architecture)
+- [Configuration](#configuration)
+  - [Model Mapping Configuration](#model-mapping-configuration)
+- [Setup](#setup)
+- [Usage](#usage)
+- [Troubleshooting](#troubleshooting)
+
+## Overview
+
+The Amp CLI integration adds specialized routing to support Amp's API patterns while maintaining full compatibility with all existing CLIProxyAPI features. This allows you to use both traditional CLIProxyAPI features and Amp CLI with the same proxy server.
+
+### Key Features
+
+- **Provider route aliases**: Maps Amp's `/api/provider/{provider}/v1...` patterns to CLIProxyAPI handlers
+- **Management proxy**: Forwards OAuth and account management requests to Amp's control plane
+- **Smart fallback**: Automatically routes unconfigured models to ampcode.com
+- **Model mapping**: Route unavailable models to alternatives you have access to (e.g., `claude-opus-4.5` → `claude-sonnet-4`)
+- **Secret management**: Configurable precedence (config > env > file) with 5-minute caching
+- **Security-first**: Management routes restricted to localhost by default
+- **Automatic gzip handling**: Decompresses responses from Amp upstream
+
+### What You Can Do
+
+- Use Amp CLI with your Google account (Gemini 3 Pro Preview, Gemini 2.5 Pro, Gemini 2.5 Flash)
+- Use Amp CLI with your ChatGPT Plus/Pro subscription (GPT-5, GPT-5 Codex models)
+- Use Amp CLI with your Claude Pro/Max subscription (Claude Sonnet 4.5, Opus 4.1)
+- Use Amp IDE extensions (VS Code, Cursor, Windsurf, etc.) with the same proxy
+- Run multiple CLI tools (Factory + Amp) through one proxy server
+- Route unconfigured models automatically through ampcode.com
+
+### Which Providers Should You Authenticate?
+
+**Important**: The providers you need to authenticate depend on which models and features your installed version of Amp currently uses. Amp employs different providers for various agent modes and specialized subagents:
+
+- **Smart mode**: Uses Google/Gemini models (Gemini 3 Pro)
+- **Rush mode**: Uses Anthropic/Claude models (Claude Haiku 4.5)
+- **Oracle subagent**: Uses OpenAI/GPT models (GPT-5 medium reasoning)
+- **Librarian subagent**: Uses Anthropic/Claude models (Claude Sonnet 4.5)
+- **Search subagent**: Uses Anthropic/Claude models (Claude Haiku 4.5)
+- **Review feature**: Uses Google/Gemini models (Gemini 2.5 Flash-Lite)
+
+For the most current information about which models Amp uses, see the **[Amp Models Documentation](https://ampcode.com/models)**.
+
+#### Fallback Behavior
+
+CLIProxyAPI uses a smart fallback system:
+
+1. **Provider authenticated locally** (`--login`, `--codex-login`, `--claude-login`):
+   - Requests use **your OAuth subscription** (ChatGPT Plus/Pro, Claude Pro/Max, Google account)
+   - You benefit from your subscription's included usage quotas
+   - No Amp credits consumed
+
+2. **Provider NOT authenticated locally**:
+   - Requests automatically forward to **ampcode.com**
+   - Uses Amp's backend provider connections
+   - **Requires Amp credits** if the provider is paid (OpenAI, Anthropic paid tiers)
+   - May result in errors if Amp credit balance is insufficient
+
+**Recommendation**: Authenticate all providers you have subscriptions for to maximize value and minimize Amp credit usage. If you don't have subscriptions to all providers Amp uses, ensure you have sufficient Amp credits available for fallback requests.
+
+## Architecture
+
+### Request Flow
+
+```
+Amp CLI/IDE
+  ↓
+  ├─ Provider API requests (/api/provider/{provider}/v1/...)
+  │   ↓
+  │   ├─ Model configured locally?
+  │   │   YES → Use local OAuth tokens (OpenAI/Claude/Gemini handlers)
+  │   │   NO  ↓
+  │   │       ├─ Model mapping configured?
+  │   │       │   YES → Rewrite model → Use local handler (free)
+  │   │       │   NO  → Forward to ampcode.com (uses Amp credits)
+  │   ↓
+  │   Response
+  │
+  └─ Management requests (/api/auth, /api/user, /api/threads, ...)
+      ↓
+      ├─ Localhost check (security)
+      ↓
+      └─ Reverse proxy to ampcode.com
+          ↓
+          Response (auto-decompressed if gzipped)
+```
+
+### Components
+
+The Amp integration is implemented as a modular routing module (`internal/api/modules/amp/`) with these components:
+
+1. **Route Aliases** (`routes.go`): Maps Amp-style paths to standard handlers
+2. **Reverse Proxy** (`proxy.go`): Forwards management requests to ampcode.com
+3. **Fallback Handler** (`fallback_handlers.go`): Routes unconfigured models to ampcode.com
+4. **Secret Management** (`secret.go`): Multi-source API key resolution with caching
+5. **Main Module** (`amp.go`): Orchestrates registration and configuration
+
+## Configuration
+
+### Basic Configuration
+
+Add these fields to your `config.yaml`:
+
+```yaml
+# Amp upstream control plane (required for management routes)
+amp-upstream-url: "https://ampcode.com"
+
+# Optional: Override API key (otherwise uses env or file)
+# amp-upstream-api-key: "your-amp-api-key"
+
+# Security: restrict management routes to localhost (recommended)
+amp-restrict-management-to-localhost: true
+```
+
+### Model Mapping Configuration
+
+When Amp CLI requests a model that you don't have access to, you can configure mappings to route those requests to alternative models that you DO have available. This avoids consuming Amp credits for models you could handle locally.
+
+```yaml
+# Route unavailable models to alternatives
+amp-model-mappings:
+  # Example: Route Claude Opus 4.5 requests to Claude Sonnet 4
+  - from: "claude-opus-4.5"
+    to: "claude-sonnet-4"
+  
+  # Example: Route GPT-5 requests to Gemini 2.5 Pro
+  - from: "gpt-5"
+    to: "gemini-2.5-pro"
+  
+  # Example: Map older model names to newer versions
+  - from: "claude-3-opus-20240229"
+    to: "claude-3-5-sonnet-20241022"
+```
+
+**How it works:**
+
+1. Amp CLI requests a model (e.g., `claude-opus-4.5`)
+2. CLIProxyAPI checks if a local provider is available for that model
+3. If not available, it checks the model mappings
+4. If a mapping exists, the request is rewritten to use the target model
+5. The request is then handled locally (free, using your OAuth subscription)
+
+**Benefits:**
+- **Save Amp credits**: Use your local subscriptions instead of forwarding to ampcode.com
+- **Hot-reload**: Mappings can be updated without restarting the proxy
+- **Structured logging**: Clear logs show when mappings are applied
+
+**Routing Decision Logs:**
+
+The proxy logs each routing decision with structured fields:
+
+```
+[AMP] Using local provider for model: gemini-2.5-pro          # Local provider (free)
+[AMP] Model mapped: claude-opus-4.5 -> claude-sonnet-4        # Mapping applied (free)
+[AMP] Forwarding to ampcode.com (uses Amp credits) - model_id: gpt-5  # Fallback (costs credits)
+```
+
+### Secret Resolution Precedence
+
+The Amp module resolves API keys using this precedence order:
+
+| Source | Key | Priority | Cache |
+|--------|-----|----------|-------|
+| Config file | `amp-upstream-api-key` | High | No |
+| Environment | `AMP_API_KEY` | Medium | No |
+| Amp secrets file | `~/.local/share/amp/secrets.json` | Low | 5 min |
+
+**Recommendation**: Use the Amp secrets file (lowest precedence) for normal usage. This file is automatically managed by `amp login`.
+
+### Security Settings
+
+**`amp-restrict-management-to-localhost`** (default: `true`)
+
+When enabled, management routes (`/api/auth`, `/api/user`, `/api/threads`, etc.) only accept connections from localhost (127.0.0.1, ::1). This prevents:
+- Drive-by browser attacks
+- Remote access to management endpoints
+- CORS-based attacks
+- Header spoofing attacks (e.g., `X-Forwarded-For: 127.0.0.1`)
+
+#### How It Works
+
+This restriction uses the **actual TCP connection address** (`RemoteAddr`), not HTTP headers like `X-Forwarded-For`. This prevents header spoofing attacks but has important implications:
+
+- ✅ **Works for direct connections**: Running CLIProxyAPI directly on your machine or server
+- ⚠️ **May not work behind reverse proxies**: If deploying behind nginx, Cloudflare, or other proxies, the connection will appear to come from the proxy's IP, not localhost
+
+#### Reverse Proxy Deployments
+
+If you need to run CLIProxyAPI behind a reverse proxy (nginx, Caddy, Cloudflare Tunnel, etc.):
+
+1. **Disable the localhost restriction**:
+   ```yaml
+   amp-restrict-management-to-localhost: false
+   ```
+
+2. **Use alternative security measures**:
+   - Firewall rules restricting access to management routes
+   - Proxy-level authentication (HTTP Basic Auth, OAuth)
+   - Network-level isolation (VPN, Tailscale, Cloudflare Access)
+   - Bind CLIProxyAPI to `127.0.0.1` only and access via SSH tunnel
+
+3. **Example nginx configuration** (blocks external access to management routes):
+   ```nginx
+   location /api/auth { deny all; }
+   location /api/user { deny all; }
+   location /api/threads { deny all; }
+   location /api/internal { deny all; }
+   ```
+
+**Important**: Only disable `amp-restrict-management-to-localhost` if you understand the security implications and have other protections in place.
+
+## Setup
+
+### 1. Configure CLIProxyAPI
+
+Create or edit `config.yaml`:
+
+```yaml
+port: 8317
+auth-dir: "~/.cli-proxy-api"
+
+# Amp integration
+amp-upstream-url: "https://ampcode.com"
+amp-restrict-management-to-localhost: true
+
+# Other standard settings...
+debug: false
+logging-to-file: true
+```
+
+### 2. Authenticate with Providers
+
+Run OAuth login for the providers you want to use:
+
+**Google Account (Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 3 Pro Preview):**
+```bash
+./cli-proxy-api --login
+```
+
+**ChatGPT Plus/Pro (GPT-5, GPT-5 Codex):**
+```bash
+./cli-proxy-api --codex-login
+```
+
+**Claude Pro/Max (Claude Sonnet 4.5, Opus 4.1):**
+```bash
+./cli-proxy-api --claude-login
+```
+
+Tokens are saved to:
+- Gemini: `~/.cli-proxy-api/gemini-<email>.json`
+- OpenAI Codex: `~/.cli-proxy-api/codex-<email>.json`
+- Claude: `~/.cli-proxy-api/claude-<email>.json`
+
+### 3. Start the Proxy
+
+```bash
+./cli-proxy-api --config config.yaml
+```
+
+Or run in background with tmux (recommended for remote servers):
+
+```bash
+tmux new-session -d -s proxy "./cli-proxy-api --config config.yaml"
+```
+
+### 4. Configure Amp CLI
+
+#### Option A: Settings File
+
+Edit `~/.config/amp/settings.json`:
+
+```json
+{
+  "amp.url": "http://localhost:8317"
+}
+```
+
+#### Option B: Environment Variable
+
+```bash
+export AMP_URL=http://localhost:8317
+```
+
+### 5. Login and Use Amp
+
+Login through the proxy (proxied to ampcode.com):
+
+```bash
+amp login
+```
+
+Use Amp as normal:
+
+```bash
+amp "Write a hello world program in Python"
+```
+
+### 6. (Optional) Configure Amp IDE Extension
+
+The proxy also works with Amp IDE extensions for VS Code, Cursor, Windsurf, etc.
+
+1. Open Amp extension settings in your IDE
+2. Set **Amp URL** to `http://localhost:8317`
+3. Login with your Amp account
+4. Start using Amp in your IDE
+
+Both CLI and IDE can use the proxy simultaneously.
+
+## Usage
+
+### Supported Routes
+
+#### Provider Aliases (Always Available)
+
+These routes work even without `amp-upstream-url` configured:
+
+- `/api/provider/openai/v1/chat/completions`
+- `/api/provider/openai/v1/responses`
+- `/api/provider/anthropic/v1/messages`
+- `/api/provider/google/v1beta/models/:action`
+
+Amp CLI calls these routes with your OAuth-authenticated models configured in CLIProxyAPI.
+
+#### Management Routes (Require `amp-upstream-url`)
+
+These routes are proxied to ampcode.com:
+
+- `/api/auth` - Authentication
+- `/api/user` - User profile
+- `/api/meta` - Metadata
+- `/api/threads` - Conversation threads
+- `/api/telemetry` - Usage telemetry
+- `/api/internal` - Internal APIs
+
+**Security**: Restricted to localhost by default.
+
+### Model Fallback Behavior
+
+When Amp requests a model:
+
+1. **Check local configuration**: Does CLIProxyAPI have OAuth tokens for this model's provider?
+2. **If YES**: Route to local handler (use your OAuth subscription)
+3. **If NO**: Check if a model mapping exists
+4. **If mapping exists**: Rewrite request to mapped model → Route to local handler (free)
+5. **If no mapping**: Forward to ampcode.com (uses Amp credits)
+
+This enables seamless mixed usage:
+- Models you've configured (Gemini, ChatGPT, Claude) → Your OAuth subscriptions
+- Models with mappings configured → Routed to alternative local models (free)
+- Models you haven't configured and have no mapping → Amp's default providers (uses credits)
+
+### Example API Calls
+
+**Chat completion with local OAuth:**
+```bash
+curl http://localhost:8317/api/provider/openai/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "gpt-5",
+    "messages": [{"role": "user", "content": "Hello"}]
+  }'
+```
+
+**Management endpoint (localhost only):**
+```bash
+curl http://localhost:8317/api/user
+```
+
+## Troubleshooting
+
+### Common Issues
+
+| Symptom | Likely Cause | Fix |
+|---------|--------------|-----|
+| 404 on `/api/provider/...` | Incorrect route path | Ensure exact path: `/api/provider/{provider}/v1...` |
+| 403 on `/api/user` | Non-localhost request | Run from same machine or disable `amp-restrict-management-to-localhost` (not recommended) |
+| 401/403 from provider | Missing/expired OAuth | Re-run `--codex-login` or `--claude-login` |
+| Amp gzip errors | Response decompression issue | Update to latest build; auto-decompression should handle this |
+| Models not using proxy | Wrong Amp URL | Verify `amp.url` setting or `AMP_URL` environment variable |
+| CORS errors | Protected management endpoint | Use CLI/terminal, not browser |
+
+### Diagnostics
+
+**Check proxy logs:**
+```bash
+# If logging-to-file: true
+tail -f logs/requests.log
+
+# If running in tmux
+tmux attach-session -t proxy
+```
+
+**Enable debug mode** (temporarily):
+```yaml
+debug: true
+```
+
+**Test basic connectivity:**
+```bash
+# Check if proxy is running
+curl http://localhost:8317/v1/models
+
+# Check Amp-specific route
+curl http://localhost:8317/api/provider/openai/v1/models
+```
+
+**Verify Amp configuration:**
+```bash
+# Check if Amp is using proxy
+amp config get amp.url
+
+# Or check environment
+echo $AMP_URL
+```
+
+### Security Checklist
+
+- ✅ Keep `amp-restrict-management-to-localhost: true` (default)
+- ✅ Don't expose proxy publicly (bind to localhost or use firewall/VPN)
+- ✅ Use the Amp secrets file (`~/.local/share/amp/secrets.json`) managed by `amp login`
+- ✅ Rotate OAuth tokens periodically by re-running login commands
+- ✅ Store config and auth-dir on encrypted disk if handling sensitive data
+- ✅ Keep proxy binary up to date for security fixes
+
+## Additional Resources
+
+- [CLIProxyAPI Main Documentation](https://help.router-for.me/)
+- [Amp CLI Official Manual](https://ampcode.com/manual)
+- [Management API Reference](https://help.router-for.me/management/api)
+- [SDK Documentation](sdk-usage.md)
+
+## Disclaimer
+
+This integration is for personal/educational use. Using reverse proxies or alternate API bases may violate provider Terms of Service. You are solely responsible for how you use this software. Accounts may be rate-limited, locked, or banned. No warranties. Use at your own risk.

docs/amp-cli-integration_CN.md

@@ -0,0 +1,392 @@
+# Amp CLI 集成指南
+
+本指南说明如何在 Amp CLI 和 Amp IDE 扩展中使用 CLIProxyAPI，通过 OAuth 让你能够把已有的 Google/ChatGPT/Claude 订阅与 Amp 的 CLI 一起使用。
+
+## 目录
+
+- [概述](#概述)
+  - [应该认证哪些服务提供商？](#应该认证哪些服务提供商)
+- [架构](#架构)
+- [配置](#配置)
+- [设置](#设置)
+- [用法](#用法)
+- [故障排查](#故障排查)
+
+## 概述
+
+Amp CLI 集成为 Amp 的 API 模式添加了专用路由，同时保持与现有 CLIProxyAPI 功能的完全兼容。这样你可以在同一个代理服务器上同时使用传统 CLIProxyAPI 功能和 Amp CLI。
+
+### 主要特性
+
+- **提供者路由别名**：将 Amp 的 `/api/provider/{provider}/v1...` 路径映射到 CLIProxyAPI 处理器
+- **管理代理**：将 OAuth 和账号管理请求转发到 Amp 控制平面
+- **智能回退**：自动将未配置的模型路由到 ampcode.com
+- **密钥管理**：可配置优先级（配置 > 环境变量 > 文件），缓存 5 分钟
+- **安全优先**：管理路由默认限制为 localhost
+- **自动 gzip 处理**：自动解压来自 Amp 上游的响应
+
+### 你可以做什么
+
+- 使用 Amp CLI 搭配你的 Google 账号（Gemini 3 Pro Preview、Gemini 2.5 Pro、Gemini 2.5 Flash）
+- 使用 Amp CLI 搭配你的 ChatGPT Plus/Pro 订阅（GPT-5、GPT-5 Codex 模型）
+- 使用 Amp CLI 搭配你的 Claude Pro/Max 订阅（Claude Sonnet 4.5、Opus 4.1）
+- 将 Amp IDE 扩展（VS Code、Cursor、Windsurf 等）与同一个代理一起使用
+- 通过一个代理同时运行多个 CLI 工具（Factory + Amp）
+- 将未配置的模型自动路由到 ampcode.com
+
+### 应该认证哪些服务提供商？
+
+**重要**：需要认证的提供商取决于你安装的 Amp 版本当前使用的模型和功能。Amp 的不同智能模式和子代理会使用不同的提供商：
+
+- **Smart 模式**：使用 Google/Gemini 模型（Gemini 3 Pro）
+- **Rush 模式**：使用 Anthropic/Claude 模型（Claude Haiku 4.5）
+- **Oracle 子代理**：使用 OpenAI/GPT 模型（GPT-5 medium reasoning）
+- **Librarian 子代理**：使用 Anthropic/Claude 模型（Claude Sonnet 4.5）
+- **Search 子代理**：使用 Anthropic/Claude 模型（Claude Haiku 4.5）
+- **Review 功能**：使用 Google/Gemini 模型（Gemini 2.5 Flash-Lite）
+
+有关 Amp 当前使用哪些模型的最新信息，请参阅 **[Amp 模型文档](https://ampcode.com/models)**。
+
+#### 回退行为
+
+CLIProxyAPI 采用智能回退机制：
+
+1. **本地已认证提供商**（`--login`、`--codex-login`、`--claude-login`）：
+   - 请求使用**你的 OAuth 订阅**（ChatGPT Plus/Pro、Claude Pro/Max、Google 账号）
+   - 享受订阅自带的额度
+   - 不消耗 Amp 额度
+
+2. **本地未认证提供商**：
+   - 请求自动转发到 **ampcode.com**
+   - 使用 Amp 的后端提供商连接
+   - 如果提供商是付费的（OpenAI、Anthropic 付费档），**需要消耗 Amp 额度**
+   - 若 Amp 额度不足，可能产生错误
+
+**建议**：对你有订阅的所有提供商都进行认证，以最大化价值并尽量减少 Amp 额度消耗。如果没有覆盖 Amp 使用的全部提供商，请确保为回退请求准备足够的 Amp 额度。
+
+## 架构
+
+### 请求流
+
+```
+Amp CLI/IDE
+  ↓
+  ├─ Provider API requests (/api/provider/{provider}/v1/...)
+  │   ↓
+  │   ├─ Model configured locally?
+  │   │   YES → Use local OAuth tokens (OpenAI/Claude/Gemini handlers)
+  │   │   NO  → Forward to ampcode.com (reverse proxy)
+  │   ↓
+  │   Response
+  │
+  └─ Management requests (/api/auth, /api/user, /api/threads, ...)
+      ↓
+      ├─ Localhost check (security)
+      ↓
+      └─ Reverse proxy to ampcode.com
+          ↓
+          Response (auto-decompressed if gzipped)
+```
+
+### 组件
+
+Amp 集成以模块化路由模块（`internal/api/modules/amp/`）实现，包含以下组件：
+
+1. **路由别名**（`routes.go`）：将 Amp 风格的路径映射到标准处理器
+2. **反向代理**（`proxy.go`）：将管理请求转发到 ampcode.com
+3. **回退处理器**（`fallback_handlers.go`）：将未配置的模型路由到 ampcode.com
+4. **密钥管理**（`secret.go`）：多来源 API 密钥解析并带缓存
+5. **主模块**（`amp.go`）：负责注册和配置
+
+## 配置
+
+### 基础配置
+
+在 `config.yaml` 中新增以下字段：
+
+```yaml
+# Amp 上游控制平面（管理路由必需）
+amp-upstream-url: "https://ampcode.com"
+
+# 可选：覆盖 API key（否则使用环境变量或文件）
+# amp-upstream-api-key: "your-amp-api-key"
+
+# 安全性：将管理路由限制为 localhost（推荐）
+amp-restrict-management-to-localhost: true
+```
+
+### 密钥解析优先级
+
+Amp 模块以如下优先级解析 API key：
+
+| 来源 | 键名 | 优先级 | 缓存 |
+|------|------|--------|------|
+| 配置文件 | `amp-upstream-api-key` | 高 | 无 |
+| 环境变量 | `AMP_API_KEY` | 中 | 无 |
+| Amp 密钥文件 | `~/.local/share/amp/secrets.json` | 低 | 5 分钟 |
+
+**建议**：日常使用时采用 Amp 密钥文件（最低优先级）。该文件由 `amp login` 自动管理。
+
+### 安全设置
+
+**`amp-restrict-management-to-localhost`**（默认：`true`）
+
+启用后，管理路由（`/api/auth`、`/api/user`、`/api/threads` 等）只接受来自 localhost（127.0.0.1、::1）的连接，可防止：
+- 浏览器探测式攻击
+- 对管理端点的远程访问
+- 基于 CORS 的攻击
+- 伪造头攻击（例如 `X-Forwarded-For: 127.0.0.1`）
+
+#### 工作原理
+
+此限制使用**实际的 TCP 连接地址**（`RemoteAddr`），而非 `X-Forwarded-For` 等 HTTP 头，能防止头部伪造，但有重要影响：
+
+- ✅ **直接连接可用**：在本机或服务器直接运行 CLIProxyAPI 时适用
+- ⚠️ **可能不适用于反向代理场景**：部署在 nginx、Cloudflare 等代理后，请求源会显示为代理 IP 而非 localhost
+
+#### 反向代理部署
+
+若需要在反向代理（nginx、Caddy、Cloudflare Tunnel 等）后运行 CLIProxyAPI：
+
+1. **关闭 localhost 限制**：
+   ```yaml
+   amp-restrict-management-to-localhost: false
+   ```
+
+2. **使用替代安全措施**：
+   - 防火墙规则限制管理路由访问
+   - 代理层认证（HTTP Basic Auth、OAuth）
+   - 网络隔离（VPN、Tailscale、Cloudflare Access）
+   - 将 CLIProxyAPI 仅绑定 `127.0.0.1`，并通过 SSH 隧道访问
+
+3. **nginx 示例配置**（阻止外部访问管理路由）：
+   ```nginx
+   location /api/auth { deny all; }
+   location /api/user { deny all; }
+   location /api/threads { deny all; }
+   location /api/internal { deny all; }
+   ```
+
+**重要**：只有在理解安全影响并已采取其他防护措施时，才关闭 `amp-restrict-management-to-localhost`。
+
+## 设置
+
+### 1. 配置 CLIProxyAPI
+
+创建或编辑 `config.yaml`：
+
+```yaml
+port: 8317
+auth-dir: "~/.cli-proxy-api"
+
+# Amp 集成
+amp-upstream-url: "https://ampcode.com"
+amp-restrict-management-to-localhost: true
+
+# 其他常规设置...
+debug: false
+logging-to-file: true
+```
+
+### 2. 认证提供商
+
+为要使用的提供商执行 OAuth 登录：
+
+**Google 账号（Gemini 2.5 Pro、Gemini 2.5 Flash、Gemini 3 Pro Preview）：**
+```bash
+./cli-proxy-api --login
+```
+
+**ChatGPT Plus/Pro（GPT-5、GPT-5 Codex）：**
+```bash
+./cli-proxy-api --codex-login
+```
+
+**Claude Pro/Max（Claude Sonnet 4.5、Opus 4.1）：**
+```bash
+./cli-proxy-api --claude-login
+```
+
+令牌会保存到：
+- Gemini: `~/.cli-proxy-api/gemini-<email>.json`
+- OpenAI Codex: `~/.cli-proxy-api/codex-<email>.json`
+- Claude: `~/.cli-proxy-api/claude-<email>.json`
+
+### 3. 启动代理
+
+```bash
+./cli-proxy-api --config config.yaml
+```
+
+或使用 tmux 在后台运行（推荐用于远程服务器）：
+
+```bash
+tmux new-session -d -s proxy "./cli-proxy-api --config config.yaml"
+```
+
+### 4. 配置 Amp CLI
+
+#### 方案 A：配置文件
+
+编辑 `~/.config/amp/settings.json`：
+
+```json
+{
+  "amp.url": "http://localhost:8317"
+}
+```
+
+#### 方案 B：环境变量
+
+```bash
+export AMP_URL=http://localhost:8317
+```
+
+### 5. 登录并使用 Amp
+
+通过代理登录（请求会被代理到 ampcode.com）：
+
+```bash
+amp login
+```
+
+像平常一样使用 Amp：
+
+```bash
+amp "Write a hello world program in Python"
+```
+
+### 6. （可选）配置 Amp IDE 扩展
+
+该代理同样适用于 VS Code、Cursor、Windsurf 等 Amp IDE 扩展。
+
+1. 在 IDE 中打开 Amp 扩展设置
+2. 将 **Amp URL** 设置为 `http://localhost:8317`
+3. 用你的 Amp 账号登录
+4. 在 IDE 中开始使用 Amp
+
+CLI 和 IDE 可同时使用该代理。
+
+## 用法
+
+### 支持的路由
+
+#### 提供商别名（始终可用）
+
+这些路由即使未配置 `amp-upstream-url` 也可使用：
+
+- `/api/provider/openai/v1/chat/completions`
+- `/api/provider/openai/v1/responses`
+- `/api/provider/anthropic/v1/messages`
+- `/api/provider/google/v1beta/models/:action`
+
+Amp CLI 会使用你在 CLIProxyAPI 中通过 OAuth 认证的模型来调用这些路由。
+
+#### 管理路由（需要 `amp-upstream-url`）
+
+这些路由会被代理到 ampcode.com：
+
+- `/api/auth` - 认证
+- `/api/user` - 用户资料
+- `/api/meta` - 元数据
+- `/api/threads` - 会话线程
+- `/api/telemetry` - 使用遥测
+- `/api/internal` - 内部 API
+
+**安全性**：默认限制为 localhost。
+
+### 模型回退行为
+
+当 Amp 请求模型时：
+
+1. **检查本地配置**：CLIProxyAPI 是否有该模型提供商的 OAuth 令牌？
+2. **如果有**：路由到本地处理器（使用你的 OAuth 订阅）
+3. **如果没有**：转发到 ampcode.com（使用 Amp 的默认路由）
+
+这实现了无缝混用：
+- 你已配置的模型（Gemini、ChatGPT、Claude）→ 你的 OAuth 订阅
+- 未配置的模型 → Amp 的默认提供商
+
+### 示例 API 调用
+
+**使用本地 OAuth 的聊天补全：**
+```bash
+curl http://localhost:8317/api/provider/openai/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "gpt-5",
+    "messages": [{"role": "user", "content": "Hello"}]
+  }'
+```
+
+**管理端点（仅限 localhost）：**
+```bash
+curl http://localhost:8317/api/user
+```
+
+## 故障排查
+
+### 常见问题
+
+| 症状 | 可能原因 | 解决方案 |
+|------|----------|----------|
+| `/api/provider/...` 返回 404 | 路径错误 | 确保路径准确：`/api/provider/{provider}/v1...` |
+| `/api/user` 返回 403 | 非 localhost 请求 | 在同一机器上访问，或关闭 `amp-restrict-management-to-localhost`（不推荐） |
+| 提供商返回 401/403 | OAuth 缺失或过期 | 重新运行 `--codex-login` 或 `--claude-login` |
+| Amp gzip 错误 | 响应解压问题 | 更新到最新构建；自动解压应能处理 |
+| 模型未走代理 | Amp URL 设置错误 | 检查 `amp.url` 设置或 `AMP_URL` 环境变量 |
+| CORS 错误 | 受保护的管理端点 | 使用 CLI/终端而非浏览器 |
+
+### 诊断
+
+**查看代理日志：**
+```bash
+# 若 logging-to-file: true
+tail -f logs/requests.log
+
+# 若运行在 tmux 中
+tmux attach-session -t proxy
+```
+
+**临时开启调试模式：**
+```yaml
+debug: true
+```
+
+**测试基础连通性：**
+```bash
+# 检查代理是否运行
+curl http://localhost:8317/v1/models
+
+# 检查 Amp 特定路由
+curl http://localhost:8317/api/provider/openai/v1/models
+```
+
+**验证 Amp 配置：**
+```bash
+# 检查 Amp 是否使用代理
+amp config get amp.url
+
+# 或检查环境变量
+echo $AMP_URL
+```
+
+### 安全清单
+
+- ✅ 保持 `amp-restrict-management-to-localhost: true`（默认）
+- ✅ 不要将代理暴露到公共网络（绑定到 localhost 或使用防火墙/VPN）
+- ✅ 使用 `amp login` 管理的 Amp 密钥文件（`~/.local/share/amp/secrets.json`）
+- ✅ 定期重新登录轮换 OAuth 令牌
+- ✅ 若处理敏感数据，使用加密磁盘存储配置和 auth-dir
+- ✅ 保持代理二进制为最新版本以获取安全修复
+
+## 其他资源
+
+- [CLIProxyAPI 主文档](https://help.router-for.me/)
+- [Amp CLI 官方手册](https://ampcode.com/manual)
+- [管理 API 参考](https://help.router-for.me/management/api)
+- [SDK 文档](sdk-usage.md)
+
+## 免责声明
+
+此集成仅用于个人或教育用途。使用反向代理或替代 API 基址可能违反提供商的服务条款。你需要对自己的使用方式负责。账号可能会被限速、锁定或封禁。软件不附带任何保证，使用风险自负。

examples/custom-provider/main.go

@@ -137,6 +137,10 @@ func (MyExecutor) Execute(ctx context.Context, a *coreauth.Auth, req clipexec.Re
 	return clipexec.Response{Payload: body}, nil
 }
 
+func (MyExecutor) CountTokens(context.Context, *coreauth.Auth, clipexec.Request, clipexec.Options) (clipexec.Response, error) {
+	return clipexec.Response{}, errors.New("count tokens not implemented")
+}
+
 func (MyExecutor) ExecuteStream(ctx context.Context, a *coreauth.Auth, req clipexec.Request, opts clipexec.Options) (<-chan clipexec.StreamChunk, error) {
 	ch := make(chan clipexec.StreamChunk, 1)
 	go func() {
@@ -146,10 +150,6 @@ func (MyExecutor) ExecuteStream(ctx context.Context, a *coreauth.Auth, req clipe
 	return ch, nil
 }
 
-func (MyExecutor) CountTokens(ctx context.Context, a *coreauth.Auth, req clipexec.Request, opts clipexec.Options) (clipexec.Response, error) {
-	return clipexec.Response{}, errors.New("not implemented")
-}
-
 func (MyExecutor) Refresh(ctx context.Context, a *coreauth.Auth) (*coreauth.Auth, error) {
 	return a, nil
 }

go.mod

@@ -3,6 +3,7 @@ module github.com/router-for-me/CLIProxyAPI/v6
 go 1.24.0
 
 require (
+	github.com/andybalholm/brotli v1.0.6
 	github.com/fsnotify/fsnotify v1.9.0
 	github.com/gin-gonic/gin v1.10.1
 	github.com/go-git/go-git/v6 v6.0.0-20251009132922-75a182125145
@@ -28,7 +29,6 @@ require (
 	cloud.google.com/go/compute/metadata v0.3.0 // indirect
 	github.com/Microsoft/go-winio v0.6.2 // indirect
 	github.com/ProtonMail/go-crypto v1.3.0 // indirect
-	github.com/andybalholm/brotli v1.0.6 // indirect
 	github.com/bytedance/sonic v1.11.6 // indirect
 	github.com/bytedance/sonic/loader v0.1.1 // indirect
 	github.com/cloudflare/circl v1.6.1 // indirect

internal/api/handlers/management/auth_files.go

@@ -220,6 +220,14 @@ func stopForwarderInstance(port int, forwarder *callbackForwarder) {
 	log.Infof("callback forwarder on port %d stopped", port)
 }
 
+func sanitizeAntigravityFileName(email string) string {
+	if strings.TrimSpace(email) == "" {
+		return "antigravity.json"
+	}
+	replacer := strings.NewReplacer("@", "_", ".", "_")
+	return fmt.Sprintf("antigravity-%s.json", replacer.Replace(email))
+}
+
 func (h *Handler) managementCallbackURL(path string) (string, error) {
 	if h == nil || h.cfg == nil || h.cfg.Port <= 0 {
 		return "", fmt.Errorf("server port is not configured")
@@ -227,7 +235,11 @@ func (h *Handler) managementCallbackURL(path string) (string, error) {
 	if !strings.HasPrefix(path, "/") {
 		path = "/" + path
 	}
-	return fmt.Sprintf("http://127.0.0.1:%d%s", h.cfg.Port, path), nil
+	scheme := "http"
+	if h.cfg.TLS.Enable {
+		scheme = "https"
+	}
+	return fmt.Sprintf("%s://127.0.0.1:%d%s", scheme, h.cfg.Port, path), nil
 }
 
 func (h *Handler) ListAuthFiles(c *gin.Context) {
@@ -292,6 +304,7 @@ func (h *Handler) buildAuthFileEntry(auth *coreauth.Auth) gin.H {
 	if auth == nil {
 		return nil
 	}
+	auth.EnsureIndex()
 	runtimeOnly := isRuntimeOnlyAuth(auth)
 	if runtimeOnly && (auth.Disabled || auth.Status == coreauth.StatusDisabled) {
 		return nil
@@ -306,6 +319,7 @@ func (h *Handler) buildAuthFileEntry(auth *coreauth.Auth) gin.H {
 	}
 	entry := gin.H{
 		"id":             auth.ID,
+		"auth_index":     auth.Index,
 		"name":           name,
 		"type":           strings.TrimSpace(auth.Provider),
 		"provider":       strings.TrimSpace(auth.Provider),
@@ -346,6 +360,10 @@ func (h *Handler) buildAuthFileEntry(auth *coreauth.Auth) gin.H {
 			entry["size"] = info.Size()
 			entry["modtime"] = info.ModTime()
 		} else if os.IsNotExist(err) {
+			// Hide credentials removed from disk but still lingering in memory.
+			if !runtimeOnly && (auth.Disabled || auth.Status == coreauth.StatusDisabled || strings.EqualFold(strings.TrimSpace(auth.StatusMessage), "removed via management api")) {
+				return nil
+			}
 			entry["source"] = "memory"
 		} else {
 			log.WithError(err).Warnf("failed to stat auth file %s", path)
@@ -695,14 +713,16 @@ func (h *Handler) RequestAnthropicToken(c *gin.Context) {
 	// Generate PKCE codes
 	pkceCodes, err := claude.GeneratePKCECodes()
 	if err != nil {
-		log.Fatalf("Failed to generate PKCE codes: %v", err)
+		log.Errorf("Failed to generate PKCE codes: %v", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate PKCE codes"})
 		return
 	}
 
 	// Generate random state parameter
 	state, err := misc.GenerateRandomState()
 	if err != nil {
-		log.Fatalf("Failed to generate state parameter: %v", err)
+		log.Errorf("Failed to generate state parameter: %v", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate state parameter"})
 		return
 	}
 
@@ -712,7 +732,8 @@ func (h *Handler) RequestAnthropicToken(c *gin.Context) {
 	// Generate authorization URL (then override redirect_uri to reuse server port)
 	authURL, state, err := anthropicAuth.GenerateAuthURL(state, pkceCodes)
 	if err != nil {
-		log.Fatalf("Failed to generate authorization URL: %v", err)
+		log.Errorf("Failed to generate authorization URL: %v", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate authorization url"})
 		return
 	}
 
@@ -854,7 +875,7 @@ func (h *Handler) RequestAnthropicToken(c *gin.Context) {
 		}
 		savedPath, errSave := h.saveTokenRecord(ctx, record)
 		if errSave != nil {
-			log.Fatalf("Failed to save authentication tokens: %v", errSave)
+			log.Errorf("Failed to save authentication tokens: %v", errSave)
 			oauthStatus[state] = "Failed to save authentication tokens"
 			return
 		}
@@ -873,6 +894,8 @@ func (h *Handler) RequestAnthropicToken(c *gin.Context) {
 
 func (h *Handler) RequestGeminiCLIToken(c *gin.Context) {
 	ctx := context.Background()
+	proxyHTTPClient := util.SetProxy(&h.cfg.SDKConfig, &http.Client{})
+	ctx = context.WithValue(ctx, oauth2.HTTPClient, proxyHTTPClient)
 
 	// Optional project ID from query
 	projectID := c.Query("project_id")
@@ -958,7 +981,7 @@ func (h *Handler) RequestGeminiCLIToken(c *gin.Context) {
 		requestedProjectID := strings.TrimSpace(projectID)
 
 		// Create token storage (mirrors internal/auth/gemini createTokenStorage)
-		httpClient := conf.Client(ctx, token)
+		authHTTPClient := conf.Client(ctx, token)
 		req, errNewRequest := http.NewRequestWithContext(ctx, "GET", "https://www.googleapis.com/oauth2/v1/userinfo?alt=json", nil)
 		if errNewRequest != nil {
 			log.Errorf("Could not get user info: %v", errNewRequest)
@@ -968,7 +991,7 @@ func (h *Handler) RequestGeminiCLIToken(c *gin.Context) {
 		req.Header.Set("Content-Type", "application/json")
 		req.Header.Set("Authorization", fmt.Sprintf("Bearer %s", token.AccessToken))
 
-		resp, errDo := httpClient.Do(req)
+		resp, errDo := authHTTPClient.Do(req)
 		if errDo != nil {
 			log.Errorf("Failed to execute request: %v", errDo)
 			oauthStatus[state] = "Failed to execute request"
@@ -1025,7 +1048,7 @@ func (h *Handler) RequestGeminiCLIToken(c *gin.Context) {
 		gemAuth := geminiAuth.NewGeminiAuth()
 		gemClient, errGetClient := gemAuth.GetAuthenticatedClient(ctx, &ts, h.cfg, true)
 		if errGetClient != nil {
-			log.Fatalf("failed to get authenticated client: %v", errGetClient)
+			log.Errorf("failed to get authenticated client: %v", errGetClient)
 			oauthStatus[state] = "Failed to get authenticated client"
 			return
 		}
@@ -1090,7 +1113,7 @@ func (h *Handler) RequestGeminiCLIToken(c *gin.Context) {
 		}
 		savedPath, errSave := h.saveTokenRecord(ctx, record)
 		if errSave != nil {
-			log.Fatalf("Failed to save token to file: %v", errSave)
+			log.Errorf("Failed to save token to file: %v", errSave)
 			oauthStatus[state] = "Failed to save token to file"
 			return
 		}
@@ -1111,14 +1134,16 @@ func (h *Handler) RequestCodexToken(c *gin.Context) {
 	// Generate PKCE codes
 	pkceCodes, err := codex.GeneratePKCECodes()
 	if err != nil {
-		log.Fatalf("Failed to generate PKCE codes: %v", err)
+		log.Errorf("Failed to generate PKCE codes: %v", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate PKCE codes"})
 		return
 	}
 
 	// Generate random state parameter
 	state, err := misc.GenerateRandomState()
 	if err != nil {
-		log.Fatalf("Failed to generate state parameter: %v", err)
+		log.Errorf("Failed to generate state parameter: %v", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate state parameter"})
 		return
 	}
 
@@ -1128,7 +1153,8 @@ func (h *Handler) RequestCodexToken(c *gin.Context) {
 	// Generate authorization URL
 	authURL, err := openaiAuth.GenerateAuthURL(state, pkceCodes)
 	if err != nil {
-		log.Fatalf("Failed to generate authorization URL: %v", err)
+		log.Errorf("Failed to generate authorization URL: %v", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate authorization url"})
 		return
 	}
 
@@ -1263,7 +1289,7 @@ func (h *Handler) RequestCodexToken(c *gin.Context) {
 		savedPath, errSave := h.saveTokenRecord(ctx, record)
 		if errSave != nil {
 			oauthStatus[state] = "Failed to save authentication tokens"
-			log.Fatalf("Failed to save authentication tokens: %v", errSave)
+			log.Errorf("Failed to save authentication tokens: %v", errSave)
 			return
 		}
 		fmt.Printf("Authentication successful! Token saved to %s\n", savedPath)
@@ -1278,6 +1304,240 @@ func (h *Handler) RequestCodexToken(c *gin.Context) {
 	c.JSON(200, gin.H{"status": "ok", "url": authURL, "state": state})
 }
 
+func (h *Handler) RequestAntigravityToken(c *gin.Context) {
+	const (
+		antigravityCallbackPort = 51121
+		antigravityClientID     = "1071006060591-tmhssin2h21lcre235vtolojh4g403ep.apps.googleusercontent.com"
+		antigravityClientSecret = "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"
+	)
+	var antigravityScopes = []string{
+		"https://www.googleapis.com/auth/cloud-platform",
+		"https://www.googleapis.com/auth/userinfo.email",
+		"https://www.googleapis.com/auth/userinfo.profile",
+		"https://www.googleapis.com/auth/cclog",
+		"https://www.googleapis.com/auth/experimentsandconfigs",
+	}
+
+	ctx := context.Background()
+
+	fmt.Println("Initializing Antigravity authentication...")
+
+	state, errState := misc.GenerateRandomState()
+	if errState != nil {
+		log.Errorf("Failed to generate state parameter: %v", errState)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate state parameter"})
+		return
+	}
+
+	redirectURI := fmt.Sprintf("http://localhost:%d/oauth-callback", antigravityCallbackPort)
+
+	params := url.Values{}
+	params.Set("access_type", "offline")
+	params.Set("client_id", antigravityClientID)
+	params.Set("prompt", "consent")
+	params.Set("redirect_uri", redirectURI)
+	params.Set("response_type", "code")
+	params.Set("scope", strings.Join(antigravityScopes, " "))
+	params.Set("state", state)
+	authURL := "https://accounts.google.com/o/oauth2/v2/auth?" + params.Encode()
+
+	isWebUI := isWebUIRequest(c)
+	if isWebUI {
+		targetURL, errTarget := h.managementCallbackURL("/antigravity/callback")
+		if errTarget != nil {
+			log.WithError(errTarget).Error("failed to compute antigravity callback target")
+			c.JSON(http.StatusInternalServerError, gin.H{"error": "callback server unavailable"})
+			return
+		}
+		if _, errStart := startCallbackForwarder(antigravityCallbackPort, "antigravity", targetURL); errStart != nil {
+			log.WithError(errStart).Error("failed to start antigravity callback forwarder")
+			c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to start callback server"})
+			return
+		}
+	}
+
+	go func() {
+		if isWebUI {
+			defer stopCallbackForwarder(antigravityCallbackPort)
+		}
+
+		waitFile := filepath.Join(h.cfg.AuthDir, fmt.Sprintf(".oauth-antigravity-%s.oauth", state))
+		deadline := time.Now().Add(5 * time.Minute)
+		var authCode string
+		for {
+			if time.Now().After(deadline) {
+				log.Error("oauth flow timed out")
+				oauthStatus[state] = "OAuth flow timed out"
+				return
+			}
+			if data, errReadFile := os.ReadFile(waitFile); errReadFile == nil {
+				var payload map[string]string
+				_ = json.Unmarshal(data, &payload)
+				_ = os.Remove(waitFile)
+				if errStr := strings.TrimSpace(payload["error"]); errStr != "" {
+					log.Errorf("Authentication failed: %s", errStr)
+					oauthStatus[state] = "Authentication failed"
+					return
+				}
+				if payloadState := strings.TrimSpace(payload["state"]); payloadState != "" && payloadState != state {
+					log.Errorf("Authentication failed: state mismatch")
+					oauthStatus[state] = "Authentication failed: state mismatch"
+					return
+				}
+				authCode = strings.TrimSpace(payload["code"])
+				if authCode == "" {
+					log.Error("Authentication failed: code not found")
+					oauthStatus[state] = "Authentication failed: code not found"
+					return
+				}
+				break
+			}
+			time.Sleep(500 * time.Millisecond)
+		}
+
+		httpClient := util.SetProxy(&h.cfg.SDKConfig, &http.Client{})
+		form := url.Values{}
+		form.Set("code", authCode)
+		form.Set("client_id", antigravityClientID)
+		form.Set("client_secret", antigravityClientSecret)
+		form.Set("redirect_uri", redirectURI)
+		form.Set("grant_type", "authorization_code")
+
+		req, errNewRequest := http.NewRequestWithContext(ctx, http.MethodPost, "https://oauth2.googleapis.com/token", strings.NewReader(form.Encode()))
+		if errNewRequest != nil {
+			log.Errorf("Failed to build token request: %v", errNewRequest)
+			oauthStatus[state] = "Failed to build token request"
+			return
+		}
+		req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
+
+		resp, errDo := httpClient.Do(req)
+		if errDo != nil {
+			log.Errorf("Failed to execute token request: %v", errDo)
+			oauthStatus[state] = "Failed to exchange token"
+			return
+		}
+		defer func() {
+			if errClose := resp.Body.Close(); errClose != nil {
+				log.Errorf("antigravity token exchange close error: %v", errClose)
+			}
+		}()
+
+		if resp.StatusCode < http.StatusOK || resp.StatusCode >= http.StatusMultipleChoices {
+			bodyBytes, _ := io.ReadAll(resp.Body)
+			log.Errorf("Antigravity token exchange failed with status %d: %s", resp.StatusCode, string(bodyBytes))
+			oauthStatus[state] = fmt.Sprintf("Token exchange failed: %d", resp.StatusCode)
+			return
+		}
+
+		var tokenResp struct {
+			AccessToken  string `json:"access_token"`
+			RefreshToken string `json:"refresh_token"`
+			ExpiresIn    int64  `json:"expires_in"`
+			TokenType    string `json:"token_type"`
+		}
+		if errDecode := json.NewDecoder(resp.Body).Decode(&tokenResp); errDecode != nil {
+			log.Errorf("Failed to parse token response: %v", errDecode)
+			oauthStatus[state] = "Failed to parse token response"
+			return
+		}
+
+		email := ""
+		if strings.TrimSpace(tokenResp.AccessToken) != "" {
+			infoReq, errInfoReq := http.NewRequestWithContext(ctx, http.MethodGet, "https://www.googleapis.com/oauth2/v1/userinfo?alt=json", nil)
+			if errInfoReq != nil {
+				log.Errorf("Failed to build user info request: %v", errInfoReq)
+				oauthStatus[state] = "Failed to build user info request"
+				return
+			}
+			infoReq.Header.Set("Authorization", "Bearer "+tokenResp.AccessToken)
+
+			infoResp, errInfo := httpClient.Do(infoReq)
+			if errInfo != nil {
+				log.Errorf("Failed to execute user info request: %v", errInfo)
+				oauthStatus[state] = "Failed to execute user info request"
+				return
+			}
+			defer func() {
+				if errClose := infoResp.Body.Close(); errClose != nil {
+					log.Errorf("antigravity user info close error: %v", errClose)
+				}
+			}()
+
+			if infoResp.StatusCode >= http.StatusOK && infoResp.StatusCode < http.StatusMultipleChoices {
+				var infoPayload struct {
+					Email string `json:"email"`
+				}
+				if errDecodeInfo := json.NewDecoder(infoResp.Body).Decode(&infoPayload); errDecodeInfo == nil {
+					email = strings.TrimSpace(infoPayload.Email)
+				}
+			} else {
+				bodyBytes, _ := io.ReadAll(infoResp.Body)
+				log.Errorf("User info request failed with status %d: %s", infoResp.StatusCode, string(bodyBytes))
+				oauthStatus[state] = fmt.Sprintf("User info request failed: %d", infoResp.StatusCode)
+				return
+			}
+		}
+
+		projectID := ""
+		if strings.TrimSpace(tokenResp.AccessToken) != "" {
+			fetchedProjectID, errProject := sdkAuth.FetchAntigravityProjectID(ctx, tokenResp.AccessToken, httpClient)
+			if errProject != nil {
+				log.Warnf("antigravity: failed to fetch project ID: %v", errProject)
+			} else {
+				projectID = fetchedProjectID
+				log.Infof("antigravity: obtained project ID %s", projectID)
+			}
+		}
+
+		now := time.Now()
+		metadata := map[string]any{
+			"type":          "antigravity",
+			"access_token":  tokenResp.AccessToken,
+			"refresh_token": tokenResp.RefreshToken,
+			"expires_in":    tokenResp.ExpiresIn,
+			"timestamp":     now.UnixMilli(),
+			"expired":       now.Add(time.Duration(tokenResp.ExpiresIn) * time.Second).Format(time.RFC3339),
+		}
+		if email != "" {
+			metadata["email"] = email
+		}
+		if projectID != "" {
+			metadata["project_id"] = projectID
+		}
+
+		fileName := sanitizeAntigravityFileName(email)
+		label := strings.TrimSpace(email)
+		if label == "" {
+			label = "antigravity"
+		}
+
+		record := &coreauth.Auth{
+			ID:       fileName,
+			Provider: "antigravity",
+			FileName: fileName,
+			Label:    label,
+			Metadata: metadata,
+		}
+		savedPath, errSave := h.saveTokenRecord(ctx, record)
+		if errSave != nil {
+			log.Errorf("Failed to save token to file: %v", errSave)
+			oauthStatus[state] = "Failed to save token to file"
+			return
+		}
+
+		delete(oauthStatus, state)
+		fmt.Printf("Authentication successful! Token saved to %s\n", savedPath)
+		if projectID != "" {
+			fmt.Printf("Using GCP project: %s\n", projectID)
+		}
+		fmt.Println("You can now use Antigravity services through this CLI")
+	}()
+
+	oauthStatus[state] = ""
+	c.JSON(200, gin.H{"status": "ok", "url": authURL, "state": state})
+}
+
 func (h *Handler) RequestQwenToken(c *gin.Context) {
 	ctx := context.Background()
 
@@ -1290,7 +1550,8 @@ func (h *Handler) RequestQwenToken(c *gin.Context) {
 	// Generate authorization URL
 	deviceFlow, err := qwenAuth.InitiateDeviceFlow(ctx)
 	if err != nil {
-		log.Fatalf("Failed to generate authorization URL: %v", err)
+		log.Errorf("Failed to generate authorization URL: %v", err)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate authorization url"})
 		return
 	}
 	authURL := deviceFlow.VerificationURIComplete
@@ -1317,7 +1578,7 @@ func (h *Handler) RequestQwenToken(c *gin.Context) {
 		}
 		savedPath, errSave := h.saveTokenRecord(ctx, record)
 		if errSave != nil {
-			log.Fatalf("Failed to save authentication tokens: %v", errSave)
+			log.Errorf("Failed to save authentication tokens: %v", errSave)
 			oauthStatus[state] = "Failed to save authentication tokens"
 			return
 		}
@@ -1421,7 +1682,7 @@ func (h *Handler) RequestIFlowToken(c *gin.Context) {
 		savedPath, errSave := h.saveTokenRecord(ctx, record)
 		if errSave != nil {
 			oauthStatus[state] = "Failed to save authentication tokens"
-			log.Fatalf("Failed to save authentication tokens: %v", errSave)
+			log.Errorf("Failed to save authentication tokens: %v", errSave)
 			return
 		}
 
@@ -1437,6 +1698,87 @@ func (h *Handler) RequestIFlowToken(c *gin.Context) {
 	c.JSON(http.StatusOK, gin.H{"status": "ok", "url": authURL, "state": state})
 }
 
+func (h *Handler) RequestIFlowCookieToken(c *gin.Context) {
+	ctx := context.Background()
+
+	var payload struct {
+		Cookie string `json:"cookie"`
+	}
+	if err := c.ShouldBindJSON(&payload); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"status": "error", "error": "cookie is required"})
+		return
+	}
+
+	cookieValue := strings.TrimSpace(payload.Cookie)
+
+	if cookieValue == "" {
+		c.JSON(http.StatusBadRequest, gin.H{"status": "error", "error": "cookie is required"})
+		return
+	}
+
+	cookieValue, errNormalize := iflowauth.NormalizeCookie(cookieValue)
+	if errNormalize != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"status": "error", "error": errNormalize.Error()})
+		return
+	}
+
+	authSvc := iflowauth.NewIFlowAuth(h.cfg)
+	tokenData, errAuth := authSvc.AuthenticateWithCookie(ctx, cookieValue)
+	if errAuth != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"status": "error", "error": errAuth.Error()})
+		return
+	}
+
+	tokenData.Cookie = cookieValue
+
+	tokenStorage := authSvc.CreateCookieTokenStorage(tokenData)
+	email := strings.TrimSpace(tokenStorage.Email)
+	if email == "" {
+		c.JSON(http.StatusBadRequest, gin.H{"status": "error", "error": "failed to extract email from token"})
+		return
+	}
+
+	fileName := iflowauth.SanitizeIFlowFileName(email)
+	if fileName == "" {
+		fileName = fmt.Sprintf("iflow-%d", time.Now().UnixMilli())
+	}
+
+	tokenStorage.Email = email
+
+	record := &coreauth.Auth{
+		ID:       fmt.Sprintf("iflow-%s.json", fileName),
+		Provider: "iflow",
+		FileName: fmt.Sprintf("iflow-%s.json", fileName),
+		Storage:  tokenStorage,
+		Metadata: map[string]any{
+			"email":        email,
+			"api_key":      tokenStorage.APIKey,
+			"expired":      tokenStorage.Expire,
+			"cookie":       tokenStorage.Cookie,
+			"type":         tokenStorage.Type,
+			"last_refresh": tokenStorage.LastRefresh,
+		},
+		Attributes: map[string]string{
+			"api_key": tokenStorage.APIKey,
+		},
+	}
+
+	savedPath, errSave := h.saveTokenRecord(ctx, record)
+	if errSave != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"status": "error", "error": "failed to save authentication tokens"})
+		return
+	}
+
+	fmt.Printf("iFlow cookie authentication successful. Token saved to %s\n", savedPath)
+	c.JSON(http.StatusOK, gin.H{
+		"status":     "ok",
+		"saved_path": savedPath,
+		"email":      email,
+		"expired":    tokenStorage.Expire,
+		"type":       tokenStorage.Type,
+	})
+}
+
 type projectSelectionRequiredError struct{}
 
 func (e *projectSelectionRequiredError) Error() string {
@@ -1769,6 +2111,7 @@ func checkCloudAPIIsEnabled(ctx context.Context, httpClient *http.Client, projec
 				continue
 			}
 		}
+		_ = resp.Body.Close()
 		return false, fmt.Errorf("project activation required: %s", errMessage)
 	}
 	return true, nil

internal/api/handlers/management/config_basic.go

@@ -1,43 +1,95 @@
 package management
 
 import (
+	"encoding/json"
+	"fmt"
 	"io"
 	"net/http"
 	"os"
 	"path/filepath"
+	"strings"
+	"time"
 
 	"github.com/gin-gonic/gin"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
+	sdkconfig "github.com/router-for-me/CLIProxyAPI/v6/sdk/config"
+	log "github.com/sirupsen/logrus"
 	"gopkg.in/yaml.v3"
 )
 
+const (
+	latestReleaseURL       = "https://api.github.com/repos/router-for-me/CLIProxyAPI/releases/latest"
+	latestReleaseUserAgent = "CLIProxyAPI"
+)
+
 func (h *Handler) GetConfig(c *gin.Context) {
 	if h == nil || h.cfg == nil {
 		c.JSON(200, gin.H{})
 		return
 	}
 	cfgCopy := *h.cfg
-	cfgCopy.GlAPIKey = geminiKeyStringsFromConfig(h.cfg)
 	c.JSON(200, &cfgCopy)
 }
 
-func (h *Handler) GetConfigYAML(c *gin.Context) {
-	data, err := os.ReadFile(h.configFilePath)
+type releaseInfo struct {
+	TagName string `json:"tag_name"`
+	Name    string `json:"name"`
+}
+
+// GetLatestVersion returns the latest release version from GitHub without downloading assets.
+func (h *Handler) GetLatestVersion(c *gin.Context) {
+	client := &http.Client{Timeout: 10 * time.Second}
+	proxyURL := ""
+	if h != nil && h.cfg != nil {
+		proxyURL = strings.TrimSpace(h.cfg.ProxyURL)
+	}
+	if proxyURL != "" {
+		sdkCfg := &sdkconfig.SDKConfig{ProxyURL: proxyURL}
+		util.SetProxy(sdkCfg, client)
+	}
+
+	req, err := http.NewRequestWithContext(c.Request.Context(), http.MethodGet, latestReleaseURL, nil)
 	if err != nil {
-		c.JSON(http.StatusInternalServerError, gin.H{"error": "read_failed", "message": err.Error()})
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "request_create_failed", "message": err.Error()})
 		return
 	}
-	var node yaml.Node
-	if err = yaml.Unmarshal(data, &node); err != nil {
-		c.JSON(http.StatusInternalServerError, gin.H{"error": "parse_failed", "message": err.Error()})
+	req.Header.Set("Accept", "application/vnd.github+json")
+	req.Header.Set("User-Agent", latestReleaseUserAgent)
+
+	resp, err := client.Do(req)
+	if err != nil {
+		c.JSON(http.StatusBadGateway, gin.H{"error": "request_failed", "message": err.Error()})
 		return
 	}
-	c.Header("Content-Type", "application/yaml; charset=utf-8")
-	c.Header("Vary", "format, Accept")
-	enc := yaml.NewEncoder(c.Writer)
-	enc.SetIndent(2)
-	_ = enc.Encode(&node)
-	_ = enc.Close()
+	defer func() {
+		if errClose := resp.Body.Close(); errClose != nil {
+			log.WithError(errClose).Debug("failed to close latest version response body")
+		}
+	}()
+
+	if resp.StatusCode != http.StatusOK {
+		body, _ := io.ReadAll(io.LimitReader(resp.Body, 1024))
+		c.JSON(http.StatusBadGateway, gin.H{"error": "unexpected_status", "message": fmt.Sprintf("status %d: %s", resp.StatusCode, strings.TrimSpace(string(body)))})
+		return
+	}
+
+	var info releaseInfo
+	if errDecode := json.NewDecoder(resp.Body).Decode(&info); errDecode != nil {
+		c.JSON(http.StatusBadGateway, gin.H{"error": "decode_failed", "message": errDecode.Error()})
+		return
+	}
+
+	version := strings.TrimSpace(info.TagName)
+	if version == "" {
+		version = strings.TrimSpace(info.Name)
+	}
+	if version == "" {
+		c.JSON(http.StatusBadGateway, gin.H{"error": "invalid_response", "message": "missing release version"})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{"latest-version": version})
 }
 
 func WriteConfig(path string, data []byte) error {
@@ -111,9 +163,9 @@ func (h *Handler) PutConfigYAML(c *gin.Context) {
 	c.JSON(http.StatusOK, gin.H{"ok": true, "changed": []string{"config"}})
 }
 
-// GetConfigFile returns the raw config.yaml file bytes without re-encoding.
+// GetConfigYAML returns the raw config.yaml file bytes without re-encoding.
 // It preserves comments and original formatting/styles.
-func (h *Handler) GetConfigFile(c *gin.Context) {
+func (h *Handler) GetConfigYAML(c *gin.Context) {
 	data, err := os.ReadFile(h.configFilePath)
 	if err != nil {
 		if os.IsNotExist(err) {
@@ -172,6 +224,14 @@ func (h *Handler) PutRequestRetry(c *gin.Context) {
 	h.updateIntField(c, func(v int) { h.cfg.RequestRetry = v })
 }
 
+// Max retry interval
+func (h *Handler) GetMaxRetryInterval(c *gin.Context) {
+	c.JSON(200, gin.H{"max-retry-interval": h.cfg.MaxRetryInterval})
+}
+func (h *Handler) PutMaxRetryInterval(c *gin.Context) {
+	h.updateIntField(c, func(v int) { h.cfg.MaxRetryInterval = v })
+}
+
 // Proxy URL
 func (h *Handler) GetProxyURL(c *gin.Context) { c.JSON(200, gin.H{"proxy-url": h.cfg.ProxyURL}) }
 func (h *Handler) PutProxyURL(c *gin.Context) {

internal/api/handlers/management/config_lists.go

@@ -104,53 +104,6 @@ func (h *Handler) deleteFromStringList(c *gin.Context, target *[]string, after f
 	c.JSON(400, gin.H{"error": "missing index or value"})
 }
 
-func sanitizeStringSlice(in []string) []string {
-	out := make([]string, 0, len(in))
-	for i := range in {
-		if trimmed := strings.TrimSpace(in[i]); trimmed != "" {
-			out = append(out, trimmed)
-		}
-	}
-	return out
-}
-
-func geminiKeyStringsFromConfig(cfg *config.Config) []string {
-	if cfg == nil || len(cfg.GeminiKey) == 0 {
-		return nil
-	}
-	out := make([]string, 0, len(cfg.GeminiKey))
-	for i := range cfg.GeminiKey {
-		if key := strings.TrimSpace(cfg.GeminiKey[i].APIKey); key != "" {
-			out = append(out, key)
-		}
-	}
-	return out
-}
-
-func (h *Handler) applyLegacyKeys(keys []string) {
-	if h == nil || h.cfg == nil {
-		return
-	}
-	sanitized := sanitizeStringSlice(keys)
-	existing := make(map[string]config.GeminiKey, len(h.cfg.GeminiKey))
-	for _, entry := range h.cfg.GeminiKey {
-		if key := strings.TrimSpace(entry.APIKey); key != "" {
-			existing[key] = entry
-		}
-	}
-	newList := make([]config.GeminiKey, 0, len(sanitized))
-	for _, key := range sanitized {
-		if entry, ok := existing[key]; ok {
-			newList = append(newList, entry)
-		} else {
-			newList = append(newList, config.GeminiKey{APIKey: key})
-		}
-	}
-	h.cfg.GeminiKey = newList
-	h.cfg.GlAPIKey = sanitized
-	h.cfg.SanitizeGeminiKeys()
-}
-
 // api-keys
 func (h *Handler) GetAPIKeys(c *gin.Context) { c.JSON(200, gin.H{"api-keys": h.cfg.APIKeys}) }
 func (h *Handler) PutAPIKeys(c *gin.Context) {
@@ -166,24 +119,6 @@ func (h *Handler) DeleteAPIKeys(c *gin.Context) {
 	h.deleteFromStringList(c, &h.cfg.APIKeys, func() { h.cfg.Access.Providers = nil })
 }
 
-// generative-language-api-key
-func (h *Handler) GetGlKeys(c *gin.Context) {
-	c.JSON(200, gin.H{"generative-language-api-key": geminiKeyStringsFromConfig(h.cfg)})
-}
-func (h *Handler) PutGlKeys(c *gin.Context) {
-	h.putStringList(c, func(v []string) {
-		h.applyLegacyKeys(v)
-	}, nil)
-}
-func (h *Handler) PatchGlKeys(c *gin.Context) {
-	target := append([]string(nil), geminiKeyStringsFromConfig(h.cfg)...)
-	h.patchStringList(c, &target, func() { h.applyLegacyKeys(target) })
-}
-func (h *Handler) DeleteGlKeys(c *gin.Context) {
-	target := append([]string(nil), geminiKeyStringsFromConfig(h.cfg)...)
-	h.deleteFromStringList(c, &target, func() { h.applyLegacyKeys(target) })
-}
-
 // gemini-api-key: []GeminiKey
 func (h *Handler) GetGeminiKeys(c *gin.Context) {
 	c.JSON(200, gin.H{"gemini-api-key": h.cfg.GeminiKey})
@@ -223,6 +158,7 @@ func (h *Handler) PatchGeminiKey(c *gin.Context) {
 	value.APIKey = strings.TrimSpace(value.APIKey)
 	value.BaseURL = strings.TrimSpace(value.BaseURL)
 	value.ProxyURL = strings.TrimSpace(value.ProxyURL)
+	value.ExcludedModels = config.NormalizeExcludedModels(value.ExcludedModels)
 	if value.APIKey == "" {
 		// Treat empty API key as delete.
 		if body.Index != nil && *body.Index >= 0 && *body.Index < len(h.cfg.GeminiKey) {
@@ -408,12 +344,9 @@ func (h *Handler) PutOpenAICompat(c *gin.Context) {
 		}
 		arr = obj.Items
 	}
-	for i := range arr {
-		normalizeOpenAICompatibilityEntry(&arr[i])
-	}
-	// Filter out providers with empty base-url -> remove provider entirely
 	filtered := make([]config.OpenAICompatibility, 0, len(arr))
 	for i := range arr {
+		normalizeOpenAICompatibilityEntry(&arr[i])
 		if strings.TrimSpace(arr[i].BaseURL) != "" {
 			filtered = append(filtered, arr[i])
 		}
@@ -505,6 +438,91 @@ func (h *Handler) DeleteOpenAICompat(c *gin.Context) {
 	c.JSON(400, gin.H{"error": "missing name or index"})
 }
 
+// oauth-excluded-models: map[string][]string
+func (h *Handler) GetOAuthExcludedModels(c *gin.Context) {
+	c.JSON(200, gin.H{"oauth-excluded-models": config.NormalizeOAuthExcludedModels(h.cfg.OAuthExcludedModels)})
+}
+
+func (h *Handler) PutOAuthExcludedModels(c *gin.Context) {
+	data, err := c.GetRawData()
+	if err != nil {
+		c.JSON(400, gin.H{"error": "failed to read body"})
+		return
+	}
+	var entries map[string][]string
+	if err = json.Unmarshal(data, &entries); err != nil {
+		var wrapper struct {
+			Items map[string][]string `json:"items"`
+		}
+		if err2 := json.Unmarshal(data, &wrapper); err2 != nil {
+			c.JSON(400, gin.H{"error": "invalid body"})
+			return
+		}
+		entries = wrapper.Items
+	}
+	h.cfg.OAuthExcludedModels = config.NormalizeOAuthExcludedModels(entries)
+	h.persist(c)
+}
+
+func (h *Handler) PatchOAuthExcludedModels(c *gin.Context) {
+	var body struct {
+		Provider *string  `json:"provider"`
+		Models   []string `json:"models"`
+	}
+	if err := c.ShouldBindJSON(&body); err != nil || body.Provider == nil {
+		c.JSON(400, gin.H{"error": "invalid body"})
+		return
+	}
+	provider := strings.ToLower(strings.TrimSpace(*body.Provider))
+	if provider == "" {
+		c.JSON(400, gin.H{"error": "invalid provider"})
+		return
+	}
+	normalized := config.NormalizeExcludedModels(body.Models)
+	if len(normalized) == 0 {
+		if h.cfg.OAuthExcludedModels == nil {
+			c.JSON(404, gin.H{"error": "provider not found"})
+			return
+		}
+		if _, ok := h.cfg.OAuthExcludedModels[provider]; !ok {
+			c.JSON(404, gin.H{"error": "provider not found"})
+			return
+		}
+		delete(h.cfg.OAuthExcludedModels, provider)
+		if len(h.cfg.OAuthExcludedModels) == 0 {
+			h.cfg.OAuthExcludedModels = nil
+		}
+		h.persist(c)
+		return
+	}
+	if h.cfg.OAuthExcludedModels == nil {
+		h.cfg.OAuthExcludedModels = make(map[string][]string)
+	}
+	h.cfg.OAuthExcludedModels[provider] = normalized
+	h.persist(c)
+}
+
+func (h *Handler) DeleteOAuthExcludedModels(c *gin.Context) {
+	provider := strings.ToLower(strings.TrimSpace(c.Query("provider")))
+	if provider == "" {
+		c.JSON(400, gin.H{"error": "missing provider"})
+		return
+	}
+	if h.cfg.OAuthExcludedModels == nil {
+		c.JSON(404, gin.H{"error": "provider not found"})
+		return
+	}
+	if _, ok := h.cfg.OAuthExcludedModels[provider]; !ok {
+		c.JSON(404, gin.H{"error": "provider not found"})
+		return
+	}
+	delete(h.cfg.OAuthExcludedModels, provider)
+	if len(h.cfg.OAuthExcludedModels) == 0 {
+		h.cfg.OAuthExcludedModels = nil
+	}
+	h.persist(c)
+}
+
 // codex-api-key: []CodexKey
 func (h *Handler) GetCodexKeys(c *gin.Context) {
 	c.JSON(200, gin.H{"codex-api-key": h.cfg.CodexKey})
@@ -534,6 +552,7 @@ func (h *Handler) PutCodexKeys(c *gin.Context) {
 		entry.BaseURL = strings.TrimSpace(entry.BaseURL)
 		entry.ProxyURL = strings.TrimSpace(entry.ProxyURL)
 		entry.Headers = config.NormalizeHeaders(entry.Headers)
+		entry.ExcludedModels = config.NormalizeExcludedModels(entry.ExcludedModels)
 		if entry.BaseURL == "" {
 			continue
 		}
@@ -558,6 +577,7 @@ func (h *Handler) PatchCodexKey(c *gin.Context) {
 	value.BaseURL = strings.TrimSpace(value.BaseURL)
 	value.ProxyURL = strings.TrimSpace(value.ProxyURL)
 	value.Headers = config.NormalizeHeaders(value.Headers)
+	value.ExcludedModels = config.NormalizeExcludedModels(value.ExcludedModels)
 	// If base-url becomes empty, delete instead of update
 	if value.BaseURL == "" {
 		if body.Index != nil && *body.Index >= 0 && *body.Index < len(h.cfg.CodexKey) {
@@ -644,21 +664,6 @@ func normalizeOpenAICompatibilityEntry(entry *config.OpenAICompatibility) {
 			existing[trimmed] = struct{}{}
 		}
 	}
-	if len(entry.APIKeys) == 0 {
-		return
-	}
-	for _, legacyKey := range entry.APIKeys {
-		trimmed := strings.TrimSpace(legacyKey)
-		if trimmed == "" {
-			continue
-		}
-		if _, ok := existing[trimmed]; ok {
-			continue
-		}
-		entry.APIKeyEntries = append(entry.APIKeyEntries, config.OpenAICompatibilityAPIKey{APIKey: trimmed})
-		existing[trimmed] = struct{}{}
-	}
-	entry.APIKeys = nil
 }
 
 func normalizedOpenAICompatibilityEntries(entries []config.OpenAICompatibility) []config.OpenAICompatibility {
@@ -671,9 +676,6 @@ func normalizedOpenAICompatibilityEntries(entries []config.OpenAICompatibility)
 		if len(copyEntry.APIKeyEntries) > 0 {
 			copyEntry.APIKeyEntries = append([]config.OpenAICompatibilityAPIKey(nil), copyEntry.APIKeyEntries...)
 		}
-		if len(copyEntry.APIKeys) > 0 {
-			copyEntry.APIKeys = append([]string(nil), copyEntry.APIKeys...)
-		}
 		normalizeOpenAICompatibilityEntry(&copyEntry)
 		out[i] = copyEntry
 	}
@@ -688,6 +690,7 @@ func normalizeClaudeKey(entry *config.ClaudeKey) {
 	entry.BaseURL = strings.TrimSpace(entry.BaseURL)
 	entry.ProxyURL = strings.TrimSpace(entry.ProxyURL)
 	entry.Headers = config.NormalizeHeaders(entry.Headers)
+	entry.ExcludedModels = config.NormalizeExcludedModels(entry.ExcludedModels)
 	if len(entry.Models) == 0 {
 		return
 	}
@@ -703,3 +706,155 @@ func normalizeClaudeKey(entry *config.ClaudeKey) {
 	}
 	entry.Models = normalized
 }
+
+// GetAmpCode returns the complete ampcode configuration.
+func (h *Handler) GetAmpCode(c *gin.Context) {
+	if h == nil || h.cfg == nil {
+		c.JSON(200, gin.H{"ampcode": config.AmpCode{}})
+		return
+	}
+	c.JSON(200, gin.H{"ampcode": h.cfg.AmpCode})
+}
+
+// GetAmpUpstreamURL returns the ampcode upstream URL.
+func (h *Handler) GetAmpUpstreamURL(c *gin.Context) {
+	if h == nil || h.cfg == nil {
+		c.JSON(200, gin.H{"upstream-url": ""})
+		return
+	}
+	c.JSON(200, gin.H{"upstream-url": h.cfg.AmpCode.UpstreamURL})
+}
+
+// PutAmpUpstreamURL updates the ampcode upstream URL.
+func (h *Handler) PutAmpUpstreamURL(c *gin.Context) {
+	h.updateStringField(c, func(v string) { h.cfg.AmpCode.UpstreamURL = strings.TrimSpace(v) })
+}
+
+// DeleteAmpUpstreamURL clears the ampcode upstream URL.
+func (h *Handler) DeleteAmpUpstreamURL(c *gin.Context) {
+	h.cfg.AmpCode.UpstreamURL = ""
+	h.persist(c)
+}
+
+// GetAmpUpstreamAPIKey returns the ampcode upstream API key.
+func (h *Handler) GetAmpUpstreamAPIKey(c *gin.Context) {
+	if h == nil || h.cfg == nil {
+		c.JSON(200, gin.H{"upstream-api-key": ""})
+		return
+	}
+	c.JSON(200, gin.H{"upstream-api-key": h.cfg.AmpCode.UpstreamAPIKey})
+}
+
+// PutAmpUpstreamAPIKey updates the ampcode upstream API key.
+func (h *Handler) PutAmpUpstreamAPIKey(c *gin.Context) {
+	h.updateStringField(c, func(v string) { h.cfg.AmpCode.UpstreamAPIKey = strings.TrimSpace(v) })
+}
+
+// DeleteAmpUpstreamAPIKey clears the ampcode upstream API key.
+func (h *Handler) DeleteAmpUpstreamAPIKey(c *gin.Context) {
+	h.cfg.AmpCode.UpstreamAPIKey = ""
+	h.persist(c)
+}
+
+// GetAmpRestrictManagementToLocalhost returns the localhost restriction setting.
+func (h *Handler) GetAmpRestrictManagementToLocalhost(c *gin.Context) {
+	if h == nil || h.cfg == nil {
+		c.JSON(200, gin.H{"restrict-management-to-localhost": true})
+		return
+	}
+	c.JSON(200, gin.H{"restrict-management-to-localhost": h.cfg.AmpCode.RestrictManagementToLocalhost})
+}
+
+// PutAmpRestrictManagementToLocalhost updates the localhost restriction setting.
+func (h *Handler) PutAmpRestrictManagementToLocalhost(c *gin.Context) {
+	h.updateBoolField(c, func(v bool) { h.cfg.AmpCode.RestrictManagementToLocalhost = v })
+}
+
+// GetAmpModelMappings returns the ampcode model mappings.
+func (h *Handler) GetAmpModelMappings(c *gin.Context) {
+	if h == nil || h.cfg == nil {
+		c.JSON(200, gin.H{"model-mappings": []config.AmpModelMapping{}})
+		return
+	}
+	c.JSON(200, gin.H{"model-mappings": h.cfg.AmpCode.ModelMappings})
+}
+
+// PutAmpModelMappings replaces all ampcode model mappings.
+func (h *Handler) PutAmpModelMappings(c *gin.Context) {
+	var body struct {
+		Value []config.AmpModelMapping `json:"value"`
+	}
+	if err := c.ShouldBindJSON(&body); err != nil {
+		c.JSON(400, gin.H{"error": "invalid body"})
+		return
+	}
+	h.cfg.AmpCode.ModelMappings = body.Value
+	h.persist(c)
+}
+
+// PatchAmpModelMappings adds or updates model mappings.
+func (h *Handler) PatchAmpModelMappings(c *gin.Context) {
+	var body struct {
+		Value []config.AmpModelMapping `json:"value"`
+	}
+	if err := c.ShouldBindJSON(&body); err != nil {
+		c.JSON(400, gin.H{"error": "invalid body"})
+		return
+	}
+
+	existing := make(map[string]int)
+	for i, m := range h.cfg.AmpCode.ModelMappings {
+		existing[strings.TrimSpace(m.From)] = i
+	}
+
+	for _, newMapping := range body.Value {
+		from := strings.TrimSpace(newMapping.From)
+		if idx, ok := existing[from]; ok {
+			h.cfg.AmpCode.ModelMappings[idx] = newMapping
+		} else {
+			h.cfg.AmpCode.ModelMappings = append(h.cfg.AmpCode.ModelMappings, newMapping)
+			existing[from] = len(h.cfg.AmpCode.ModelMappings) - 1
+		}
+	}
+	h.persist(c)
+}
+
+// DeleteAmpModelMappings removes specified model mappings by "from" field.
+func (h *Handler) DeleteAmpModelMappings(c *gin.Context) {
+	var body struct {
+		Value []string `json:"value"`
+	}
+	if err := c.ShouldBindJSON(&body); err != nil || len(body.Value) == 0 {
+		h.cfg.AmpCode.ModelMappings = nil
+		h.persist(c)
+		return
+	}
+
+	toRemove := make(map[string]bool)
+	for _, from := range body.Value {
+		toRemove[strings.TrimSpace(from)] = true
+	}
+
+	newMappings := make([]config.AmpModelMapping, 0, len(h.cfg.AmpCode.ModelMappings))
+	for _, m := range h.cfg.AmpCode.ModelMappings {
+		if !toRemove[strings.TrimSpace(m.From)] {
+			newMappings = append(newMappings, m)
+		}
+	}
+	h.cfg.AmpCode.ModelMappings = newMappings
+	h.persist(c)
+}
+
+// GetAmpForceModelMappings returns whether model mappings are forced.
+func (h *Handler) GetAmpForceModelMappings(c *gin.Context) {
+	if h == nil || h.cfg == nil {
+		c.JSON(200, gin.H{"force-model-mappings": false})
+		return
+	}
+	c.JSON(200, gin.H{"force-model-mappings": h.cfg.AmpCode.ForceModelMappings})
+}
+
+// PutAmpForceModelMappings updates the force model mappings setting.
+func (h *Handler) PutAmpForceModelMappings(c *gin.Context) {
+	h.updateBoolField(c, func(v bool) { h.cfg.AmpCode.ForceModelMappings = v })
+}

internal/api/handlers/management/handler.go

@@ -240,16 +240,6 @@ func (h *Handler) updateBoolField(c *gin.Context, set func(bool)) {
 		Value *bool `json:"value"`
 	}
 	if err := c.ShouldBindJSON(&body); err != nil || body.Value == nil {
-		var m map[string]any
-		if err2 := c.ShouldBindJSON(&m); err2 == nil {
-			for _, v := range m {
-				if b, ok := v.(bool); ok {
-					set(b)
-					h.persist(c)
-					return
-				}
-			}
-		}
 		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid body"})
 		return
 	}

internal/api/handlers/management/logs.go

@@ -58,8 +58,14 @@ func (h *Handler) GetLogs(c *gin.Context) {
 		return
 	}
 
+	limit, errLimit := parseLimit(c.Query("limit"))
+	if errLimit != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("invalid limit: %v", errLimit)})
+		return
+	}
+
 	cutoff := parseCutoff(c.Query("after"))
-	acc := newLogAccumulator(cutoff)
+	acc := newLogAccumulator(cutoff, limit)
 	for i := range files {
 		if errProcess := acc.consumeFile(files[i]); errProcess != nil {
 			c.JSON(http.StatusInternalServerError, gin.H{"error": fmt.Sprintf("failed to read log file %s: %v", files[i], errProcess)})
@@ -139,6 +145,126 @@ func (h *Handler) DeleteLogs(c *gin.Context) {
 	})
 }
 
+// GetRequestErrorLogs lists error request log files when RequestLog is disabled.
+// It returns an empty list when RequestLog is enabled.
+func (h *Handler) GetRequestErrorLogs(c *gin.Context) {
+	if h == nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "handler unavailable"})
+		return
+	}
+	if h.cfg == nil {
+		c.JSON(http.StatusServiceUnavailable, gin.H{"error": "configuration unavailable"})
+		return
+	}
+	if h.cfg.RequestLog {
+		c.JSON(http.StatusOK, gin.H{"files": []any{}})
+		return
+	}
+
+	dir := h.logDirectory()
+	if strings.TrimSpace(dir) == "" {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "log directory not configured"})
+		return
+	}
+
+	entries, err := os.ReadDir(dir)
+	if err != nil {
+		if os.IsNotExist(err) {
+			c.JSON(http.StatusOK, gin.H{"files": []any{}})
+			return
+		}
+		c.JSON(http.StatusInternalServerError, gin.H{"error": fmt.Sprintf("failed to list request error logs: %v", err)})
+		return
+	}
+
+	type errorLog struct {
+		Name     string `json:"name"`
+		Size     int64  `json:"size"`
+		Modified int64  `json:"modified"`
+	}
+
+	files := make([]errorLog, 0, len(entries))
+	for _, entry := range entries {
+		if entry.IsDir() {
+			continue
+		}
+		name := entry.Name()
+		if !strings.HasPrefix(name, "error-") || !strings.HasSuffix(name, ".log") {
+			continue
+		}
+		info, errInfo := entry.Info()
+		if errInfo != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": fmt.Sprintf("failed to read log info for %s: %v", name, errInfo)})
+			return
+		}
+		files = append(files, errorLog{
+			Name:     name,
+			Size:     info.Size(),
+			Modified: info.ModTime().Unix(),
+		})
+	}
+
+	sort.Slice(files, func(i, j int) bool { return files[i].Modified > files[j].Modified })
+
+	c.JSON(http.StatusOK, gin.H{"files": files})
+}
+
+// DownloadRequestErrorLog downloads a specific error request log file by name.
+func (h *Handler) DownloadRequestErrorLog(c *gin.Context) {
+	if h == nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "handler unavailable"})
+		return
+	}
+	if h.cfg == nil {
+		c.JSON(http.StatusServiceUnavailable, gin.H{"error": "configuration unavailable"})
+		return
+	}
+
+	dir := h.logDirectory()
+	if strings.TrimSpace(dir) == "" {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "log directory not configured"})
+		return
+	}
+
+	name := strings.TrimSpace(c.Param("name"))
+	if name == "" || strings.Contains(name, "/") || strings.Contains(name, "\\") {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid log file name"})
+		return
+	}
+	if !strings.HasPrefix(name, "error-") || !strings.HasSuffix(name, ".log") {
+		c.JSON(http.StatusNotFound, gin.H{"error": "log file not found"})
+		return
+	}
+
+	dirAbs, errAbs := filepath.Abs(dir)
+	if errAbs != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": fmt.Sprintf("failed to resolve log directory: %v", errAbs)})
+		return
+	}
+	fullPath := filepath.Clean(filepath.Join(dirAbs, name))
+	prefix := dirAbs + string(os.PathSeparator)
+	if !strings.HasPrefix(fullPath, prefix) {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid log file path"})
+		return
+	}
+
+	info, errStat := os.Stat(fullPath)
+	if errStat != nil {
+		if os.IsNotExist(errStat) {
+			c.JSON(http.StatusNotFound, gin.H{"error": "log file not found"})
+			return
+		}
+		c.JSON(http.StatusInternalServerError, gin.H{"error": fmt.Sprintf("failed to read log file: %v", errStat)})
+		return
+	}
+	if info.IsDir() {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid log file"})
+		return
+	}
+
+	c.FileAttachment(fullPath, name)
+}
+
 func (h *Handler) logDirectory() string {
 	if h == nil {
 		return ""
@@ -194,16 +320,22 @@ func (h *Handler) collectLogFiles(dir string) ([]string, error) {
 
 type logAccumulator struct {
 	cutoff  int64
+	limit   int
 	lines   []string
 	total   int
 	latest  int64
 	include bool
 }
 
-func newLogAccumulator(cutoff int64) *logAccumulator {
+func newLogAccumulator(cutoff int64, limit int) *logAccumulator {
+	capacity := 256
+	if limit > 0 && limit < capacity {
+		capacity = limit
+	}
 	return &logAccumulator{
 		cutoff: cutoff,
-		lines:  make([]string, 0, 256),
+		limit:  limit,
+		lines:  make([]string, 0, capacity),
 	}
 }
 
@@ -215,7 +347,9 @@ func (acc *logAccumulator) consumeFile(path string) error {
 		}
 		return err
 	}
-	defer file.Close()
+	defer func() {
+		_ = file.Close()
+	}()
 
 	scanner := bufio.NewScanner(file)
 	buf := make([]byte, 0, logScannerInitialBuffer)
@@ -239,12 +373,19 @@ func (acc *logAccumulator) addLine(raw string) {
 	if ts > 0 {
 		acc.include = acc.cutoff == 0 || ts > acc.cutoff
 		if acc.cutoff == 0 || acc.include {
-			acc.lines = append(acc.lines, line)
+			acc.append(line)
 		}
 		return
 	}
 	if acc.cutoff == 0 || acc.include {
-		acc.lines = append(acc.lines, line)
+		acc.append(line)
+	}
+}
+
+func (acc *logAccumulator) append(line string) {
+	acc.lines = append(acc.lines, line)
+	if acc.limit > 0 && len(acc.lines) > acc.limit {
+		acc.lines = acc.lines[len(acc.lines)-acc.limit:]
 	}
 }
 
@@ -267,6 +408,21 @@ func parseCutoff(raw string) int64 {
 	return ts
 }
 
+func parseLimit(raw string) (int, error) {
+	value := strings.TrimSpace(raw)
+	if value == "" {
+		return 0, nil
+	}
+	limit, err := strconv.Atoi(value)
+	if err != nil {
+		return 0, fmt.Errorf("must be a positive integer")
+	}
+	if limit <= 0 {
+		return 0, fmt.Errorf("must be greater than zero")
+	}
+	return limit, nil
+}
+
 func parseTimestamp(line string) int64 {
 	if strings.HasPrefix(line, "[") {
 		line = line[1:]

internal/api/middleware/request_logging.go

@@ -6,6 +6,7 @@ package middleware
 import (
 	"bytes"
 	"io"
+	"net/http"
 	"strings"
 
 	"github.com/gin-gonic/gin"
@@ -15,23 +16,22 @@ import (
 
 // RequestLoggingMiddleware creates a Gin middleware that logs HTTP requests and responses.
 // It captures detailed information about the request and response, including headers and body,
-// and uses the provided RequestLogger to record this data. If logging is disabled in the
-// logger, the middleware has minimal overhead.
+// and uses the provided RequestLogger to record this data. When logging is disabled in the
+// logger, it still captures data so that upstream errors can be persisted.
 func RequestLoggingMiddleware(logger logging.RequestLogger) gin.HandlerFunc {
 	return func(c *gin.Context) {
-		path := c.Request.URL.Path
-		shouldLog := false
-		if strings.HasPrefix(path, "/v1") {
-			shouldLog = true
-		}
-
-		if !shouldLog {
+		if logger == nil {
 			c.Next()
 			return
 		}
 
-		// Early return if logging is disabled (zero overhead)
-		if !logger.IsEnabled() {
+		if c.Request.Method == http.MethodGet {
+			c.Next()
+			return
+		}
+
+		path := c.Request.URL.Path
+		if !shouldLogRequest(path) {
 			c.Next()
 			return
 		}
@@ -47,6 +47,9 @@ func RequestLoggingMiddleware(logger logging.RequestLogger) gin.HandlerFunc {
 
 		// Create response writer wrapper
 		wrapper := NewResponseWriterWrapper(c.Writer, logger, requestInfo)
+		if !logger.IsEnabled() {
+			wrapper.logOnErrorOnly = true
+		}
 		c.Writer = wrapper
 
 		// Process the request
@@ -101,3 +104,18 @@ func captureRequestInfo(c *gin.Context) (*RequestInfo, error) {
 		Body:    body,
 	}, nil
 }
+
+// shouldLogRequest determines whether the request should be logged.
+// It skips management endpoints to avoid leaking secrets but allows
+// all other routes, including module-provided ones, to honor request-log.
+func shouldLogRequest(path string) bool {
+	if strings.HasPrefix(path, "/v0/management") || strings.HasPrefix(path, "/management") {
+		return false
+	}
+
+	if strings.HasPrefix(path, "/api") {
+		return strings.HasPrefix(path, "/api/provider")
+	}
+
+	return true
+}

internal/api/middleware/response_writer.go

@@ -5,6 +5,7 @@ package middleware
 
 import (
 	"bytes"
+	"net/http"
 	"strings"
 
 	"github.com/gin-gonic/gin"
@@ -24,15 +25,16 @@ type RequestInfo struct {
 // It is designed to handle both standard and streaming responses, ensuring that logging operations do not block the client response.
 type ResponseWriterWrapper struct {
 	gin.ResponseWriter
-	body         *bytes.Buffer              // body is a buffer to store the response body for non-streaming responses.
-	isStreaming  bool                       // isStreaming indicates whether the response is a streaming type (e.g., text/event-stream).
-	streamWriter logging.StreamingLogWriter // streamWriter is a writer for handling streaming log entries.
-	chunkChannel chan []byte                // chunkChannel is a channel for asynchronously passing response chunks to the logger.
-	streamDone   chan struct{}              // streamDone signals when the streaming goroutine completes.
-	logger       logging.RequestLogger      // logger is the instance of the request logger service.
-	requestInfo  *RequestInfo               // requestInfo holds the details of the original request.
-	statusCode   int                        // statusCode stores the HTTP status code of the response.
-	headers      map[string][]string        // headers stores the response headers.
+	body           *bytes.Buffer              // body is a buffer to store the response body for non-streaming responses.
+	isStreaming    bool                       // isStreaming indicates whether the response is a streaming type (e.g., text/event-stream).
+	streamWriter   logging.StreamingLogWriter // streamWriter is a writer for handling streaming log entries.
+	chunkChannel   chan []byte                // chunkChannel is a channel for asynchronously passing response chunks to the logger.
+	streamDone     chan struct{}              // streamDone signals when the streaming goroutine completes.
+	logger         logging.RequestLogger      // logger is the instance of the request logger service.
+	requestInfo    *RequestInfo               // requestInfo holds the details of the original request.
+	statusCode     int                        // statusCode stores the HTTP status code of the response.
+	headers        map[string][]string        // headers stores the response headers.
+	logOnErrorOnly bool                       // logOnErrorOnly enables logging only when an error response is detected.
 }
 
 // NewResponseWriterWrapper creates and initializes a new ResponseWriterWrapper.
@@ -192,12 +194,34 @@ func (w *ResponseWriterWrapper) processStreamingChunks(done chan struct{}) {
 // For non-streaming responses, it logs the complete request and response details,
 // including any API-specific request/response data stored in the Gin context.
 func (w *ResponseWriterWrapper) Finalize(c *gin.Context) error {
-	if !w.logger.IsEnabled() {
+	if w.logger == nil {
+		return nil
+	}
+
+	finalStatusCode := w.statusCode
+	if finalStatusCode == 0 {
+		if statusWriter, ok := w.ResponseWriter.(interface{ Status() int }); ok {
+			finalStatusCode = statusWriter.Status()
+		} else {
+			finalStatusCode = 200
+		}
+	}
+
+	var slicesAPIResponseError []*interfaces.ErrorMessage
+	apiResponseError, isExist := c.Get("API_RESPONSE_ERROR")
+	if isExist {
+		if apiErrors, ok := apiResponseError.([]*interfaces.ErrorMessage); ok {
+			slicesAPIResponseError = apiErrors
+		}
+	}
+
+	hasAPIError := len(slicesAPIResponseError) > 0 || finalStatusCode >= http.StatusBadRequest
+	forceLog := w.logOnErrorOnly && hasAPIError && !w.logger.IsEnabled()
+	if !w.logger.IsEnabled() && !forceLog {
 		return nil
 	}
 
 	if w.isStreaming {
-		// Close streaming channel and writer
 		if w.chunkChannel != nil {
 			close(w.chunkChannel)
 			w.chunkChannel = nil
@@ -208,81 +232,108 @@ func (w *ResponseWriterWrapper) Finalize(c *gin.Context) error {
 			w.streamDone = nil
 		}
 
+		// Write API Request and Response to the streaming log before closing
 		if w.streamWriter != nil {
-			err := w.streamWriter.Close()
+			apiRequest := w.extractAPIRequest(c)
+			if len(apiRequest) > 0 {
+				_ = w.streamWriter.WriteAPIRequest(apiRequest)
+			}
+			apiResponse := w.extractAPIResponse(c)
+			if len(apiResponse) > 0 {
+				_ = w.streamWriter.WriteAPIResponse(apiResponse)
+			}
+			if err := w.streamWriter.Close(); err != nil {
+				w.streamWriter = nil
+				return err
+			}
 			w.streamWriter = nil
-			return err
 		}
-	} else {
-		// Capture final status code and headers if not already captured
-		finalStatusCode := w.statusCode
-		if finalStatusCode == 0 {
-			// Get status from underlying ResponseWriter if available
-			if statusWriter, ok := w.ResponseWriter.(interface{ Status() int }); ok {
-				finalStatusCode = statusWriter.Status()
-			} else {
-				finalStatusCode = 200 // Default
-			}
+		if forceLog {
+			return w.logRequest(finalStatusCode, w.cloneHeaders(), w.body.Bytes(), w.extractAPIRequest(c), w.extractAPIResponse(c), slicesAPIResponseError, forceLog)
 		}
+		return nil
+	}
 
-		// Ensure we have the latest headers before finalizing
-		w.ensureHeadersCaptured()
+	return w.logRequest(finalStatusCode, w.cloneHeaders(), w.body.Bytes(), w.extractAPIRequest(c), w.extractAPIResponse(c), slicesAPIResponseError, forceLog)
+}
 
-		// Use the captured headers as the final headers
-		finalHeaders := make(map[string][]string)
-		for key, values := range w.headers {
-			// Make a copy of the values slice to avoid reference issues
-			headerValues := make([]string, len(values))
-			copy(headerValues, values)
-			finalHeaders[key] = headerValues
-		}
+func (w *ResponseWriterWrapper) cloneHeaders() map[string][]string {
+	w.ensureHeadersCaptured()
 
-		var apiRequestBody []byte
-		apiRequest, isExist := c.Get("API_REQUEST")
-		if isExist {
-			var ok bool
-			apiRequestBody, ok = apiRequest.([]byte)
-			if !ok {
-				apiRequestBody = nil
-			}
-		}
+	finalHeaders := make(map[string][]string, len(w.headers))
+	for key, values := range w.headers {
+		headerValues := make([]string, len(values))
+		copy(headerValues, values)
+		finalHeaders[key] = headerValues
+	}
 
-		var apiResponseBody []byte
-		apiResponse, isExist := c.Get("API_RESPONSE")
-		if isExist {
-			var ok bool
-			apiResponseBody, ok = apiResponse.([]byte)
-			if !ok {
-				apiResponseBody = nil
-			}
-		}
+	return finalHeaders
+}
 
-		var slicesAPIResponseError []*interfaces.ErrorMessage
-		apiResponseError, isExist := c.Get("API_RESPONSE_ERROR")
-		if isExist {
-			var ok bool
-			slicesAPIResponseError, ok = apiResponseError.([]*interfaces.ErrorMessage)
-			if !ok {
-				slicesAPIResponseError = nil
-			}
-		}
+func (w *ResponseWriterWrapper) extractAPIRequest(c *gin.Context) []byte {
+	apiRequest, isExist := c.Get("API_REQUEST")
+	if !isExist {
+		return nil
+	}
+	data, ok := apiRequest.([]byte)
+	if !ok || len(data) == 0 {
+		return nil
+	}
+	return data
+}
 
-		// Log complete non-streaming response
-		return w.logger.LogRequest(
+func (w *ResponseWriterWrapper) extractAPIResponse(c *gin.Context) []byte {
+	apiResponse, isExist := c.Get("API_RESPONSE")
+	if !isExist {
+		return nil
+	}
+	data, ok := apiResponse.([]byte)
+	if !ok || len(data) == 0 {
+		return nil
+	}
+	return data
+}
+
+func (w *ResponseWriterWrapper) logRequest(statusCode int, headers map[string][]string, body []byte, apiRequestBody, apiResponseBody []byte, apiResponseErrors []*interfaces.ErrorMessage, forceLog bool) error {
+	if w.requestInfo == nil {
+		return nil
+	}
+
+	var requestBody []byte
+	if len(w.requestInfo.Body) > 0 {
+		requestBody = w.requestInfo.Body
+	}
+
+	if loggerWithOptions, ok := w.logger.(interface {
+		LogRequestWithOptions(string, string, map[string][]string, []byte, int, map[string][]string, []byte, []byte, []byte, []*interfaces.ErrorMessage, bool) error
+	}); ok {
+		return loggerWithOptions.LogRequestWithOptions(
 			w.requestInfo.URL,
 			w.requestInfo.Method,
 			w.requestInfo.Headers,
-			w.requestInfo.Body,
-			finalStatusCode,
-			finalHeaders,
-			w.body.Bytes(),
+			requestBody,
+			statusCode,
+			headers,
+			body,
 			apiRequestBody,
 			apiResponseBody,
-			slicesAPIResponseError,
+			apiResponseErrors,
+			forceLog,
 		)
 	}
 
-	return nil
+	return w.logger.LogRequest(
+		w.requestInfo.URL,
+		w.requestInfo.Method,
+		w.requestInfo.Headers,
+		requestBody,
+		statusCode,
+		headers,
+		body,
+		apiRequestBody,
+		apiResponseBody,
+		apiResponseErrors,
+	)
 }
 
 // Status returns the HTTP response status code captured by the wrapper.

internal/api/modules/amp/amp.go

@@ -0,0 +1,342 @@
+// Package amp implements the Amp CLI routing module, providing OAuth-based
+// integration with Amp CLI for ChatGPT and Anthropic subscriptions.
+package amp
+
+import (
+	"fmt"
+	"net/http/httputil"
+	"strings"
+	"sync"
+
+	"github.com/gin-gonic/gin"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/api/modules"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	sdkaccess "github.com/router-for-me/CLIProxyAPI/v6/sdk/access"
+	log "github.com/sirupsen/logrus"
+)
+
+// Option configures the AmpModule.
+type Option func(*AmpModule)
+
+// AmpModule implements the RouteModuleV2 interface for Amp CLI integration.
+// It provides:
+//   - Reverse proxy to Amp control plane for OAuth/management
+//   - Provider-specific route aliases (/api/provider/{provider}/...)
+//   - Automatic gzip decompression for misconfigured upstreams
+//   - Model mapping for routing unavailable models to alternatives
+type AmpModule struct {
+	secretSource    SecretSource
+	proxy           *httputil.ReverseProxy
+	proxyMu         sync.RWMutex // protects proxy for hot-reload
+	accessManager   *sdkaccess.Manager
+	authMiddleware_ gin.HandlerFunc
+	modelMapper     *DefaultModelMapper
+	enabled         bool
+	registerOnce    sync.Once
+
+	// restrictToLocalhost controls localhost-only access for management routes (hot-reloadable)
+	restrictToLocalhost bool
+	restrictMu          sync.RWMutex
+
+	// configMu protects lastConfig for partial reload comparison
+	configMu   sync.RWMutex
+	lastConfig *config.AmpCode
+}
+
+// New creates a new Amp routing module with the given options.
+// This is the preferred constructor using the Option pattern.
+//
+// Example:
+//
+//	ampModule := amp.New(
+//	    amp.WithAccessManager(accessManager),
+//	    amp.WithAuthMiddleware(authMiddleware),
+//	    amp.WithSecretSource(customSecret),
+//	)
+func New(opts ...Option) *AmpModule {
+	m := &AmpModule{
+		secretSource: nil, // Will be created on demand if not provided
+	}
+	for _, opt := range opts {
+		opt(m)
+	}
+	return m
+}
+
+// NewLegacy creates a new Amp routing module using the legacy constructor signature.
+// This is provided for backwards compatibility.
+//
+// DEPRECATED: Use New with options instead.
+func NewLegacy(accessManager *sdkaccess.Manager, authMiddleware gin.HandlerFunc) *AmpModule {
+	return New(
+		WithAccessManager(accessManager),
+		WithAuthMiddleware(authMiddleware),
+	)
+}
+
+// WithSecretSource sets a custom secret source for the module.
+func WithSecretSource(source SecretSource) Option {
+	return func(m *AmpModule) {
+		m.secretSource = source
+	}
+}
+
+// WithAccessManager sets the access manager for the module.
+func WithAccessManager(am *sdkaccess.Manager) Option {
+	return func(m *AmpModule) {
+		m.accessManager = am
+	}
+}
+
+// WithAuthMiddleware sets the authentication middleware for provider routes.
+func WithAuthMiddleware(middleware gin.HandlerFunc) Option {
+	return func(m *AmpModule) {
+		m.authMiddleware_ = middleware
+	}
+}
+
+// Name returns the module identifier
+func (m *AmpModule) Name() string {
+	return "amp-routing"
+}
+
+// forceModelMappings returns whether model mappings should take precedence over local API keys
+func (m *AmpModule) forceModelMappings() bool {
+	m.configMu.RLock()
+	defer m.configMu.RUnlock()
+	if m.lastConfig == nil {
+		return false
+	}
+	return m.lastConfig.ForceModelMappings
+}
+
+// Register sets up Amp routes if configured.
+// This implements the RouteModuleV2 interface with Context.
+// Routes are registered only once via sync.Once for idempotent behavior.
+func (m *AmpModule) Register(ctx modules.Context) error {
+	settings := ctx.Config.AmpCode
+	upstreamURL := strings.TrimSpace(settings.UpstreamURL)
+
+	// Determine auth middleware (from module or context)
+	auth := m.getAuthMiddleware(ctx)
+
+	// Use registerOnce to ensure routes are only registered once
+	var regErr error
+	m.registerOnce.Do(func() {
+		// Initialize model mapper from config (for routing unavailable models to alternatives)
+		m.modelMapper = NewModelMapper(settings.ModelMappings)
+
+		// Store initial config for partial reload comparison
+		settingsCopy := settings
+		m.lastConfig = &settingsCopy
+
+		// Initialize localhost restriction setting (hot-reloadable)
+		m.setRestrictToLocalhost(settings.RestrictManagementToLocalhost)
+
+		// Always register provider aliases - these work without an upstream
+		m.registerProviderAliases(ctx.Engine, ctx.BaseHandler, auth)
+
+		// Register management proxy routes once; middleware will gate access when upstream is unavailable.
+		m.registerManagementRoutes(ctx.Engine, ctx.BaseHandler)
+
+		// If no upstream URL, skip proxy routes but provider aliases are still available
+		if upstreamURL == "" {
+			log.Debug("amp upstream proxy disabled (no upstream URL configured)")
+			log.Debug("amp provider alias routes registered")
+			m.enabled = false
+			return
+		}
+
+		if err := m.enableUpstreamProxy(upstreamURL, &settings); err != nil {
+			regErr = fmt.Errorf("failed to create amp proxy: %w", err)
+			return
+		}
+
+		log.Debug("amp provider alias routes registered")
+	})
+
+	return regErr
+}
+
+// getAuthMiddleware returns the authentication middleware, preferring the
+// module's configured middleware, then the context middleware, then a fallback.
+func (m *AmpModule) getAuthMiddleware(ctx modules.Context) gin.HandlerFunc {
+	if m.authMiddleware_ != nil {
+		return m.authMiddleware_
+	}
+	if ctx.AuthMiddleware != nil {
+		return ctx.AuthMiddleware
+	}
+	// Fallback: no authentication (should not happen in production)
+	log.Warn("amp module: no auth middleware provided, allowing all requests")
+	return func(c *gin.Context) {
+		c.Next()
+	}
+}
+
+// OnConfigUpdated handles configuration updates with partial reload support.
+// Only updates components that have actually changed to avoid unnecessary work.
+// Supports hot-reload for: model-mappings, upstream-api-key, upstream-url, restrict-management-to-localhost.
+func (m *AmpModule) OnConfigUpdated(cfg *config.Config) error {
+	newSettings := cfg.AmpCode
+
+	// Get previous config for comparison
+	m.configMu.RLock()
+	oldSettings := m.lastConfig
+	m.configMu.RUnlock()
+
+	if oldSettings != nil && oldSettings.RestrictManagementToLocalhost != newSettings.RestrictManagementToLocalhost {
+		m.setRestrictToLocalhost(newSettings.RestrictManagementToLocalhost)
+		if !newSettings.RestrictManagementToLocalhost {
+			log.Warnf("amp management routes now accessible from any IP - this is insecure!")
+		}
+	}
+
+	newUpstreamURL := strings.TrimSpace(newSettings.UpstreamURL)
+	oldUpstreamURL := ""
+	if oldSettings != nil {
+		oldUpstreamURL = strings.TrimSpace(oldSettings.UpstreamURL)
+	}
+
+	if !m.enabled && newUpstreamURL != "" {
+		if err := m.enableUpstreamProxy(newUpstreamURL, &newSettings); err != nil {
+			log.Errorf("amp config: failed to enable upstream proxy for %s: %v", newUpstreamURL, err)
+		}
+	}
+
+	// Check model mappings change
+	modelMappingsChanged := m.hasModelMappingsChanged(oldSettings, &newSettings)
+	if modelMappingsChanged {
+		if m.modelMapper != nil {
+			m.modelMapper.UpdateMappings(newSettings.ModelMappings)
+		} else if m.enabled {
+			log.Warnf("amp model mapper not initialized, skipping model mapping update")
+		}
+	}
+
+	if m.enabled {
+		// Check upstream URL change - now supports hot-reload
+		if newUpstreamURL == "" && oldUpstreamURL != "" {
+			m.setProxy(nil)
+			m.enabled = false
+		} else if oldUpstreamURL != "" && newUpstreamURL != oldUpstreamURL && newUpstreamURL != "" {
+			// Recreate proxy with new URL
+			proxy, err := createReverseProxy(newUpstreamURL, m.secretSource)
+			if err != nil {
+				log.Errorf("amp config: failed to create proxy for new upstream URL %s: %v", newUpstreamURL, err)
+			} else {
+				m.setProxy(proxy)
+			}
+		}
+
+		// Check API key change
+		apiKeyChanged := m.hasAPIKeyChanged(oldSettings, &newSettings)
+		if apiKeyChanged {
+			if m.secretSource != nil {
+				if ms, ok := m.secretSource.(*MultiSourceSecret); ok {
+					ms.UpdateExplicitKey(newSettings.UpstreamAPIKey)
+					ms.InvalidateCache()
+				}
+			}
+		}
+
+	}
+
+	// Store current config for next comparison
+	m.configMu.Lock()
+	settingsCopy := newSettings // copy struct
+	m.lastConfig = &settingsCopy
+	m.configMu.Unlock()
+
+	return nil
+}
+
+func (m *AmpModule) enableUpstreamProxy(upstreamURL string, settings *config.AmpCode) error {
+	if m.secretSource == nil {
+		m.secretSource = NewMultiSourceSecret(settings.UpstreamAPIKey, 0 /* default 5min */)
+	} else if ms, ok := m.secretSource.(*MultiSourceSecret); ok {
+		ms.UpdateExplicitKey(settings.UpstreamAPIKey)
+		ms.InvalidateCache()
+	}
+
+	proxy, err := createReverseProxy(upstreamURL, m.secretSource)
+	if err != nil {
+		return err
+	}
+
+	m.setProxy(proxy)
+	m.enabled = true
+
+	log.Infof("amp upstream proxy enabled for: %s", upstreamURL)
+	return nil
+}
+
+// hasModelMappingsChanged compares old and new model mappings.
+func (m *AmpModule) hasModelMappingsChanged(old *config.AmpCode, new *config.AmpCode) bool {
+	if old == nil {
+		return len(new.ModelMappings) > 0
+	}
+
+	if len(old.ModelMappings) != len(new.ModelMappings) {
+		return true
+	}
+
+	// Build map for efficient comparison
+	oldMap := make(map[string]string, len(old.ModelMappings))
+	for _, mapping := range old.ModelMappings {
+		oldMap[strings.TrimSpace(mapping.From)] = strings.TrimSpace(mapping.To)
+	}
+
+	for _, mapping := range new.ModelMappings {
+		from := strings.TrimSpace(mapping.From)
+		to := strings.TrimSpace(mapping.To)
+		if oldTo, exists := oldMap[from]; !exists || oldTo != to {
+			return true
+		}
+	}
+
+	return false
+}
+
+// hasAPIKeyChanged compares old and new API keys.
+func (m *AmpModule) hasAPIKeyChanged(old *config.AmpCode, new *config.AmpCode) bool {
+	oldKey := ""
+	if old != nil {
+		oldKey = strings.TrimSpace(old.UpstreamAPIKey)
+	}
+	newKey := strings.TrimSpace(new.UpstreamAPIKey)
+	return oldKey != newKey
+}
+
+// GetModelMapper returns the model mapper instance (for testing/debugging).
+func (m *AmpModule) GetModelMapper() *DefaultModelMapper {
+	return m.modelMapper
+}
+
+// getProxy returns the current proxy instance (thread-safe for hot-reload).
+func (m *AmpModule) getProxy() *httputil.ReverseProxy {
+	m.proxyMu.RLock()
+	defer m.proxyMu.RUnlock()
+	return m.proxy
+}
+
+// setProxy updates the proxy instance (thread-safe for hot-reload).
+func (m *AmpModule) setProxy(proxy *httputil.ReverseProxy) {
+	m.proxyMu.Lock()
+	defer m.proxyMu.Unlock()
+	m.proxy = proxy
+}
+
+// IsRestrictedToLocalhost returns whether management routes are restricted to localhost.
+func (m *AmpModule) IsRestrictedToLocalhost() bool {
+	m.restrictMu.RLock()
+	defer m.restrictMu.RUnlock()
+	return m.restrictToLocalhost
+}
+
+// setRestrictToLocalhost updates the localhost restriction setting.
+func (m *AmpModule) setRestrictToLocalhost(restrict bool) {
+	m.restrictMu.Lock()
+	defer m.restrictMu.Unlock()
+	m.restrictToLocalhost = restrict
+}

internal/api/modules/amp/amp_test.go

@@ -0,0 +1,311 @@
+package amp
+
+import (
+	"context"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"github.com/gin-gonic/gin"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/api/modules"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	sdkaccess "github.com/router-for-me/CLIProxyAPI/v6/sdk/access"
+	"github.com/router-for-me/CLIProxyAPI/v6/sdk/api/handlers"
+)
+
+func TestAmpModule_Name(t *testing.T) {
+	m := New()
+	if m.Name() != "amp-routing" {
+		t.Fatalf("want amp-routing, got %s", m.Name())
+	}
+}
+
+func TestAmpModule_New(t *testing.T) {
+	accessManager := sdkaccess.NewManager()
+	authMiddleware := func(c *gin.Context) { c.Next() }
+
+	m := NewLegacy(accessManager, authMiddleware)
+
+	if m.accessManager != accessManager {
+		t.Fatal("accessManager not set")
+	}
+	if m.authMiddleware_ == nil {
+		t.Fatal("authMiddleware not set")
+	}
+	if m.enabled {
+		t.Fatal("enabled should be false initially")
+	}
+	if m.proxy != nil {
+		t.Fatal("proxy should be nil initially")
+	}
+}
+
+func TestAmpModule_Register_WithUpstream(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+
+	// Fake upstream to ensure URL is valid
+	upstream := httptest.NewServer(nil)
+	defer upstream.Close()
+
+	accessManager := sdkaccess.NewManager()
+	base := &handlers.BaseAPIHandler{}
+
+	m := NewLegacy(accessManager, func(c *gin.Context) { c.Next() })
+
+	cfg := &config.Config{
+		AmpCode: config.AmpCode{
+			UpstreamURL:    upstream.URL,
+			UpstreamAPIKey: "test-key",
+		},
+	}
+
+	ctx := modules.Context{Engine: r, BaseHandler: base, Config: cfg, AuthMiddleware: func(c *gin.Context) { c.Next() }}
+	if err := m.Register(ctx); err != nil {
+		t.Fatalf("register error: %v", err)
+	}
+
+	if !m.enabled {
+		t.Fatal("module should be enabled with upstream URL")
+	}
+	if m.proxy == nil {
+		t.Fatal("proxy should be initialized")
+	}
+	if m.secretSource == nil {
+		t.Fatal("secretSource should be initialized")
+	}
+}
+
+func TestAmpModule_Register_WithoutUpstream(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+
+	accessManager := sdkaccess.NewManager()
+	base := &handlers.BaseAPIHandler{}
+
+	m := NewLegacy(accessManager, func(c *gin.Context) { c.Next() })
+
+	cfg := &config.Config{
+		AmpCode: config.AmpCode{
+			UpstreamURL: "", // No upstream
+		},
+	}
+
+	ctx := modules.Context{Engine: r, BaseHandler: base, Config: cfg, AuthMiddleware: func(c *gin.Context) { c.Next() }}
+	if err := m.Register(ctx); err != nil {
+		t.Fatalf("register should not error without upstream: %v", err)
+	}
+
+	if m.enabled {
+		t.Fatal("module should be disabled without upstream URL")
+	}
+	if m.proxy != nil {
+		t.Fatal("proxy should not be initialized without upstream")
+	}
+
+	// But provider aliases should still be registered
+	req := httptest.NewRequest("GET", "/api/provider/openai/models", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code == 404 {
+		t.Fatal("provider aliases should be registered even without upstream")
+	}
+}
+
+func TestAmpModule_Register_InvalidUpstream(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+
+	accessManager := sdkaccess.NewManager()
+	base := &handlers.BaseAPIHandler{}
+
+	m := NewLegacy(accessManager, func(c *gin.Context) { c.Next() })
+
+	cfg := &config.Config{
+		AmpCode: config.AmpCode{
+			UpstreamURL: "://invalid-url",
+		},
+	}
+
+	ctx := modules.Context{Engine: r, BaseHandler: base, Config: cfg, AuthMiddleware: func(c *gin.Context) { c.Next() }}
+	if err := m.Register(ctx); err == nil {
+		t.Fatal("expected error for invalid upstream URL")
+	}
+}
+
+func TestAmpModule_OnConfigUpdated_CacheInvalidation(t *testing.T) {
+	tmpDir := t.TempDir()
+	p := filepath.Join(tmpDir, "secrets.json")
+	if err := os.WriteFile(p, []byte(`{"apiKey@https://ampcode.com/":"v1"}`), 0600); err != nil {
+		t.Fatal(err)
+	}
+
+	m := &AmpModule{enabled: true}
+	ms := NewMultiSourceSecretWithPath("", p, time.Minute)
+	m.secretSource = ms
+
+	// Warm the cache
+	if _, err := ms.Get(context.Background()); err != nil {
+		t.Fatal(err)
+	}
+
+	if ms.cache == nil {
+		t.Fatal("expected cache to be set")
+	}
+
+	// Update config - should invalidate cache
+	if err := m.OnConfigUpdated(&config.Config{AmpCode: config.AmpCode{UpstreamURL: "http://x"}}); err != nil {
+		t.Fatal(err)
+	}
+
+	if ms.cache != nil {
+		t.Fatal("expected cache to be invalidated")
+	}
+}
+
+func TestAmpModule_OnConfigUpdated_NotEnabled(t *testing.T) {
+	m := &AmpModule{enabled: false}
+
+	// Should not error or panic when disabled
+	if err := m.OnConfigUpdated(&config.Config{}); err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+}
+
+func TestAmpModule_OnConfigUpdated_URLRemoved(t *testing.T) {
+	m := &AmpModule{enabled: true}
+	ms := NewMultiSourceSecret("", 0)
+	m.secretSource = ms
+
+	// Config update with empty URL - should log warning but not error
+	cfg := &config.Config{AmpCode: config.AmpCode{UpstreamURL: ""}}
+
+	if err := m.OnConfigUpdated(cfg); err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+}
+
+func TestAmpModule_OnConfigUpdated_NonMultiSourceSecret(t *testing.T) {
+	// Test that OnConfigUpdated doesn't panic with StaticSecretSource
+	m := &AmpModule{enabled: true}
+	m.secretSource = NewStaticSecretSource("static-key")
+
+	cfg := &config.Config{AmpCode: config.AmpCode{UpstreamURL: "http://example.com"}}
+
+	// Should not error or panic
+	if err := m.OnConfigUpdated(cfg); err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+}
+
+func TestAmpModule_AuthMiddleware_Fallback(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+
+	// Create module with no auth middleware
+	m := &AmpModule{authMiddleware_: nil}
+
+	// Get the fallback middleware via getAuthMiddleware
+	ctx := modules.Context{Engine: r, AuthMiddleware: nil}
+	middleware := m.getAuthMiddleware(ctx)
+
+	if middleware == nil {
+		t.Fatal("getAuthMiddleware should return a fallback, not nil")
+	}
+
+	// Test that it works
+	called := false
+	r.GET("/test", middleware, func(c *gin.Context) {
+		called = true
+		c.String(200, "ok")
+	})
+
+	req := httptest.NewRequest("GET", "/test", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if !called {
+		t.Fatal("fallback middleware should allow requests through")
+	}
+}
+
+func TestAmpModule_SecretSource_FromConfig(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+
+	upstream := httptest.NewServer(nil)
+	defer upstream.Close()
+
+	accessManager := sdkaccess.NewManager()
+	base := &handlers.BaseAPIHandler{}
+
+	m := NewLegacy(accessManager, func(c *gin.Context) { c.Next() })
+
+	// Config with explicit API key
+	cfg := &config.Config{
+		AmpCode: config.AmpCode{
+			UpstreamURL:    upstream.URL,
+			UpstreamAPIKey: "config-key",
+		},
+	}
+
+	ctx := modules.Context{Engine: r, BaseHandler: base, Config: cfg, AuthMiddleware: func(c *gin.Context) { c.Next() }}
+	if err := m.Register(ctx); err != nil {
+		t.Fatalf("register error: %v", err)
+	}
+
+	// Secret source should be MultiSourceSecret with config key
+	if m.secretSource == nil {
+		t.Fatal("secretSource should be set")
+	}
+
+	// Verify it returns the config key
+	key, err := m.secretSource.Get(context.Background())
+	if err != nil {
+		t.Fatalf("Get error: %v", err)
+	}
+	if key != "config-key" {
+		t.Fatalf("want config-key, got %s", key)
+	}
+}
+
+func TestAmpModule_ProviderAliasesAlwaysRegistered(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	scenarios := []struct {
+		name      string
+		configURL string
+	}{
+		{"with_upstream", "http://example.com"},
+		{"without_upstream", ""},
+	}
+
+	for _, scenario := range scenarios {
+		t.Run(scenario.name, func(t *testing.T) {
+			r := gin.New()
+			accessManager := sdkaccess.NewManager()
+			base := &handlers.BaseAPIHandler{}
+
+			m := NewLegacy(accessManager, func(c *gin.Context) { c.Next() })
+
+			cfg := &config.Config{AmpCode: config.AmpCode{UpstreamURL: scenario.configURL}}
+
+			ctx := modules.Context{Engine: r, BaseHandler: base, Config: cfg, AuthMiddleware: func(c *gin.Context) { c.Next() }}
+			if err := m.Register(ctx); err != nil && scenario.configURL != "" {
+				t.Fatalf("register error: %v", err)
+			}
+
+			// Provider aliases should always be available
+			req := httptest.NewRequest("GET", "/api/provider/openai/models", nil)
+			w := httptest.NewRecorder()
+			r.ServeHTTP(w, req)
+
+			if w.Code == 404 {
+				t.Fatal("provider aliases should be registered")
+			}
+		})
+	}
+}

internal/api/modules/amp/fallback_handlers.go

@@ -0,0 +1,305 @@
+package amp
+
+import (
+	"bytes"
+	"io"
+	"net/http/httputil"
+	"strings"
+	"time"
+
+	"github.com/gin-gonic/gin"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
+	log "github.com/sirupsen/logrus"
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+// AmpRouteType represents the type of routing decision made for an Amp request
+type AmpRouteType string
+
+const (
+	// RouteTypeLocalProvider indicates the request is handled by a local OAuth provider (free)
+	RouteTypeLocalProvider AmpRouteType = "LOCAL_PROVIDER"
+	// RouteTypeModelMapping indicates the request was remapped to another available model (free)
+	RouteTypeModelMapping AmpRouteType = "MODEL_MAPPING"
+	// RouteTypeAmpCredits indicates the request is forwarded to ampcode.com (uses Amp credits)
+	RouteTypeAmpCredits AmpRouteType = "AMP_CREDITS"
+	// RouteTypeNoProvider indicates no provider or fallback available
+	RouteTypeNoProvider AmpRouteType = "NO_PROVIDER"
+)
+
+// MappedModelContextKey is the Gin context key for passing mapped model names.
+const MappedModelContextKey = "mapped_model"
+
+// logAmpRouting logs the routing decision for an Amp request with structured fields
+func logAmpRouting(routeType AmpRouteType, requestedModel, resolvedModel, provider, path string) {
+	fields := log.Fields{
+		"component":       "amp-routing",
+		"route_type":      string(routeType),
+		"requested_model": requestedModel,
+		"path":            path,
+		"timestamp":       time.Now().Format(time.RFC3339),
+	}
+
+	if resolvedModel != "" && resolvedModel != requestedModel {
+		fields["resolved_model"] = resolvedModel
+	}
+	if provider != "" {
+		fields["provider"] = provider
+	}
+
+	switch routeType {
+	case RouteTypeLocalProvider:
+		fields["cost"] = "free"
+		fields["source"] = "local_oauth"
+		log.WithFields(fields).Debugf("amp using local provider for model: %s", requestedModel)
+
+	case RouteTypeModelMapping:
+		fields["cost"] = "free"
+		fields["source"] = "local_oauth"
+		fields["mapping"] = requestedModel + " -> " + resolvedModel
+		// model mapping already logged in mapper; avoid duplicate here
+
+	case RouteTypeAmpCredits:
+		fields["cost"] = "amp_credits"
+		fields["source"] = "ampcode.com"
+		fields["model_id"] = requestedModel // Explicit model_id for easy config reference
+		log.WithFields(fields).Warnf("forwarding to ampcode.com (uses amp credits) - model_id: %s | To use local proxy, add to config: amp-model-mappings: [{from: \"%s\", to: \"<your-local-model>\"}]", requestedModel, requestedModel)
+
+	case RouteTypeNoProvider:
+		fields["cost"] = "none"
+		fields["source"] = "error"
+		fields["model_id"] = requestedModel // Explicit model_id for easy config reference
+		log.WithFields(fields).Warnf("no provider available for model_id: %s", requestedModel)
+	}
+}
+
+// FallbackHandler wraps a standard handler with fallback logic to ampcode.com
+// when the model's provider is not available in CLIProxyAPI
+type FallbackHandler struct {
+	getProxy           func() *httputil.ReverseProxy
+	modelMapper        ModelMapper
+	forceModelMappings func() bool
+}
+
+// NewFallbackHandler creates a new fallback handler wrapper
+// The getProxy function allows lazy evaluation of the proxy (useful when proxy is created after routes)
+func NewFallbackHandler(getProxy func() *httputil.ReverseProxy) *FallbackHandler {
+	return &FallbackHandler{
+		getProxy:           getProxy,
+		forceModelMappings: func() bool { return false },
+	}
+}
+
+// NewFallbackHandlerWithMapper creates a new fallback handler with model mapping support
+func NewFallbackHandlerWithMapper(getProxy func() *httputil.ReverseProxy, mapper ModelMapper, forceModelMappings func() bool) *FallbackHandler {
+	if forceModelMappings == nil {
+		forceModelMappings = func() bool { return false }
+	}
+	return &FallbackHandler{
+		getProxy:           getProxy,
+		modelMapper:        mapper,
+		forceModelMappings: forceModelMappings,
+	}
+}
+
+// SetModelMapper sets the model mapper for this handler (allows late binding)
+func (fh *FallbackHandler) SetModelMapper(mapper ModelMapper) {
+	fh.modelMapper = mapper
+}
+
+// WrapHandler wraps a gin.HandlerFunc with fallback logic
+// If the model's provider is not configured in CLIProxyAPI, it forwards to ampcode.com
+func (fh *FallbackHandler) WrapHandler(handler gin.HandlerFunc) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		requestPath := c.Request.URL.Path
+
+		// Read the request body to extract the model name
+		bodyBytes, err := io.ReadAll(c.Request.Body)
+		if err != nil {
+			log.Errorf("amp fallback: failed to read request body: %v", err)
+			handler(c)
+			return
+		}
+
+		// Restore the body for the handler to read
+		c.Request.Body = io.NopCloser(bytes.NewReader(bodyBytes))
+
+		// Try to extract model from request body or URL path (for Gemini)
+		modelName := extractModelFromRequest(bodyBytes, c)
+		if modelName == "" {
+			// Can't determine model, proceed with normal handler
+			handler(c)
+			return
+		}
+
+		// Normalize model (handles dynamic thinking suffixes)
+		normalizedModel, _ := util.NormalizeThinkingModel(modelName)
+
+		// Track resolved model for logging (may change if mapping is applied)
+		resolvedModel := normalizedModel
+		usedMapping := false
+		var providers []string
+
+		// Check if model mappings should be forced ahead of local API keys
+		forceMappings := fh.forceModelMappings != nil && fh.forceModelMappings()
+
+		if forceMappings {
+			// FORCE MODE: Check model mappings FIRST (takes precedence over local API keys)
+			// This allows users to route Amp requests to their preferred OAuth providers
+			if fh.modelMapper != nil {
+				if mappedModel := fh.modelMapper.MapModel(normalizedModel); mappedModel != "" {
+					// Mapping found - check if we have a provider for the mapped model
+					mappedProviders := util.GetProviderName(mappedModel)
+					if len(mappedProviders) > 0 {
+						// Mapping found and provider available - rewrite the model in request body
+						bodyBytes = rewriteModelInRequest(bodyBytes, mappedModel)
+						c.Request.Body = io.NopCloser(bytes.NewReader(bodyBytes))
+						// Store mapped model in context for handlers that check it (like gemini bridge)
+						c.Set(MappedModelContextKey, mappedModel)
+						resolvedModel = mappedModel
+						usedMapping = true
+						providers = mappedProviders
+					}
+				}
+			}
+
+			// If no mapping applied, check for local providers
+			if !usedMapping {
+				providers = util.GetProviderName(normalizedModel)
+			}
+		} else {
+			// DEFAULT MODE: Check local providers first, then mappings as fallback
+			providers = util.GetProviderName(normalizedModel)
+
+			if len(providers) == 0 {
+				// No providers configured - check if we have a model mapping
+				if fh.modelMapper != nil {
+					if mappedModel := fh.modelMapper.MapModel(normalizedModel); mappedModel != "" {
+						// Mapping found - check if we have a provider for the mapped model
+						mappedProviders := util.GetProviderName(mappedModel)
+						if len(mappedProviders) > 0 {
+							// Mapping found and provider available - rewrite the model in request body
+							bodyBytes = rewriteModelInRequest(bodyBytes, mappedModel)
+							c.Request.Body = io.NopCloser(bytes.NewReader(bodyBytes))
+							// Store mapped model in context for handlers that check it (like gemini bridge)
+							c.Set(MappedModelContextKey, mappedModel)
+							resolvedModel = mappedModel
+							usedMapping = true
+							providers = mappedProviders
+						}
+					}
+				}
+			}
+		}
+
+		// If no providers available, fallback to ampcode.com
+		if len(providers) == 0 {
+			proxy := fh.getProxy()
+			if proxy != nil {
+				// Log: Forwarding to ampcode.com (uses Amp credits)
+				logAmpRouting(RouteTypeAmpCredits, modelName, "", "", requestPath)
+
+				// Restore body again for the proxy
+				c.Request.Body = io.NopCloser(bytes.NewReader(bodyBytes))
+
+				// Forward to ampcode.com
+				proxy.ServeHTTP(c.Writer, c.Request)
+				return
+			}
+
+			// No proxy available, let the normal handler return the error
+			logAmpRouting(RouteTypeNoProvider, modelName, "", "", requestPath)
+		}
+
+		// Log the routing decision
+		providerName := ""
+		if len(providers) > 0 {
+			providerName = providers[0]
+		}
+
+		if usedMapping {
+			// Log: Model was mapped to another model
+			log.Debugf("amp model mapping: request %s -> %s", normalizedModel, resolvedModel)
+			logAmpRouting(RouteTypeModelMapping, modelName, resolvedModel, providerName, requestPath)
+			rewriter := NewResponseRewriter(c.Writer, normalizedModel)
+			c.Writer = rewriter
+			// Filter Anthropic-Beta header only for local handling paths
+			filterAntropicBetaHeader(c)
+			c.Request.Body = io.NopCloser(bytes.NewReader(bodyBytes))
+			handler(c)
+			rewriter.Flush()
+			log.Debugf("amp model mapping: response %s -> %s", resolvedModel, normalizedModel)
+		} else if len(providers) > 0 {
+			// Log: Using local provider (free)
+			logAmpRouting(RouteTypeLocalProvider, modelName, resolvedModel, providerName, requestPath)
+			// Filter Anthropic-Beta header only for local handling paths
+			filterAntropicBetaHeader(c)
+			c.Request.Body = io.NopCloser(bytes.NewReader(bodyBytes))
+			handler(c)
+		} else {
+			// No provider, no mapping, no proxy: fall back to the wrapped handler so it can return an error response
+			c.Request.Body = io.NopCloser(bytes.NewReader(bodyBytes))
+			handler(c)
+		}
+	}
+}
+
+// filterAntropicBetaHeader filters Anthropic-Beta header to remove features requiring special subscription
+// This is needed when using local providers (bypassing the Amp proxy)
+func filterAntropicBetaHeader(c *gin.Context) {
+	if betaHeader := c.Request.Header.Get("Anthropic-Beta"); betaHeader != "" {
+		if filtered := filterBetaFeatures(betaHeader, "context-1m-2025-08-07"); filtered != "" {
+			c.Request.Header.Set("Anthropic-Beta", filtered)
+		} else {
+			c.Request.Header.Del("Anthropic-Beta")
+		}
+	}
+}
+
+// rewriteModelInRequest replaces the model name in a JSON request body
+func rewriteModelInRequest(body []byte, newModel string) []byte {
+	if !gjson.GetBytes(body, "model").Exists() {
+		return body
+	}
+	result, err := sjson.SetBytes(body, "model", newModel)
+	if err != nil {
+		log.Warnf("amp model mapping: failed to rewrite model in request body: %v", err)
+		return body
+	}
+	return result
+}
+
+// extractModelFromRequest attempts to extract the model name from various request formats
+func extractModelFromRequest(body []byte, c *gin.Context) string {
+	// First try to parse from JSON body (OpenAI, Claude, etc.)
+	// Check common model field names
+	if result := gjson.GetBytes(body, "model"); result.Exists() && result.Type == gjson.String {
+		return result.String()
+	}
+
+	// For Gemini requests, model is in the URL path
+	// Standard format: /models/{model}:generateContent -> :action parameter
+	if action := c.Param("action"); action != "" {
+		// Split by colon to get model name (e.g., "gemini-pro:generateContent" -> "gemini-pro")
+		parts := strings.Split(action, ":")
+		if len(parts) > 0 && parts[0] != "" {
+			return parts[0]
+		}
+	}
+
+	// AMP CLI format: /publishers/google/models/{model}:method -> *path parameter
+	// Example: /publishers/google/models/gemini-3-pro-preview:streamGenerateContent
+	if path := c.Param("path"); path != "" {
+		// Look for /models/{model}:method pattern
+		if idx := strings.Index(path, "/models/"); idx >= 0 {
+			modelPart := path[idx+8:] // Skip "/models/"
+			// Split by colon to get model name
+			if colonIdx := strings.Index(modelPart, ":"); colonIdx > 0 {
+				return modelPart[:colonIdx]
+			}
+		}
+	}
+
+	return ""
+}

internal/api/modules/amp/gemini_bridge.go

@@ -0,0 +1,59 @@
+package amp
+
+import (
+	"strings"
+
+	"github.com/gin-gonic/gin"
+)
+
+// createGeminiBridgeHandler creates a handler that bridges AMP CLI's non-standard Gemini paths
+// to our standard Gemini handler by rewriting the request context.
+//
+// AMP CLI format: /publishers/google/models/gemini-3-pro-preview:streamGenerateContent
+// Standard format: /models/gemini-3-pro-preview:streamGenerateContent
+//
+// This extracts the model+method from the AMP path and sets it as the :action parameter
+// so the standard Gemini handler can process it.
+//
+// The handler parameter should be a Gemini-compatible handler that expects the :action param.
+func createGeminiBridgeHandler(handler gin.HandlerFunc) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		// Get the full path from the catch-all parameter
+		path := c.Param("path")
+
+		// Extract model:method from AMP CLI path format
+		// Example: /publishers/google/models/gemini-3-pro-preview:streamGenerateContent
+		const modelsPrefix = "/models/"
+		if idx := strings.Index(path, modelsPrefix); idx >= 0 {
+			// Extract everything after modelsPrefix
+			actionPart := path[idx+len(modelsPrefix):]
+
+			// Check if model was mapped by FallbackHandler
+			if mappedModel, exists := c.Get(MappedModelContextKey); exists {
+				if strModel, ok := mappedModel.(string); ok && strModel != "" {
+					// Replace the model part in the action
+					// actionPart is like "model-name:method"
+					if colonIdx := strings.Index(actionPart, ":"); colonIdx > 0 {
+						method := actionPart[colonIdx:] // ":method"
+						actionPart = strModel + method
+					}
+				}
+			}
+
+			// Set this as the :action parameter that the Gemini handler expects
+			c.Params = append(c.Params, gin.Param{
+				Key:   "action",
+				Value: actionPart,
+			})
+
+			// Call the handler
+			handler(c)
+			return
+		}
+
+		// If we can't parse the path, return 400
+		c.JSON(400, gin.H{
+			"error": "Invalid Gemini API path format",
+		})
+	}
+}

internal/api/modules/amp/gemini_bridge_test.go

@@ -0,0 +1,93 @@
+package amp
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+)
+
+func TestCreateGeminiBridgeHandler_ActionParameterExtraction(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	tests := []struct {
+		name           string
+		path           string
+		mappedModel    string // empty string means no mapping
+		expectedAction string
+	}{
+		{
+			name:           "no_mapping_uses_url_model",
+			path:           "/publishers/google/models/gemini-pro:generateContent",
+			mappedModel:    "",
+			expectedAction: "gemini-pro:generateContent",
+		},
+		{
+			name:           "mapped_model_replaces_url_model",
+			path:           "/publishers/google/models/gemini-exp:generateContent",
+			mappedModel:    "gemini-2.0-flash",
+			expectedAction: "gemini-2.0-flash:generateContent",
+		},
+		{
+			name:           "mapping_preserves_method",
+			path:           "/publishers/google/models/gemini-2.5-preview:streamGenerateContent",
+			mappedModel:    "gemini-flash",
+			expectedAction: "gemini-flash:streamGenerateContent",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			var capturedAction string
+
+			mockGeminiHandler := func(c *gin.Context) {
+				capturedAction = c.Param("action")
+				c.JSON(http.StatusOK, gin.H{"captured": capturedAction})
+			}
+
+			// Use the actual createGeminiBridgeHandler function
+			bridgeHandler := createGeminiBridgeHandler(mockGeminiHandler)
+
+			r := gin.New()
+			if tt.mappedModel != "" {
+				r.Use(func(c *gin.Context) {
+					c.Set(MappedModelContextKey, tt.mappedModel)
+					c.Next()
+				})
+			}
+			r.POST("/api/provider/google/v1beta1/*path", bridgeHandler)
+
+			req := httptest.NewRequest(http.MethodPost, "/api/provider/google/v1beta1"+tt.path, nil)
+			w := httptest.NewRecorder()
+			r.ServeHTTP(w, req)
+
+			if w.Code != http.StatusOK {
+				t.Fatalf("Expected status 200, got %d", w.Code)
+			}
+			if capturedAction != tt.expectedAction {
+				t.Errorf("Expected action '%s', got '%s'", tt.expectedAction, capturedAction)
+			}
+		})
+	}
+}
+
+func TestCreateGeminiBridgeHandler_InvalidPath(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	mockHandler := func(c *gin.Context) {
+		c.JSON(http.StatusOK, gin.H{"ok": true})
+	}
+	bridgeHandler := createGeminiBridgeHandler(mockHandler)
+
+	r := gin.New()
+	r.POST("/api/provider/google/v1beta1/*path", bridgeHandler)
+
+	req := httptest.NewRequest(http.MethodPost, "/api/provider/google/v1beta1/invalid/path", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusBadRequest {
+		t.Errorf("Expected status 400 for invalid path, got %d", w.Code)
+	}
+}

internal/api/modules/amp/model_mapping.go

@@ -0,0 +1,112 @@
+// Package amp provides model mapping functionality for routing Amp CLI requests
+// to alternative models when the requested model is not available locally.
+package amp
+
+import (
+	"strings"
+	"sync"
+
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
+	log "github.com/sirupsen/logrus"
+)
+
+// ModelMapper provides model name mapping/aliasing for Amp CLI requests.
+// When an Amp request comes in for a model that isn't available locally,
+// this mapper can redirect it to an alternative model that IS available.
+type ModelMapper interface {
+	// MapModel returns the target model name if a mapping exists and the target
+	// model has available providers. Returns empty string if no mapping applies.
+	MapModel(requestedModel string) string
+
+	// UpdateMappings refreshes the mapping configuration (for hot-reload).
+	UpdateMappings(mappings []config.AmpModelMapping)
+}
+
+// DefaultModelMapper implements ModelMapper with thread-safe mapping storage.
+type DefaultModelMapper struct {
+	mu       sync.RWMutex
+	mappings map[string]string // from -> to (normalized lowercase keys)
+}
+
+// NewModelMapper creates a new model mapper with the given initial mappings.
+func NewModelMapper(mappings []config.AmpModelMapping) *DefaultModelMapper {
+	m := &DefaultModelMapper{
+		mappings: make(map[string]string),
+	}
+	m.UpdateMappings(mappings)
+	return m
+}
+
+// MapModel checks if a mapping exists for the requested model and if the
+// target model has available local providers. Returns the mapped model name
+// or empty string if no valid mapping exists.
+func (m *DefaultModelMapper) MapModel(requestedModel string) string {
+	if requestedModel == "" {
+		return ""
+	}
+
+	m.mu.RLock()
+	defer m.mu.RUnlock()
+
+	// Normalize the requested model for lookup
+	normalizedRequest := strings.ToLower(strings.TrimSpace(requestedModel))
+
+	// Check for direct mapping
+	targetModel, exists := m.mappings[normalizedRequest]
+	if !exists {
+		return ""
+	}
+
+	// Verify target model has available providers
+	providers := util.GetProviderName(targetModel)
+	if len(providers) == 0 {
+		log.Debugf("amp model mapping: target model %s has no available providers, skipping mapping", targetModel)
+		return ""
+	}
+
+	// Note: Detailed routing log is handled by logAmpRouting in fallback_handlers.go
+	return targetModel
+}
+
+// UpdateMappings refreshes the mapping configuration from config.
+// This is called during initialization and on config hot-reload.
+func (m *DefaultModelMapper) UpdateMappings(mappings []config.AmpModelMapping) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+
+	// Clear and rebuild mappings
+	m.mappings = make(map[string]string, len(mappings))
+
+	for _, mapping := range mappings {
+		from := strings.TrimSpace(mapping.From)
+		to := strings.TrimSpace(mapping.To)
+
+		if from == "" || to == "" {
+			log.Warnf("amp model mapping: skipping invalid mapping (from=%q, to=%q)", from, to)
+			continue
+		}
+
+		// Store with normalized lowercase key for case-insensitive lookup
+		normalizedFrom := strings.ToLower(from)
+		m.mappings[normalizedFrom] = to
+
+		log.Debugf("amp model mapping registered: %s -> %s", from, to)
+	}
+
+	if len(m.mappings) > 0 {
+		log.Infof("amp model mapping: loaded %d mapping(s)", len(m.mappings))
+	}
+}
+
+// GetMappings returns a copy of current mappings (for debugging/status).
+func (m *DefaultModelMapper) GetMappings() map[string]string {
+	m.mu.RLock()
+	defer m.mu.RUnlock()
+
+	result := make(map[string]string, len(m.mappings))
+	for k, v := range m.mappings {
+		result[k] = v
+	}
+	return result
+}

internal/api/modules/amp/model_mapping_test.go

@@ -0,0 +1,186 @@
+package amp
+
+import (
+	"testing"
+
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
+)
+
+func TestNewModelMapper(t *testing.T) {
+	mappings := []config.AmpModelMapping{
+		{From: "claude-opus-4.5", To: "claude-sonnet-4"},
+		{From: "gpt-5", To: "gemini-2.5-pro"},
+	}
+
+	mapper := NewModelMapper(mappings)
+	if mapper == nil {
+		t.Fatal("Expected non-nil mapper")
+	}
+
+	result := mapper.GetMappings()
+	if len(result) != 2 {
+		t.Errorf("Expected 2 mappings, got %d", len(result))
+	}
+}
+
+func TestNewModelMapper_Empty(t *testing.T) {
+	mapper := NewModelMapper(nil)
+	if mapper == nil {
+		t.Fatal("Expected non-nil mapper")
+	}
+
+	result := mapper.GetMappings()
+	if len(result) != 0 {
+		t.Errorf("Expected 0 mappings, got %d", len(result))
+	}
+}
+
+func TestModelMapper_MapModel_NoProvider(t *testing.T) {
+	mappings := []config.AmpModelMapping{
+		{From: "claude-opus-4.5", To: "claude-sonnet-4"},
+	}
+
+	mapper := NewModelMapper(mappings)
+
+	// Without a registered provider for the target, mapping should return empty
+	result := mapper.MapModel("claude-opus-4.5")
+	if result != "" {
+		t.Errorf("Expected empty result when target has no provider, got %s", result)
+	}
+}
+
+func TestModelMapper_MapModel_WithProvider(t *testing.T) {
+	// Register a mock provider for the target model
+	reg := registry.GetGlobalRegistry()
+	reg.RegisterClient("test-client", "claude", []*registry.ModelInfo{
+		{ID: "claude-sonnet-4", OwnedBy: "anthropic", Type: "claude"},
+	})
+	defer reg.UnregisterClient("test-client")
+
+	mappings := []config.AmpModelMapping{
+		{From: "claude-opus-4.5", To: "claude-sonnet-4"},
+	}
+
+	mapper := NewModelMapper(mappings)
+
+	// With a registered provider, mapping should work
+	result := mapper.MapModel("claude-opus-4.5")
+	if result != "claude-sonnet-4" {
+		t.Errorf("Expected claude-sonnet-4, got %s", result)
+	}
+}
+
+func TestModelMapper_MapModel_CaseInsensitive(t *testing.T) {
+	reg := registry.GetGlobalRegistry()
+	reg.RegisterClient("test-client2", "claude", []*registry.ModelInfo{
+		{ID: "claude-sonnet-4", OwnedBy: "anthropic", Type: "claude"},
+	})
+	defer reg.UnregisterClient("test-client2")
+
+	mappings := []config.AmpModelMapping{
+		{From: "Claude-Opus-4.5", To: "claude-sonnet-4"},
+	}
+
+	mapper := NewModelMapper(mappings)
+
+	// Should match case-insensitively
+	result := mapper.MapModel("claude-opus-4.5")
+	if result != "claude-sonnet-4" {
+		t.Errorf("Expected claude-sonnet-4, got %s", result)
+	}
+}
+
+func TestModelMapper_MapModel_NotFound(t *testing.T) {
+	mappings := []config.AmpModelMapping{
+		{From: "claude-opus-4.5", To: "claude-sonnet-4"},
+	}
+
+	mapper := NewModelMapper(mappings)
+
+	// Unknown model should return empty
+	result := mapper.MapModel("unknown-model")
+	if result != "" {
+		t.Errorf("Expected empty for unknown model, got %s", result)
+	}
+}
+
+func TestModelMapper_MapModel_EmptyInput(t *testing.T) {
+	mappings := []config.AmpModelMapping{
+		{From: "claude-opus-4.5", To: "claude-sonnet-4"},
+	}
+
+	mapper := NewModelMapper(mappings)
+
+	result := mapper.MapModel("")
+	if result != "" {
+		t.Errorf("Expected empty for empty input, got %s", result)
+	}
+}
+
+func TestModelMapper_UpdateMappings(t *testing.T) {
+	mapper := NewModelMapper(nil)
+
+	// Initially empty
+	if len(mapper.GetMappings()) != 0 {
+		t.Error("Expected 0 initial mappings")
+	}
+
+	// Update with new mappings
+	mapper.UpdateMappings([]config.AmpModelMapping{
+		{From: "model-a", To: "model-b"},
+		{From: "model-c", To: "model-d"},
+	})
+
+	result := mapper.GetMappings()
+	if len(result) != 2 {
+		t.Errorf("Expected 2 mappings after update, got %d", len(result))
+	}
+
+	// Update again should replace, not append
+	mapper.UpdateMappings([]config.AmpModelMapping{
+		{From: "model-x", To: "model-y"},
+	})
+
+	result = mapper.GetMappings()
+	if len(result) != 1 {
+		t.Errorf("Expected 1 mapping after second update, got %d", len(result))
+	}
+}
+
+func TestModelMapper_UpdateMappings_SkipsInvalid(t *testing.T) {
+	mapper := NewModelMapper(nil)
+
+	mapper.UpdateMappings([]config.AmpModelMapping{
+		{From: "", To: "model-b"},        // Invalid: empty from
+		{From: "model-a", To: ""},        // Invalid: empty to
+		{From: "  ", To: "model-b"},      // Invalid: whitespace from
+		{From: "model-c", To: "model-d"}, // Valid
+	})
+
+	result := mapper.GetMappings()
+	if len(result) != 1 {
+		t.Errorf("Expected 1 valid mapping, got %d", len(result))
+	}
+}
+
+func TestModelMapper_GetMappings_ReturnsCopy(t *testing.T) {
+	mappings := []config.AmpModelMapping{
+		{From: "model-a", To: "model-b"},
+	}
+
+	mapper := NewModelMapper(mappings)
+
+	// Get mappings and modify the returned map
+	result := mapper.GetMappings()
+	result["new-key"] = "new-value"
+
+	// Original should be unchanged
+	original := mapper.GetMappings()
+	if len(original) != 1 {
+		t.Errorf("Expected original to have 1 mapping, got %d", len(original))
+	}
+	if _, exists := original["new-key"]; exists {
+		t.Error("Original map was modified")
+	}
+}

internal/api/modules/amp/proxy.go

@@ -0,0 +1,195 @@
+package amp
+
+import (
+	"bytes"
+	"compress/gzip"
+	"fmt"
+	"io"
+	"net/http"
+	"net/http/httputil"
+	"net/url"
+	"strconv"
+	"strings"
+
+	"github.com/gin-gonic/gin"
+	log "github.com/sirupsen/logrus"
+)
+
+// readCloser wraps a reader and forwards Close to a separate closer.
+// Used to restore peeked bytes while preserving upstream body Close behavior.
+type readCloser struct {
+	r io.Reader
+	c io.Closer
+}
+
+func (rc *readCloser) Read(p []byte) (int, error) { return rc.r.Read(p) }
+func (rc *readCloser) Close() error               { return rc.c.Close() }
+
+// createReverseProxy creates a reverse proxy handler for Amp upstream
+// with automatic gzip decompression via ModifyResponse
+func createReverseProxy(upstreamURL string, secretSource SecretSource) (*httputil.ReverseProxy, error) {
+	parsed, err := url.Parse(upstreamURL)
+	if err != nil {
+		return nil, fmt.Errorf("invalid amp upstream url: %w", err)
+	}
+
+	proxy := httputil.NewSingleHostReverseProxy(parsed)
+	originalDirector := proxy.Director
+
+	// Modify outgoing requests to inject API key and fix routing
+	proxy.Director = func(req *http.Request) {
+		originalDirector(req)
+		req.Host = parsed.Host
+
+		// Preserve correlation headers for debugging
+		if req.Header.Get("X-Request-ID") == "" {
+			// Could generate one here if needed
+		}
+
+		// Note: We do NOT filter Anthropic-Beta headers in the proxy path
+		// Users going through ampcode.com proxy are paying for the service and should get all features
+		// including 1M context window (context-1m-2025-08-07)
+
+		// Inject API key from secret source (precedence: config > env > file)
+		if key, err := secretSource.Get(req.Context()); err == nil && key != "" {
+			req.Header.Set("X-Api-Key", key)
+			req.Header.Set("Authorization", fmt.Sprintf("Bearer %s", key))
+		} else if err != nil {
+			log.Warnf("amp secret source error (continuing without auth): %v", err)
+		}
+	}
+
+	// Modify incoming responses to handle gzip without Content-Encoding
+	// This addresses the same issue as inline handler gzip handling, but at the proxy level
+	proxy.ModifyResponse = func(resp *http.Response) error {
+		// Only process successful responses
+		if resp.StatusCode < 200 || resp.StatusCode >= 300 {
+			return nil
+		}
+
+		// Skip if already marked as gzip (Content-Encoding set)
+		if resp.Header.Get("Content-Encoding") != "" {
+			return nil
+		}
+
+		// Skip streaming responses (SSE, chunked)
+		if isStreamingResponse(resp) {
+			return nil
+		}
+
+		// Save reference to original upstream body for proper cleanup
+		originalBody := resp.Body
+
+		// Peek at first 2 bytes to detect gzip magic bytes
+		header := make([]byte, 2)
+		n, _ := io.ReadFull(originalBody, header)
+
+		// Check for gzip magic bytes (0x1f 0x8b)
+		// If n < 2, we didn't get enough bytes, so it's not gzip
+		if n >= 2 && header[0] == 0x1f && header[1] == 0x8b {
+			// It's gzip - read the rest of the body
+			rest, err := io.ReadAll(originalBody)
+			if err != nil {
+				// Restore what we read and return original body (preserve Close behavior)
+				resp.Body = &readCloser{
+					r: io.MultiReader(bytes.NewReader(header[:n]), originalBody),
+					c: originalBody,
+				}
+				return nil
+			}
+
+			// Reconstruct complete gzipped data
+			gzippedData := append(header[:n], rest...)
+
+			// Decompress
+			gzipReader, err := gzip.NewReader(bytes.NewReader(gzippedData))
+			if err != nil {
+				log.Warnf("amp proxy: gzip header detected but decompress failed: %v", err)
+				// Close original body and return in-memory copy
+				_ = originalBody.Close()
+				resp.Body = io.NopCloser(bytes.NewReader(gzippedData))
+				return nil
+			}
+
+			decompressed, err := io.ReadAll(gzipReader)
+			_ = gzipReader.Close()
+			if err != nil {
+				log.Warnf("amp proxy: gzip decompress error: %v", err)
+				// Close original body and return in-memory copy
+				_ = originalBody.Close()
+				resp.Body = io.NopCloser(bytes.NewReader(gzippedData))
+				return nil
+			}
+
+			// Close original body since we're replacing with in-memory decompressed content
+			_ = originalBody.Close()
+
+			// Replace body with decompressed content
+			resp.Body = io.NopCloser(bytes.NewReader(decompressed))
+			resp.ContentLength = int64(len(decompressed))
+
+			// Update headers to reflect decompressed state
+			resp.Header.Del("Content-Encoding")                                          // No longer compressed
+			resp.Header.Del("Content-Length")                                            // Remove stale compressed length
+			resp.Header.Set("Content-Length", strconv.FormatInt(resp.ContentLength, 10)) // Set decompressed length
+
+			log.Debugf("amp proxy: decompressed gzip response (%d -> %d bytes)", len(gzippedData), len(decompressed))
+		} else {
+			// Not gzip - restore peeked bytes while preserving Close behavior
+			// Handle edge cases: n might be 0, 1, or 2 depending on EOF
+			resp.Body = &readCloser{
+				r: io.MultiReader(bytes.NewReader(header[:n]), originalBody),
+				c: originalBody,
+			}
+		}
+
+		return nil
+	}
+
+	// Error handler for proxy failures
+	proxy.ErrorHandler = func(rw http.ResponseWriter, req *http.Request, err error) {
+		log.Errorf("amp upstream proxy error for %s %s: %v", req.Method, req.URL.Path, err)
+		rw.Header().Set("Content-Type", "application/json")
+		rw.WriteHeader(http.StatusBadGateway)
+		_, _ = rw.Write([]byte(`{"error":"amp_upstream_proxy_error","message":"Failed to reach Amp upstream"}`))
+	}
+
+	return proxy, nil
+}
+
+// isStreamingResponse detects if the response is streaming (SSE only)
+// Note: We only treat text/event-stream as streaming. Chunked transfer encoding
+// is a transport-level detail and doesn't mean we can't decompress the full response.
+// Many JSON APIs use chunked encoding for normal responses.
+func isStreamingResponse(resp *http.Response) bool {
+	contentType := resp.Header.Get("Content-Type")
+
+	// Only Server-Sent Events are true streaming responses
+	if strings.Contains(contentType, "text/event-stream") {
+		return true
+	}
+
+	return false
+}
+
+// proxyHandler converts httputil.ReverseProxy to gin.HandlerFunc
+func proxyHandler(proxy *httputil.ReverseProxy) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		proxy.ServeHTTP(c.Writer, c.Request)
+	}
+}
+
+// filterBetaFeatures removes a specific beta feature from comma-separated list
+func filterBetaFeatures(header, featureToRemove string) string {
+	features := strings.Split(header, ",")
+	filtered := make([]string, 0, len(features))
+
+	for _, feature := range features {
+		trimmed := strings.TrimSpace(feature)
+		if trimmed != "" && trimmed != featureToRemove {
+			filtered = append(filtered, trimmed)
+		}
+	}
+
+	return strings.Join(filtered, ",")
+}

internal/api/modules/amp/proxy_test.go

@@ -0,0 +1,500 @@
+package amp
+
+import (
+	"bytes"
+	"compress/gzip"
+	"fmt"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+)
+
+// Helper: compress data with gzip
+func gzipBytes(b []byte) []byte {
+	var buf bytes.Buffer
+	zw := gzip.NewWriter(&buf)
+	zw.Write(b)
+	zw.Close()
+	return buf.Bytes()
+}
+
+// Helper: create a mock http.Response
+func mkResp(status int, hdr http.Header, body []byte) *http.Response {
+	if hdr == nil {
+		hdr = http.Header{}
+	}
+	return &http.Response{
+		StatusCode:    status,
+		Header:        hdr,
+		Body:          io.NopCloser(bytes.NewReader(body)),
+		ContentLength: int64(len(body)),
+	}
+}
+
+func TestCreateReverseProxy_ValidURL(t *testing.T) {
+	proxy, err := createReverseProxy("http://example.com", NewStaticSecretSource("key"))
+	if err != nil {
+		t.Fatalf("expected no error, got: %v", err)
+	}
+	if proxy == nil {
+		t.Fatal("expected proxy to be created")
+	}
+}
+
+func TestCreateReverseProxy_InvalidURL(t *testing.T) {
+	_, err := createReverseProxy("://invalid", NewStaticSecretSource("key"))
+	if err == nil {
+		t.Fatal("expected error for invalid URL")
+	}
+}
+
+func TestModifyResponse_GzipScenarios(t *testing.T) {
+	proxy, err := createReverseProxy("http://example.com", NewStaticSecretSource("k"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	goodJSON := []byte(`{"ok":true}`)
+	good := gzipBytes(goodJSON)
+	truncated := good[:10]
+	corrupted := append([]byte{0x1f, 0x8b}, []byte("notgzip")...)
+
+	cases := []struct {
+		name     string
+		header   http.Header
+		body     []byte
+		status   int
+		wantBody []byte
+		wantCE   string
+	}{
+		{
+			name:     "decompresses_valid_gzip_no_header",
+			header:   http.Header{},
+			body:     good,
+			status:   200,
+			wantBody: goodJSON,
+			wantCE:   "",
+		},
+		{
+			name:     "skips_when_ce_present",
+			header:   http.Header{"Content-Encoding": []string{"gzip"}},
+			body:     good,
+			status:   200,
+			wantBody: good,
+			wantCE:   "gzip",
+		},
+		{
+			name:     "passes_truncated_unchanged",
+			header:   http.Header{},
+			body:     truncated,
+			status:   200,
+			wantBody: truncated,
+			wantCE:   "",
+		},
+		{
+			name:     "passes_corrupted_unchanged",
+			header:   http.Header{},
+			body:     corrupted,
+			status:   200,
+			wantBody: corrupted,
+			wantCE:   "",
+		},
+		{
+			name:     "non_gzip_unchanged",
+			header:   http.Header{},
+			body:     []byte("plain"),
+			status:   200,
+			wantBody: []byte("plain"),
+			wantCE:   "",
+		},
+		{
+			name:     "empty_body",
+			header:   http.Header{},
+			body:     []byte{},
+			status:   200,
+			wantBody: []byte{},
+			wantCE:   "",
+		},
+		{
+			name:     "single_byte_body",
+			header:   http.Header{},
+			body:     []byte{0x1f},
+			status:   200,
+			wantBody: []byte{0x1f},
+			wantCE:   "",
+		},
+		{
+			name:     "skips_non_2xx_status",
+			header:   http.Header{},
+			body:     good,
+			status:   404,
+			wantBody: good,
+			wantCE:   "",
+		},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			resp := mkResp(tc.status, tc.header, tc.body)
+			if err := proxy.ModifyResponse(resp); err != nil {
+				t.Fatalf("ModifyResponse error: %v", err)
+			}
+			got, err := io.ReadAll(resp.Body)
+			if err != nil {
+				t.Fatalf("ReadAll error: %v", err)
+			}
+			if !bytes.Equal(got, tc.wantBody) {
+				t.Fatalf("body mismatch:\nwant: %q\ngot:  %q", tc.wantBody, got)
+			}
+			if ce := resp.Header.Get("Content-Encoding"); ce != tc.wantCE {
+				t.Fatalf("Content-Encoding: want %q, got %q", tc.wantCE, ce)
+			}
+		})
+	}
+}
+
+func TestModifyResponse_UpdatesContentLengthHeader(t *testing.T) {
+	proxy, err := createReverseProxy("http://example.com", NewStaticSecretSource("k"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	goodJSON := []byte(`{"message":"test response"}`)
+	gzipped := gzipBytes(goodJSON)
+
+	// Simulate upstream response with gzip body AND Content-Length header
+	// (this is the scenario the bot flagged - stale Content-Length after decompression)
+	resp := mkResp(200, http.Header{
+		"Content-Length": []string{fmt.Sprintf("%d", len(gzipped))}, // Compressed size
+	}, gzipped)
+
+	if err := proxy.ModifyResponse(resp); err != nil {
+		t.Fatalf("ModifyResponse error: %v", err)
+	}
+
+	// Verify body is decompressed
+	got, _ := io.ReadAll(resp.Body)
+	if !bytes.Equal(got, goodJSON) {
+		t.Fatalf("body should be decompressed, got: %q, want: %q", got, goodJSON)
+	}
+
+	// Verify Content-Length header is updated to decompressed size
+	wantCL := fmt.Sprintf("%d", len(goodJSON))
+	gotCL := resp.Header.Get("Content-Length")
+	if gotCL != wantCL {
+		t.Fatalf("Content-Length header mismatch: want %q (decompressed), got %q", wantCL, gotCL)
+	}
+
+	// Verify struct field also matches
+	if resp.ContentLength != int64(len(goodJSON)) {
+		t.Fatalf("resp.ContentLength mismatch: want %d, got %d", len(goodJSON), resp.ContentLength)
+	}
+}
+
+func TestModifyResponse_SkipsStreamingResponses(t *testing.T) {
+	proxy, err := createReverseProxy("http://example.com", NewStaticSecretSource("k"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	goodJSON := []byte(`{"ok":true}`)
+	gzipped := gzipBytes(goodJSON)
+
+	t.Run("sse_skips_decompression", func(t *testing.T) {
+		resp := mkResp(200, http.Header{"Content-Type": []string{"text/event-stream"}}, gzipped)
+		if err := proxy.ModifyResponse(resp); err != nil {
+			t.Fatalf("ModifyResponse error: %v", err)
+		}
+		// SSE should NOT be decompressed
+		got, _ := io.ReadAll(resp.Body)
+		if !bytes.Equal(got, gzipped) {
+			t.Fatal("SSE response should not be decompressed")
+		}
+	})
+}
+
+func TestModifyResponse_DecompressesChunkedJSON(t *testing.T) {
+	proxy, err := createReverseProxy("http://example.com", NewStaticSecretSource("k"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	goodJSON := []byte(`{"ok":true}`)
+	gzipped := gzipBytes(goodJSON)
+
+	t.Run("chunked_json_decompresses", func(t *testing.T) {
+		// Chunked JSON responses (like thread APIs) should be decompressed
+		resp := mkResp(200, http.Header{"Transfer-Encoding": []string{"chunked"}}, gzipped)
+		if err := proxy.ModifyResponse(resp); err != nil {
+			t.Fatalf("ModifyResponse error: %v", err)
+		}
+		// Should decompress because it's not SSE
+		got, _ := io.ReadAll(resp.Body)
+		if !bytes.Equal(got, goodJSON) {
+			t.Fatalf("chunked JSON should be decompressed, got: %q, want: %q", got, goodJSON)
+		}
+	})
+}
+
+func TestReverseProxy_InjectsHeaders(t *testing.T) {
+	gotHeaders := make(chan http.Header, 1)
+	upstream := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		gotHeaders <- r.Header.Clone()
+		w.WriteHeader(200)
+		w.Write([]byte(`ok`))
+	}))
+	defer upstream.Close()
+
+	proxy, err := createReverseProxy(upstream.URL, NewStaticSecretSource("secret"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		proxy.ServeHTTP(w, r)
+	}))
+	defer srv.Close()
+
+	res, err := http.Get(srv.URL + "/test")
+	if err != nil {
+		t.Fatal(err)
+	}
+	res.Body.Close()
+
+	hdr := <-gotHeaders
+	if hdr.Get("X-Api-Key") != "secret" {
+		t.Fatalf("X-Api-Key missing or wrong, got: %q", hdr.Get("X-Api-Key"))
+	}
+	if hdr.Get("Authorization") != "Bearer secret" {
+		t.Fatalf("Authorization missing or wrong, got: %q", hdr.Get("Authorization"))
+	}
+}
+
+func TestReverseProxy_EmptySecret(t *testing.T) {
+	gotHeaders := make(chan http.Header, 1)
+	upstream := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		gotHeaders <- r.Header.Clone()
+		w.WriteHeader(200)
+		w.Write([]byte(`ok`))
+	}))
+	defer upstream.Close()
+
+	proxy, err := createReverseProxy(upstream.URL, NewStaticSecretSource(""))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		proxy.ServeHTTP(w, r)
+	}))
+	defer srv.Close()
+
+	res, err := http.Get(srv.URL + "/test")
+	if err != nil {
+		t.Fatal(err)
+	}
+	res.Body.Close()
+
+	hdr := <-gotHeaders
+	// Should NOT inject headers when secret is empty
+	if hdr.Get("X-Api-Key") != "" {
+		t.Fatalf("X-Api-Key should not be set, got: %q", hdr.Get("X-Api-Key"))
+	}
+	if authVal := hdr.Get("Authorization"); authVal != "" && authVal != "Bearer " {
+		t.Fatalf("Authorization should not be set, got: %q", authVal)
+	}
+}
+
+func TestReverseProxy_ErrorHandler(t *testing.T) {
+	// Point proxy to a non-routable address to trigger error
+	proxy, err := createReverseProxy("http://127.0.0.1:1", NewStaticSecretSource(""))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		proxy.ServeHTTP(w, r)
+	}))
+	defer srv.Close()
+
+	res, err := http.Get(srv.URL + "/any")
+	if err != nil {
+		t.Fatal(err)
+	}
+	body, _ := io.ReadAll(res.Body)
+	res.Body.Close()
+
+	if res.StatusCode != http.StatusBadGateway {
+		t.Fatalf("want 502, got %d", res.StatusCode)
+	}
+	if !bytes.Contains(body, []byte(`"amp_upstream_proxy_error"`)) {
+		t.Fatalf("unexpected body: %s", body)
+	}
+	if ct := res.Header.Get("Content-Type"); ct != "application/json" {
+		t.Fatalf("content-type: want application/json, got %s", ct)
+	}
+}
+
+func TestReverseProxy_FullRoundTrip_Gzip(t *testing.T) {
+	// Upstream returns gzipped JSON without Content-Encoding header
+	upstream := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(200)
+		w.Write(gzipBytes([]byte(`{"upstream":"ok"}`)))
+	}))
+	defer upstream.Close()
+
+	proxy, err := createReverseProxy(upstream.URL, NewStaticSecretSource("key"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		proxy.ServeHTTP(w, r)
+	}))
+	defer srv.Close()
+
+	res, err := http.Get(srv.URL + "/test")
+	if err != nil {
+		t.Fatal(err)
+	}
+	body, _ := io.ReadAll(res.Body)
+	res.Body.Close()
+
+	expected := []byte(`{"upstream":"ok"}`)
+	if !bytes.Equal(body, expected) {
+		t.Fatalf("want decompressed JSON, got: %s", body)
+	}
+}
+
+func TestReverseProxy_FullRoundTrip_PlainJSON(t *testing.T) {
+	// Upstream returns plain JSON
+	upstream := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		w.WriteHeader(200)
+		w.Write([]byte(`{"plain":"json"}`))
+	}))
+	defer upstream.Close()
+
+	proxy, err := createReverseProxy(upstream.URL, NewStaticSecretSource("key"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		proxy.ServeHTTP(w, r)
+	}))
+	defer srv.Close()
+
+	res, err := http.Get(srv.URL + "/test")
+	if err != nil {
+		t.Fatal(err)
+	}
+	body, _ := io.ReadAll(res.Body)
+	res.Body.Close()
+
+	expected := []byte(`{"plain":"json"}`)
+	if !bytes.Equal(body, expected) {
+		t.Fatalf("want plain JSON unchanged, got: %s", body)
+	}
+}
+
+func TestIsStreamingResponse(t *testing.T) {
+	cases := []struct {
+		name   string
+		header http.Header
+		want   bool
+	}{
+		{
+			name:   "sse",
+			header: http.Header{"Content-Type": []string{"text/event-stream"}},
+			want:   true,
+		},
+		{
+			name:   "chunked_not_streaming",
+			header: http.Header{"Transfer-Encoding": []string{"chunked"}},
+			want:   false, // Chunked is transport-level, not streaming
+		},
+		{
+			name:   "normal_json",
+			header: http.Header{"Content-Type": []string{"application/json"}},
+			want:   false,
+		},
+		{
+			name:   "empty",
+			header: http.Header{},
+			want:   false,
+		},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			resp := &http.Response{Header: tc.header}
+			got := isStreamingResponse(resp)
+			if got != tc.want {
+				t.Fatalf("want %v, got %v", tc.want, got)
+			}
+		})
+	}
+}
+
+func TestFilterBetaFeatures(t *testing.T) {
+	tests := []struct {
+		name            string
+		header          string
+		featureToRemove string
+		expected        string
+	}{
+		{
+			name:            "Remove context-1m from middle",
+			header:          "fine-grained-tool-streaming-2025-05-14,context-1m-2025-08-07,oauth-2025-04-20",
+			featureToRemove: "context-1m-2025-08-07",
+			expected:        "fine-grained-tool-streaming-2025-05-14,oauth-2025-04-20",
+		},
+		{
+			name:            "Remove context-1m from start",
+			header:          "context-1m-2025-08-07,fine-grained-tool-streaming-2025-05-14",
+			featureToRemove: "context-1m-2025-08-07",
+			expected:        "fine-grained-tool-streaming-2025-05-14",
+		},
+		{
+			name:            "Remove context-1m from end",
+			header:          "fine-grained-tool-streaming-2025-05-14,context-1m-2025-08-07",
+			featureToRemove: "context-1m-2025-08-07",
+			expected:        "fine-grained-tool-streaming-2025-05-14",
+		},
+		{
+			name:            "Feature not present",
+			header:          "fine-grained-tool-streaming-2025-05-14,oauth-2025-04-20",
+			featureToRemove: "context-1m-2025-08-07",
+			expected:        "fine-grained-tool-streaming-2025-05-14,oauth-2025-04-20",
+		},
+		{
+			name:            "Only feature to remove",
+			header:          "context-1m-2025-08-07",
+			featureToRemove: "context-1m-2025-08-07",
+			expected:        "",
+		},
+		{
+			name:            "Empty header",
+			header:          "",
+			featureToRemove: "context-1m-2025-08-07",
+			expected:        "",
+		},
+		{
+			name:            "Header with spaces",
+			header:          "fine-grained-tool-streaming-2025-05-14, context-1m-2025-08-07 , oauth-2025-04-20",
+			featureToRemove: "context-1m-2025-08-07",
+			expected:        "fine-grained-tool-streaming-2025-05-14,oauth-2025-04-20",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := filterBetaFeatures(tt.header, tt.featureToRemove)
+			if result != tt.expected {
+				t.Errorf("filterBetaFeatures() = %q, want %q", result, tt.expected)
+			}
+		})
+	}
+}

internal/api/modules/amp/response_rewriter.go

@@ -0,0 +1,98 @@
+package amp
+
+import (
+	"bytes"
+	"net/http"
+	"strings"
+
+	"github.com/gin-gonic/gin"
+	log "github.com/sirupsen/logrus"
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+// ResponseRewriter wraps a gin.ResponseWriter to intercept and modify the response body
+// It's used to rewrite model names in responses when model mapping is used
+type ResponseRewriter struct {
+	gin.ResponseWriter
+	body          *bytes.Buffer
+	originalModel string
+	isStreaming   bool
+}
+
+// NewResponseRewriter creates a new response rewriter for model name substitution
+func NewResponseRewriter(w gin.ResponseWriter, originalModel string) *ResponseRewriter {
+	return &ResponseRewriter{
+		ResponseWriter: w,
+		body:           &bytes.Buffer{},
+		originalModel:  originalModel,
+	}
+}
+
+// Write intercepts response writes and buffers them for model name replacement
+func (rw *ResponseRewriter) Write(data []byte) (int, error) {
+	// Detect streaming on first write
+	if rw.body.Len() == 0 && !rw.isStreaming {
+		contentType := rw.Header().Get("Content-Type")
+		rw.isStreaming = strings.Contains(contentType, "text/event-stream") ||
+			strings.Contains(contentType, "stream")
+	}
+
+	if rw.isStreaming {
+		return rw.ResponseWriter.Write(rw.rewriteStreamChunk(data))
+	}
+	return rw.body.Write(data)
+}
+
+// Flush writes the buffered response with model names rewritten
+func (rw *ResponseRewriter) Flush() {
+	if rw.isStreaming {
+		if flusher, ok := rw.ResponseWriter.(http.Flusher); ok {
+			flusher.Flush()
+		}
+		return
+	}
+	if rw.body.Len() > 0 {
+		if _, err := rw.ResponseWriter.Write(rw.rewriteModelInResponse(rw.body.Bytes())); err != nil {
+			log.Warnf("amp response rewriter: failed to write rewritten response: %v", err)
+		}
+	}
+}
+
+// modelFieldPaths lists all JSON paths where model name may appear
+var modelFieldPaths = []string{"model", "modelVersion", "response.modelVersion", "message.model"}
+
+// rewriteModelInResponse replaces all occurrences of the mapped model with the original model in JSON
+func (rw *ResponseRewriter) rewriteModelInResponse(data []byte) []byte {
+	if rw.originalModel == "" {
+		return data
+	}
+	for _, path := range modelFieldPaths {
+		if gjson.GetBytes(data, path).Exists() {
+			data, _ = sjson.SetBytes(data, path, rw.originalModel)
+		}
+	}
+	return data
+}
+
+// rewriteStreamChunk rewrites model names in SSE stream chunks
+func (rw *ResponseRewriter) rewriteStreamChunk(chunk []byte) []byte {
+	if rw.originalModel == "" {
+		return chunk
+	}
+
+	// SSE format: "data: {json}\n\n"
+	lines := bytes.Split(chunk, []byte("\n"))
+	for i, line := range lines {
+		if bytes.HasPrefix(line, []byte("data: ")) {
+			jsonData := bytes.TrimPrefix(line, []byte("data: "))
+			if len(jsonData) > 0 && jsonData[0] == '{' {
+				// Rewrite JSON in the data line
+				rewritten := rw.rewriteModelInResponse(jsonData)
+				lines[i] = append([]byte("data: "), rewritten...)
+			}
+		}
+	}
+
+	return bytes.Join(lines, []byte("\n"))
+}

internal/api/modules/amp/routes.go

@@ -0,0 +1,268 @@
+package amp
+
+import (
+	"errors"
+	"net"
+	"net/http"
+	"net/http/httputil"
+	"strings"
+
+	"github.com/gin-gonic/gin"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/logging"
+	"github.com/router-for-me/CLIProxyAPI/v6/sdk/api/handlers"
+	"github.com/router-for-me/CLIProxyAPI/v6/sdk/api/handlers/claude"
+	"github.com/router-for-me/CLIProxyAPI/v6/sdk/api/handlers/gemini"
+	"github.com/router-for-me/CLIProxyAPI/v6/sdk/api/handlers/openai"
+	log "github.com/sirupsen/logrus"
+)
+
+// localhostOnlyMiddleware returns a middleware that dynamically checks the module's
+// localhost restriction setting. This allows hot-reload of the restriction without restarting.
+func (m *AmpModule) localhostOnlyMiddleware() gin.HandlerFunc {
+	return func(c *gin.Context) {
+		// Check current setting (hot-reloadable)
+		if !m.IsRestrictedToLocalhost() {
+			c.Next()
+			return
+		}
+
+		// Use actual TCP connection address (RemoteAddr) to prevent header spoofing
+		// This cannot be forged by X-Forwarded-For or other client-controlled headers
+		remoteAddr := c.Request.RemoteAddr
+
+		// RemoteAddr format is "IP:port" or "[IPv6]:port", extract just the IP
+		host, _, err := net.SplitHostPort(remoteAddr)
+		if err != nil {
+			// Try parsing as raw IP (shouldn't happen with standard HTTP, but be defensive)
+			host = remoteAddr
+		}
+
+		// Parse the IP to handle both IPv4 and IPv6
+		ip := net.ParseIP(host)
+		if ip == nil {
+			log.Warnf("amp management: invalid RemoteAddr %s, denying access", remoteAddr)
+			c.AbortWithStatusJSON(403, gin.H{
+				"error": "Access denied: management routes restricted to localhost",
+			})
+			return
+		}
+
+		// Check if IP is loopback (127.0.0.1 or ::1)
+		if !ip.IsLoopback() {
+			log.Warnf("amp management: non-localhost connection from %s attempted access, denying", remoteAddr)
+			c.AbortWithStatusJSON(403, gin.H{
+				"error": "Access denied: management routes restricted to localhost",
+			})
+			return
+		}
+
+		c.Next()
+	}
+}
+
+// noCORSMiddleware disables CORS for management routes to prevent browser-based attacks.
+// This overwrites any global CORS headers set by the server.
+func noCORSMiddleware() gin.HandlerFunc {
+	return func(c *gin.Context) {
+		// Remove CORS headers to prevent cross-origin access from browsers
+		c.Header("Access-Control-Allow-Origin", "")
+		c.Header("Access-Control-Allow-Methods", "")
+		c.Header("Access-Control-Allow-Headers", "")
+		c.Header("Access-Control-Allow-Credentials", "")
+
+		// For OPTIONS preflight, deny with 403
+		if c.Request.Method == "OPTIONS" {
+			c.AbortWithStatus(403)
+			return
+		}
+
+		c.Next()
+	}
+}
+
+// managementAvailabilityMiddleware short-circuits management routes when the upstream
+// proxy is disabled, preventing noisy localhost warnings and accidental exposure.
+func (m *AmpModule) managementAvailabilityMiddleware() gin.HandlerFunc {
+	return func(c *gin.Context) {
+		if m.getProxy() == nil {
+			logging.SkipGinRequestLogging(c)
+			c.AbortWithStatusJSON(http.StatusServiceUnavailable, gin.H{
+				"error": "amp upstream proxy not available",
+			})
+			return
+		}
+		c.Next()
+	}
+}
+
+// registerManagementRoutes registers Amp management proxy routes
+// These routes proxy through to the Amp control plane for OAuth, user management, etc.
+// Uses dynamic middleware and proxy getter for hot-reload support.
+func (m *AmpModule) registerManagementRoutes(engine *gin.Engine, baseHandler *handlers.BaseAPIHandler) {
+	ampAPI := engine.Group("/api")
+
+	// Always disable CORS for management routes to prevent browser-based attacks
+	ampAPI.Use(m.managementAvailabilityMiddleware(), noCORSMiddleware())
+
+	// Apply dynamic localhost-only restriction (hot-reloadable via m.IsRestrictedToLocalhost())
+	ampAPI.Use(m.localhostOnlyMiddleware())
+
+	if !m.IsRestrictedToLocalhost() {
+		log.Warn("amp management routes are NOT restricted to localhost - this is insecure!")
+	}
+
+	// Dynamic proxy handler that uses m.getProxy() for hot-reload support
+	proxyHandler := func(c *gin.Context) {
+		// Swallow ErrAbortHandler panics from ReverseProxy copyResponse to avoid noisy stack traces
+		defer func() {
+			if rec := recover(); rec != nil {
+				if err, ok := rec.(error); ok && errors.Is(err, http.ErrAbortHandler) {
+					// Upstream already wrote the status (often 404) before the client/stream ended.
+					return
+				}
+				panic(rec)
+			}
+		}()
+
+		proxy := m.getProxy()
+		if proxy == nil {
+			c.JSON(503, gin.H{"error": "amp upstream proxy not available"})
+			return
+		}
+		proxy.ServeHTTP(c.Writer, c.Request)
+	}
+
+	// Management routes - these are proxied directly to Amp upstream
+	ampAPI.Any("/internal", proxyHandler)
+	ampAPI.Any("/internal/*path", proxyHandler)
+	ampAPI.Any("/user", proxyHandler)
+	ampAPI.Any("/user/*path", proxyHandler)
+	ampAPI.Any("/auth", proxyHandler)
+	ampAPI.Any("/auth/*path", proxyHandler)
+	ampAPI.Any("/meta", proxyHandler)
+	ampAPI.Any("/meta/*path", proxyHandler)
+	ampAPI.Any("/ads", proxyHandler)
+	ampAPI.Any("/telemetry", proxyHandler)
+	ampAPI.Any("/telemetry/*path", proxyHandler)
+	ampAPI.Any("/threads", proxyHandler)
+	ampAPI.Any("/threads/*path", proxyHandler)
+	ampAPI.Any("/otel", proxyHandler)
+	ampAPI.Any("/otel/*path", proxyHandler)
+	ampAPI.Any("/tab", proxyHandler)
+	ampAPI.Any("/tab/*path", proxyHandler)
+
+	// Root-level routes that AMP CLI expects without /api prefix
+	// These need the same security middleware as the /api/* routes (dynamic for hot-reload)
+	rootMiddleware := []gin.HandlerFunc{m.managementAvailabilityMiddleware(), noCORSMiddleware(), m.localhostOnlyMiddleware()}
+	engine.GET("/threads/*path", append(rootMiddleware, proxyHandler)...)
+	engine.GET("/threads.rss", append(rootMiddleware, proxyHandler)...)
+	engine.GET("/news.rss", append(rootMiddleware, proxyHandler)...)
+
+	// Root-level auth routes for CLI login flow
+	// Amp uses multiple auth routes: /auth/cli-login, /auth/callback, /auth/sign-in, /auth/logout
+	// We proxy all /auth/* to support the complete OAuth flow
+	engine.Any("/auth", append(rootMiddleware, proxyHandler)...)
+	engine.Any("/auth/*path", append(rootMiddleware, proxyHandler)...)
+
+	// Google v1beta1 passthrough with OAuth fallback
+	// AMP CLI uses non-standard paths like /publishers/google/models/...
+	// We bridge these to our standard Gemini handler to enable local OAuth.
+	// If no local OAuth is available, falls back to ampcode.com proxy.
+	geminiHandlers := gemini.NewGeminiAPIHandler(baseHandler)
+	geminiBridge := createGeminiBridgeHandler(geminiHandlers.GeminiHandler)
+	geminiV1Beta1Fallback := NewFallbackHandlerWithMapper(func() *httputil.ReverseProxy {
+		return m.getProxy()
+	}, m.modelMapper, m.forceModelMappings)
+	geminiV1Beta1Handler := geminiV1Beta1Fallback.WrapHandler(geminiBridge)
+
+	// Route POST model calls through Gemini bridge with FallbackHandler.
+	// FallbackHandler checks provider -> mapping -> proxy fallback automatically.
+	// All other methods (e.g., GET model listing) always proxy to upstream to preserve Amp CLI behavior.
+	ampAPI.Any("/provider/google/v1beta1/*path", func(c *gin.Context) {
+		if c.Request.Method == "POST" {
+			if path := c.Param("path"); strings.Contains(path, "/models/") {
+				// POST with /models/ path -> use Gemini bridge with fallback handler
+				// FallbackHandler will check provider/mapping and proxy if needed
+				geminiV1Beta1Handler(c)
+				return
+			}
+		}
+		// Non-POST or no local provider available -> proxy upstream
+		proxyHandler(c)
+	})
+}
+
+// registerProviderAliases registers /api/provider/{provider}/... routes
+// These allow Amp CLI to route requests like:
+//
+//	/api/provider/openai/v1/chat/completions
+//	/api/provider/anthropic/v1/messages
+//	/api/provider/google/v1beta/models
+func (m *AmpModule) registerProviderAliases(engine *gin.Engine, baseHandler *handlers.BaseAPIHandler, auth gin.HandlerFunc) {
+	// Create handler instances for different providers
+	openaiHandlers := openai.NewOpenAIAPIHandler(baseHandler)
+	geminiHandlers := gemini.NewGeminiAPIHandler(baseHandler)
+	claudeCodeHandlers := claude.NewClaudeCodeAPIHandler(baseHandler)
+	openaiResponsesHandlers := openai.NewOpenAIResponsesAPIHandler(baseHandler)
+
+	// Create fallback handler wrapper that forwards to ampcode.com when provider not found
+	// Uses m.getProxy() for hot-reload support (proxy can be updated at runtime)
+	// Also includes model mapping support for routing unavailable models to alternatives
+	fallbackHandler := NewFallbackHandlerWithMapper(func() *httputil.ReverseProxy {
+		return m.getProxy()
+	}, m.modelMapper, m.forceModelMappings)
+
+	// Provider-specific routes under /api/provider/:provider
+	ampProviders := engine.Group("/api/provider")
+	if auth != nil {
+		ampProviders.Use(auth)
+	}
+
+	provider := ampProviders.Group("/:provider")
+
+	// Dynamic models handler - routes to appropriate provider based on path parameter
+	ampModelsHandler := func(c *gin.Context) {
+		providerName := strings.ToLower(c.Param("provider"))
+
+		switch providerName {
+		case "anthropic":
+			claudeCodeHandlers.ClaudeModels(c)
+		case "google":
+			geminiHandlers.GeminiModels(c)
+		default:
+			// Default to OpenAI-compatible (works for openai, groq, cerebras, etc.)
+			openaiHandlers.OpenAIModels(c)
+		}
+	}
+
+	// Root-level routes (for providers that omit /v1, like groq/cerebras)
+	// Wrap handlers with fallback logic to forward to ampcode.com when provider not found
+	provider.GET("/models", ampModelsHandler) // Models endpoint doesn't need fallback (no body to check)
+	provider.POST("/chat/completions", fallbackHandler.WrapHandler(openaiHandlers.ChatCompletions))
+	provider.POST("/completions", fallbackHandler.WrapHandler(openaiHandlers.Completions))
+	provider.POST("/responses", fallbackHandler.WrapHandler(openaiResponsesHandlers.Responses))
+
+	// /v1 routes (OpenAI/Claude-compatible endpoints)
+	v1Amp := provider.Group("/v1")
+	{
+		v1Amp.GET("/models", ampModelsHandler) // Models endpoint doesn't need fallback
+
+		// OpenAI-compatible endpoints with fallback
+		v1Amp.POST("/chat/completions", fallbackHandler.WrapHandler(openaiHandlers.ChatCompletions))
+		v1Amp.POST("/completions", fallbackHandler.WrapHandler(openaiHandlers.Completions))
+		v1Amp.POST("/responses", fallbackHandler.WrapHandler(openaiResponsesHandlers.Responses))
+
+		// Claude/Anthropic-compatible endpoints with fallback
+		v1Amp.POST("/messages", fallbackHandler.WrapHandler(claudeCodeHandlers.ClaudeMessages))
+		v1Amp.POST("/messages/count_tokens", fallbackHandler.WrapHandler(claudeCodeHandlers.ClaudeCountTokens))
+	}
+
+	// /v1beta routes (Gemini native API)
+	// Note: Gemini handler extracts model from URL path, so fallback logic needs special handling
+	v1betaAmp := provider.Group("/v1beta")
+	{
+		v1betaAmp.GET("/models", geminiHandlers.GeminiModels)
+		v1betaAmp.POST("/models/:action", fallbackHandler.WrapHandler(geminiHandlers.GeminiHandler))
+		v1betaAmp.GET("/models/:action", geminiHandlers.GeminiGetHandler)
+	}
+}

internal/api/modules/amp/routes_test.go

@@ -0,0 +1,373 @@
+package amp
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+	"github.com/router-for-me/CLIProxyAPI/v6/sdk/api/handlers"
+)
+
+func TestRegisterManagementRoutes(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+
+	// Create module with proxy for testing
+	m := &AmpModule{
+		restrictToLocalhost: false, // disable localhost restriction for tests
+	}
+
+	// Create a mock proxy that tracks calls
+	proxyCalled := false
+	mockProxy := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		proxyCalled = true
+		w.WriteHeader(200)
+		w.Write([]byte("proxied"))
+	}))
+	defer mockProxy.Close()
+
+	// Create real proxy to mock server
+	proxy, _ := createReverseProxy(mockProxy.URL, NewStaticSecretSource(""))
+	m.setProxy(proxy)
+
+	base := &handlers.BaseAPIHandler{}
+	m.registerManagementRoutes(r, base)
+
+	managementPaths := []struct {
+		path   string
+		method string
+	}{
+		{"/api/internal", http.MethodGet},
+		{"/api/internal/some/path", http.MethodGet},
+		{"/api/user", http.MethodGet},
+		{"/api/user/profile", http.MethodGet},
+		{"/api/auth", http.MethodGet},
+		{"/api/auth/login", http.MethodGet},
+		{"/api/meta", http.MethodGet},
+		{"/api/telemetry", http.MethodGet},
+		{"/api/threads", http.MethodGet},
+		{"/threads/", http.MethodGet},
+		{"/threads.rss", http.MethodGet}, // Root-level route (no /api prefix)
+		{"/api/otel", http.MethodGet},
+		{"/api/tab", http.MethodGet},
+		{"/api/tab/some/path", http.MethodGet},
+		{"/auth", http.MethodGet},           // Root-level auth route
+		{"/auth/cli-login", http.MethodGet}, // CLI login flow
+		{"/auth/callback", http.MethodGet},  // OAuth callback
+		// Google v1beta1 bridge should still proxy non-model requests (GET) and allow POST
+		{"/api/provider/google/v1beta1/models", http.MethodGet},
+		{"/api/provider/google/v1beta1/models", http.MethodPost},
+	}
+
+	for _, path := range managementPaths {
+		t.Run(path.path, func(t *testing.T) {
+			proxyCalled = false
+			req := httptest.NewRequest(path.method, path.path, nil)
+			w := httptest.NewRecorder()
+			r.ServeHTTP(w, req)
+
+			if w.Code == http.StatusNotFound {
+				t.Fatalf("route %s not registered", path.path)
+			}
+			if !proxyCalled {
+				t.Fatalf("proxy handler not called for %s", path.path)
+			}
+		})
+	}
+}
+
+func TestRegisterProviderAliases_AllProvidersRegistered(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+
+	// Minimal base handler setup (no need to initialize, just check routing)
+	base := &handlers.BaseAPIHandler{}
+
+	// Track if auth middleware was called
+	authCalled := false
+	authMiddleware := func(c *gin.Context) {
+		authCalled = true
+		c.Header("X-Auth", "ok")
+		// Abort with success to avoid calling the actual handler (which needs full setup)
+		c.AbortWithStatus(http.StatusOK)
+	}
+
+	m := &AmpModule{authMiddleware_: authMiddleware}
+	m.registerProviderAliases(r, base, authMiddleware)
+
+	paths := []struct {
+		path   string
+		method string
+	}{
+		{"/api/provider/openai/models", http.MethodGet},
+		{"/api/provider/anthropic/models", http.MethodGet},
+		{"/api/provider/google/models", http.MethodGet},
+		{"/api/provider/groq/models", http.MethodGet},
+		{"/api/provider/openai/chat/completions", http.MethodPost},
+		{"/api/provider/anthropic/v1/messages", http.MethodPost},
+		{"/api/provider/google/v1beta/models", http.MethodGet},
+	}
+
+	for _, tc := range paths {
+		t.Run(tc.path, func(t *testing.T) {
+			authCalled = false
+			req := httptest.NewRequest(tc.method, tc.path, nil)
+			w := httptest.NewRecorder()
+			r.ServeHTTP(w, req)
+
+			if w.Code == http.StatusNotFound {
+				t.Fatalf("route %s %s not registered", tc.method, tc.path)
+			}
+			if !authCalled {
+				t.Fatalf("auth middleware not executed for %s", tc.path)
+			}
+			if w.Header().Get("X-Auth") != "ok" {
+				t.Fatalf("auth middleware header not set for %s", tc.path)
+			}
+		})
+	}
+}
+
+func TestRegisterProviderAliases_DynamicModelsHandler(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+
+	base := &handlers.BaseAPIHandler{}
+
+	m := &AmpModule{authMiddleware_: func(c *gin.Context) { c.AbortWithStatus(http.StatusOK) }}
+	m.registerProviderAliases(r, base, func(c *gin.Context) { c.AbortWithStatus(http.StatusOK) })
+
+	providers := []string{"openai", "anthropic", "google", "groq", "cerebras"}
+
+	for _, provider := range providers {
+		t.Run(provider, func(t *testing.T) {
+			path := "/api/provider/" + provider + "/models"
+			req := httptest.NewRequest(http.MethodGet, path, nil)
+			w := httptest.NewRecorder()
+			r.ServeHTTP(w, req)
+
+			// Should not 404
+			if w.Code == http.StatusNotFound {
+				t.Fatalf("models route not found for provider: %s", provider)
+			}
+		})
+	}
+}
+
+func TestRegisterProviderAliases_V1Routes(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+
+	base := &handlers.BaseAPIHandler{}
+
+	m := &AmpModule{authMiddleware_: func(c *gin.Context) { c.AbortWithStatus(http.StatusOK) }}
+	m.registerProviderAliases(r, base, func(c *gin.Context) { c.AbortWithStatus(http.StatusOK) })
+
+	v1Paths := []struct {
+		path   string
+		method string
+	}{
+		{"/api/provider/openai/v1/models", http.MethodGet},
+		{"/api/provider/openai/v1/chat/completions", http.MethodPost},
+		{"/api/provider/openai/v1/completions", http.MethodPost},
+		{"/api/provider/anthropic/v1/messages", http.MethodPost},
+		{"/api/provider/anthropic/v1/messages/count_tokens", http.MethodPost},
+	}
+
+	for _, tc := range v1Paths {
+		t.Run(tc.path, func(t *testing.T) {
+			req := httptest.NewRequest(tc.method, tc.path, nil)
+			w := httptest.NewRecorder()
+			r.ServeHTTP(w, req)
+
+			if w.Code == http.StatusNotFound {
+				t.Fatalf("v1 route %s %s not registered", tc.method, tc.path)
+			}
+		})
+	}
+}
+
+func TestRegisterProviderAliases_V1BetaRoutes(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+
+	base := &handlers.BaseAPIHandler{}
+
+	m := &AmpModule{authMiddleware_: func(c *gin.Context) { c.AbortWithStatus(http.StatusOK) }}
+	m.registerProviderAliases(r, base, func(c *gin.Context) { c.AbortWithStatus(http.StatusOK) })
+
+	v1betaPaths := []struct {
+		path   string
+		method string
+	}{
+		{"/api/provider/google/v1beta/models", http.MethodGet},
+		{"/api/provider/google/v1beta/models/generateContent", http.MethodPost},
+	}
+
+	for _, tc := range v1betaPaths {
+		t.Run(tc.path, func(t *testing.T) {
+			req := httptest.NewRequest(tc.method, tc.path, nil)
+			w := httptest.NewRecorder()
+			r.ServeHTTP(w, req)
+
+			if w.Code == http.StatusNotFound {
+				t.Fatalf("v1beta route %s %s not registered", tc.method, tc.path)
+			}
+		})
+	}
+}
+
+func TestRegisterProviderAliases_NoAuthMiddleware(t *testing.T) {
+	// Test that routes still register even if auth middleware is nil (fallback behavior)
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+
+	base := &handlers.BaseAPIHandler{}
+
+	m := &AmpModule{authMiddleware_: nil} // No auth middleware
+	m.registerProviderAliases(r, base, func(c *gin.Context) { c.AbortWithStatus(http.StatusOK) })
+
+	req := httptest.NewRequest(http.MethodGet, "/api/provider/openai/models", nil)
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	// Should still work (with fallback no-op auth)
+	if w.Code == http.StatusNotFound {
+		t.Fatal("routes should register even without auth middleware")
+	}
+}
+
+func TestLocalhostOnlyMiddleware_PreventsSpoofing(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+
+	// Create module with localhost restriction enabled
+	m := &AmpModule{
+		restrictToLocalhost: true,
+	}
+
+	// Apply dynamic localhost-only middleware
+	r.Use(m.localhostOnlyMiddleware())
+	r.GET("/test", func(c *gin.Context) {
+		c.String(http.StatusOK, "ok")
+	})
+
+	tests := []struct {
+		name           string
+		remoteAddr     string
+		forwardedFor   string
+		expectedStatus int
+		description    string
+	}{
+		{
+			name:           "spoofed_header_remote_connection",
+			remoteAddr:     "192.168.1.100:12345",
+			forwardedFor:   "127.0.0.1",
+			expectedStatus: http.StatusForbidden,
+			description:    "Spoofed X-Forwarded-For header should be ignored",
+		},
+		{
+			name:           "real_localhost_ipv4",
+			remoteAddr:     "127.0.0.1:54321",
+			forwardedFor:   "",
+			expectedStatus: http.StatusOK,
+			description:    "Real localhost IPv4 connection should work",
+		},
+		{
+			name:           "real_localhost_ipv6",
+			remoteAddr:     "[::1]:54321",
+			forwardedFor:   "",
+			expectedStatus: http.StatusOK,
+			description:    "Real localhost IPv6 connection should work",
+		},
+		{
+			name:           "remote_ipv4",
+			remoteAddr:     "203.0.113.42:8080",
+			forwardedFor:   "",
+			expectedStatus: http.StatusForbidden,
+			description:    "Remote IPv4 connection should be blocked",
+		},
+		{
+			name:           "remote_ipv6",
+			remoteAddr:     "[2001:db8::1]:9090",
+			forwardedFor:   "",
+			expectedStatus: http.StatusForbidden,
+			description:    "Remote IPv6 connection should be blocked",
+		},
+		{
+			name:           "spoofed_localhost_ipv6",
+			remoteAddr:     "203.0.113.42:8080",
+			forwardedFor:   "::1",
+			expectedStatus: http.StatusForbidden,
+			description:    "Spoofed X-Forwarded-For with IPv6 localhost should be ignored",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			req := httptest.NewRequest(http.MethodGet, "/test", nil)
+			req.RemoteAddr = tt.remoteAddr
+			if tt.forwardedFor != "" {
+				req.Header.Set("X-Forwarded-For", tt.forwardedFor)
+			}
+
+			w := httptest.NewRecorder()
+			r.ServeHTTP(w, req)
+
+			if w.Code != tt.expectedStatus {
+				t.Errorf("%s: expected status %d, got %d", tt.description, tt.expectedStatus, w.Code)
+			}
+		})
+	}
+}
+
+func TestLocalhostOnlyMiddleware_HotReload(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	r := gin.New()
+
+	// Create module with localhost restriction initially enabled
+	m := &AmpModule{
+		restrictToLocalhost: true,
+	}
+
+	// Apply dynamic localhost-only middleware
+	r.Use(m.localhostOnlyMiddleware())
+	r.GET("/test", func(c *gin.Context) {
+		c.String(http.StatusOK, "ok")
+	})
+
+	// Test 1: Remote IP should be blocked when restriction is enabled
+	req := httptest.NewRequest(http.MethodGet, "/test", nil)
+	req.RemoteAddr = "192.168.1.100:12345"
+	w := httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusForbidden {
+		t.Errorf("Expected 403 when restriction enabled, got %d", w.Code)
+	}
+
+	// Test 2: Hot-reload - disable restriction
+	m.setRestrictToLocalhost(false)
+
+	req = httptest.NewRequest(http.MethodGet, "/test", nil)
+	req.RemoteAddr = "192.168.1.100:12345"
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusOK {
+		t.Errorf("Expected 200 after disabling restriction, got %d", w.Code)
+	}
+
+	// Test 3: Hot-reload - re-enable restriction
+	m.setRestrictToLocalhost(true)
+
+	req = httptest.NewRequest(http.MethodGet, "/test", nil)
+	req.RemoteAddr = "192.168.1.100:12345"
+	w = httptest.NewRecorder()
+	r.ServeHTTP(w, req)
+
+	if w.Code != http.StatusForbidden {
+		t.Errorf("Expected 403 after re-enabling restriction, got %d", w.Code)
+	}
+}

internal/api/modules/amp/secret.go

@@ -0,0 +1,166 @@
+package amp
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+	"time"
+)
+
+// SecretSource provides Amp API keys with configurable precedence and caching
+type SecretSource interface {
+	Get(ctx context.Context) (string, error)
+}
+
+// cachedSecret holds a secret value with expiration
+type cachedSecret struct {
+	value     string
+	expiresAt time.Time
+}
+
+// MultiSourceSecret implements precedence-based secret lookup:
+// 1. Explicit config value (highest priority)
+// 2. Environment variable AMP_API_KEY
+// 3. File-based secret (lowest priority)
+type MultiSourceSecret struct {
+	explicitKey string
+	envKey      string
+	filePath    string
+	cacheTTL    time.Duration
+
+	mu    sync.RWMutex
+	cache *cachedSecret
+}
+
+// NewMultiSourceSecret creates a secret source with precedence and caching
+func NewMultiSourceSecret(explicitKey string, cacheTTL time.Duration) *MultiSourceSecret {
+	if cacheTTL == 0 {
+		cacheTTL = 5 * time.Minute // Default 5 minute cache
+	}
+
+	home, _ := os.UserHomeDir()
+	filePath := filepath.Join(home, ".local", "share", "amp", "secrets.json")
+
+	return &MultiSourceSecret{
+		explicitKey: strings.TrimSpace(explicitKey),
+		envKey:      "AMP_API_KEY",
+		filePath:    filePath,
+		cacheTTL:    cacheTTL,
+	}
+}
+
+// NewMultiSourceSecretWithPath creates a secret source with a custom file path (for testing)
+func NewMultiSourceSecretWithPath(explicitKey string, filePath string, cacheTTL time.Duration) *MultiSourceSecret {
+	if cacheTTL == 0 {
+		cacheTTL = 5 * time.Minute
+	}
+
+	return &MultiSourceSecret{
+		explicitKey: strings.TrimSpace(explicitKey),
+		envKey:      "AMP_API_KEY",
+		filePath:    filePath,
+		cacheTTL:    cacheTTL,
+	}
+}
+
+// Get retrieves the Amp API key using precedence: config > env > file
+// Results are cached for cacheTTL duration to avoid excessive file reads
+func (s *MultiSourceSecret) Get(ctx context.Context) (string, error) {
+	// Precedence 1: Explicit config key (highest priority, no caching needed)
+	if s.explicitKey != "" {
+		return s.explicitKey, nil
+	}
+
+	// Precedence 2: Environment variable
+	if envValue := strings.TrimSpace(os.Getenv(s.envKey)); envValue != "" {
+		return envValue, nil
+	}
+
+	// Precedence 3: File-based secret (lowest priority, cached)
+	// Check cache first
+	s.mu.RLock()
+	if s.cache != nil && time.Now().Before(s.cache.expiresAt) {
+		value := s.cache.value
+		s.mu.RUnlock()
+		return value, nil
+	}
+	s.mu.RUnlock()
+
+	// Cache miss or expired - read from file
+	key, err := s.readFromFile()
+	if err != nil {
+		// Cache empty result to avoid repeated file reads on missing files
+		s.updateCache("")
+		return "", err
+	}
+
+	// Cache the result
+	s.updateCache(key)
+	return key, nil
+}
+
+// readFromFile reads the Amp API key from the secrets file
+func (s *MultiSourceSecret) readFromFile() (string, error) {
+	content, err := os.ReadFile(s.filePath)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return "", nil // Missing file is not an error, just no key available
+		}
+		return "", fmt.Errorf("failed to read amp secrets from %s: %w", s.filePath, err)
+	}
+
+	var secrets map[string]string
+	if err := json.Unmarshal(content, &secrets); err != nil {
+		return "", fmt.Errorf("failed to parse amp secrets from %s: %w", s.filePath, err)
+	}
+
+	key := strings.TrimSpace(secrets["apiKey@https://ampcode.com/"])
+	return key, nil
+}
+
+// updateCache updates the cached secret value
+func (s *MultiSourceSecret) updateCache(value string) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	s.cache = &cachedSecret{
+		value:     value,
+		expiresAt: time.Now().Add(s.cacheTTL),
+	}
+}
+
+// InvalidateCache clears the cached secret, forcing a fresh read on next Get
+func (s *MultiSourceSecret) InvalidateCache() {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	s.cache = nil
+}
+
+// UpdateExplicitKey refreshes the config-provided key and clears cache.
+func (s *MultiSourceSecret) UpdateExplicitKey(key string) {
+	if s == nil {
+		return
+	}
+	s.mu.Lock()
+	s.explicitKey = strings.TrimSpace(key)
+	s.cache = nil
+	s.mu.Unlock()
+}
+
+// StaticSecretSource returns a fixed API key (for testing)
+type StaticSecretSource struct {
+	key string
+}
+
+// NewStaticSecretSource creates a secret source with a fixed key
+func NewStaticSecretSource(key string) *StaticSecretSource {
+	return &StaticSecretSource{key: strings.TrimSpace(key)}
+}
+
+// Get returns the static API key
+func (s *StaticSecretSource) Get(ctx context.Context) (string, error) {
+	return s.key, nil
+}

internal/api/modules/amp/secret_test.go

@@ -0,0 +1,280 @@
+package amp
+
+import (
+	"context"
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"sync"
+	"testing"
+	"time"
+)
+
+func TestMultiSourceSecret_PrecedenceOrder(t *testing.T) {
+	ctx := context.Background()
+
+	cases := []struct {
+		name      string
+		configKey string
+		envKey    string
+		fileJSON  string
+		want      string
+	}{
+		{"config_wins", "cfg", "env", `{"apiKey@https://ampcode.com/":"file"}`, "cfg"},
+		{"env_wins_when_no_cfg", "", "env", `{"apiKey@https://ampcode.com/":"file"}`, "env"},
+		{"file_when_no_cfg_env", "", "", `{"apiKey@https://ampcode.com/":"file"}`, "file"},
+		{"empty_cfg_trims_then_env", "   ", "env", `{"apiKey@https://ampcode.com/":"file"}`, "env"},
+		{"empty_env_then_file", "", "   ", `{"apiKey@https://ampcode.com/":"file"}`, "file"},
+		{"missing_file_returns_empty", "", "", "", ""},
+		{"all_empty_returns_empty", "  ", "  ", `{"apiKey@https://ampcode.com/":"  "}`, ""},
+	}
+
+	for _, tc := range cases {
+		tc := tc // capture range variable
+		t.Run(tc.name, func(t *testing.T) {
+			tmpDir := t.TempDir()
+			secretsPath := filepath.Join(tmpDir, "secrets.json")
+
+			if tc.fileJSON != "" {
+				if err := os.WriteFile(secretsPath, []byte(tc.fileJSON), 0600); err != nil {
+					t.Fatal(err)
+				}
+			}
+
+			t.Setenv("AMP_API_KEY", tc.envKey)
+
+			s := NewMultiSourceSecretWithPath(tc.configKey, secretsPath, 100*time.Millisecond)
+			got, err := s.Get(ctx)
+			if err != nil && tc.fileJSON != "" && json.Valid([]byte(tc.fileJSON)) {
+				t.Fatalf("unexpected error: %v", err)
+			}
+			if got != tc.want {
+				t.Fatalf("want %q, got %q", tc.want, got)
+			}
+		})
+	}
+}
+
+func TestMultiSourceSecret_CacheBehavior(t *testing.T) {
+	ctx := context.Background()
+	tmpDir := t.TempDir()
+	p := filepath.Join(tmpDir, "secrets.json")
+
+	// Initial value
+	if err := os.WriteFile(p, []byte(`{"apiKey@https://ampcode.com/":"v1"}`), 0600); err != nil {
+		t.Fatal(err)
+	}
+
+	s := NewMultiSourceSecretWithPath("", p, 50*time.Millisecond)
+
+	// First read - should return v1
+	got1, err := s.Get(ctx)
+	if err != nil {
+		t.Fatalf("Get failed: %v", err)
+	}
+	if got1 != "v1" {
+		t.Fatalf("expected v1, got %s", got1)
+	}
+
+	// Change file; within TTL we should still see v1 (cached)
+	if err := os.WriteFile(p, []byte(`{"apiKey@https://ampcode.com/":"v2"}`), 0600); err != nil {
+		t.Fatal(err)
+	}
+	got2, _ := s.Get(ctx)
+	if got2 != "v1" {
+		t.Fatalf("cache hit expected v1, got %s", got2)
+	}
+
+	// After TTL expires, should see v2
+	time.Sleep(60 * time.Millisecond)
+	got3, _ := s.Get(ctx)
+	if got3 != "v2" {
+		t.Fatalf("cache miss expected v2, got %s", got3)
+	}
+
+	// Invalidate forces re-read immediately
+	if err := os.WriteFile(p, []byte(`{"apiKey@https://ampcode.com/":"v3"}`), 0600); err != nil {
+		t.Fatal(err)
+	}
+	s.InvalidateCache()
+	got4, _ := s.Get(ctx)
+	if got4 != "v3" {
+		t.Fatalf("invalidate expected v3, got %s", got4)
+	}
+}
+
+func TestMultiSourceSecret_FileHandling(t *testing.T) {
+	ctx := context.Background()
+
+	t.Run("missing_file_no_error", func(t *testing.T) {
+		s := NewMultiSourceSecretWithPath("", "/nonexistent/path/secrets.json", 100*time.Millisecond)
+		got, err := s.Get(ctx)
+		if err != nil {
+			t.Fatalf("expected no error for missing file, got: %v", err)
+		}
+		if got != "" {
+			t.Fatalf("expected empty string, got %q", got)
+		}
+	})
+
+	t.Run("invalid_json", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		p := filepath.Join(tmpDir, "secrets.json")
+		if err := os.WriteFile(p, []byte(`{invalid json`), 0600); err != nil {
+			t.Fatal(err)
+		}
+
+		s := NewMultiSourceSecretWithPath("", p, 100*time.Millisecond)
+		_, err := s.Get(ctx)
+		if err == nil {
+			t.Fatal("expected error for invalid JSON")
+		}
+	})
+
+	t.Run("missing_key_in_json", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		p := filepath.Join(tmpDir, "secrets.json")
+		if err := os.WriteFile(p, []byte(`{"other":"value"}`), 0600); err != nil {
+			t.Fatal(err)
+		}
+
+		s := NewMultiSourceSecretWithPath("", p, 100*time.Millisecond)
+		got, err := s.Get(ctx)
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		if got != "" {
+			t.Fatalf("expected empty string for missing key, got %q", got)
+		}
+	})
+
+	t.Run("empty_key_value", func(t *testing.T) {
+		tmpDir := t.TempDir()
+		p := filepath.Join(tmpDir, "secrets.json")
+		if err := os.WriteFile(p, []byte(`{"apiKey@https://ampcode.com/":"   "}`), 0600); err != nil {
+			t.Fatal(err)
+		}
+
+		s := NewMultiSourceSecretWithPath("", p, 100*time.Millisecond)
+		got, _ := s.Get(ctx)
+		if got != "" {
+			t.Fatalf("expected empty after trim, got %q", got)
+		}
+	})
+}
+
+func TestMultiSourceSecret_Concurrency(t *testing.T) {
+	tmpDir := t.TempDir()
+	p := filepath.Join(tmpDir, "secrets.json")
+	if err := os.WriteFile(p, []byte(`{"apiKey@https://ampcode.com/":"concurrent"}`), 0600); err != nil {
+		t.Fatal(err)
+	}
+
+	s := NewMultiSourceSecretWithPath("", p, 5*time.Second)
+	ctx := context.Background()
+
+	// Spawn many goroutines calling Get concurrently
+	const goroutines = 50
+	const iterations = 100
+
+	var wg sync.WaitGroup
+	errors := make(chan error, goroutines)
+
+	for i := 0; i < goroutines; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			for j := 0; j < iterations; j++ {
+				val, err := s.Get(ctx)
+				if err != nil {
+					errors <- err
+					return
+				}
+				if val != "concurrent" {
+					errors <- err
+					return
+				}
+			}
+		}()
+	}
+
+	wg.Wait()
+	close(errors)
+
+	for err := range errors {
+		t.Errorf("concurrency error: %v", err)
+	}
+}
+
+func TestStaticSecretSource(t *testing.T) {
+	ctx := context.Background()
+
+	t.Run("returns_provided_key", func(t *testing.T) {
+		s := NewStaticSecretSource("test-key-123")
+		got, err := s.Get(ctx)
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		if got != "test-key-123" {
+			t.Fatalf("want test-key-123, got %q", got)
+		}
+	})
+
+	t.Run("trims_whitespace", func(t *testing.T) {
+		s := NewStaticSecretSource("  test-key  ")
+		got, err := s.Get(ctx)
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		if got != "test-key" {
+			t.Fatalf("want test-key, got %q", got)
+		}
+	})
+
+	t.Run("empty_string", func(t *testing.T) {
+		s := NewStaticSecretSource("")
+		got, err := s.Get(ctx)
+		if err != nil {
+			t.Fatalf("unexpected error: %v", err)
+		}
+		if got != "" {
+			t.Fatalf("want empty string, got %q", got)
+		}
+	})
+}
+
+func TestMultiSourceSecret_CacheEmptyResult(t *testing.T) {
+	// Test that missing file results are cached to avoid repeated file reads
+	tmpDir := t.TempDir()
+	p := filepath.Join(tmpDir, "nonexistent.json")
+
+	s := NewMultiSourceSecretWithPath("", p, 100*time.Millisecond)
+	ctx := context.Background()
+
+	// First call - file doesn't exist, should cache empty result
+	got1, err := s.Get(ctx)
+	if err != nil {
+		t.Fatalf("expected no error for missing file, got: %v", err)
+	}
+	if got1 != "" {
+		t.Fatalf("expected empty string, got %q", got1)
+	}
+
+	// Create the file now
+	if err := os.WriteFile(p, []byte(`{"apiKey@https://ampcode.com/":"new-value"}`), 0600); err != nil {
+		t.Fatal(err)
+	}
+
+	// Second call - should still return empty (cached), not read the new file
+	got2, _ := s.Get(ctx)
+	if got2 != "" {
+		t.Fatalf("cache should return empty, got %q", got2)
+	}
+
+	// After TTL expires, should see the new value
+	time.Sleep(110 * time.Millisecond)
+	got3, _ := s.Get(ctx)
+	if got3 != "new-value" {
+		t.Fatalf("after cache expiry, expected new-value, got %q", got3)
+	}
+}

internal/api/modules/modules.go

@@ -0,0 +1,92 @@
+// Package modules provides a pluggable routing module system for extending
+// the API server with optional features without modifying core routing logic.
+package modules
+
+import (
+	"fmt"
+
+	"github.com/gin-gonic/gin"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	"github.com/router-for-me/CLIProxyAPI/v6/sdk/api/handlers"
+)
+
+// Context encapsulates the dependencies exposed to routing modules during
+// registration. Modules can use the Gin engine to attach routes, the shared
+// BaseAPIHandler for constructing SDK-specific handlers, and the resolved
+// authentication middleware for protecting routes that require API keys.
+type Context struct {
+	Engine         *gin.Engine
+	BaseHandler    *handlers.BaseAPIHandler
+	Config         *config.Config
+	AuthMiddleware gin.HandlerFunc
+}
+
+// RouteModule represents a pluggable routing module that can register routes
+// and handle configuration updates independently of the core server.
+//
+// DEPRECATED: Use RouteModuleV2 for new modules. This interface is kept for
+// backwards compatibility and will be removed in a future version.
+type RouteModule interface {
+	// Name returns a human-readable identifier for the module
+	Name() string
+
+	// Register sets up routes and handlers for this module.
+	// It receives the Gin engine, base handlers, and current configuration.
+	// Returns an error if registration fails (errors are logged but don't stop the server).
+	Register(engine *gin.Engine, baseHandler *handlers.BaseAPIHandler, cfg *config.Config) error
+
+	// OnConfigUpdated is called when the configuration is reloaded.
+	// Modules can respond to configuration changes here.
+	// Returns an error if the update cannot be applied.
+	OnConfigUpdated(cfg *config.Config) error
+}
+
+// RouteModuleV2 represents a pluggable bundle of routes that can integrate with
+// the API server without modifying its core routing logic. Implementations can
+// attach routes during Register and react to configuration updates via
+// OnConfigUpdated.
+//
+// This is the preferred interface for new modules. It uses Context for cleaner
+// dependency injection and supports idempotent registration.
+type RouteModuleV2 interface {
+	// Name returns a unique identifier for logging and diagnostics.
+	Name() string
+
+	// Register wires the module's routes into the provided Gin engine. Modules
+	// should treat multiple calls as idempotent and avoid duplicate route
+	// registration when invoked more than once.
+	Register(ctx Context) error
+
+	// OnConfigUpdated notifies the module when the server configuration changes
+	// via hot reload. Implementations can refresh cached state or emit warnings.
+	OnConfigUpdated(cfg *config.Config) error
+}
+
+// RegisterModule is a helper that registers a module using either the V1 or V2
+// interface. This allows gradual migration from V1 to V2 without breaking
+// existing modules.
+//
+// Example usage:
+//
+//	ctx := modules.Context{
+//	    Engine:         engine,
+//	    BaseHandler:    baseHandler,
+//	    Config:         cfg,
+//	    AuthMiddleware: authMiddleware,
+//	}
+//	if err := modules.RegisterModule(ctx, ampModule); err != nil {
+//	    log.Errorf("Failed to register module: %v", err)
+//	}
+func RegisterModule(ctx Context, mod interface{}) error {
+	// Try V2 interface first (preferred)
+	if v2, ok := mod.(RouteModuleV2); ok {
+		return v2.Register(ctx)
+	}
+
+	// Fall back to V1 interface for backwards compatibility
+	if v1, ok := mod.(RouteModule); ok {
+		return v1.Register(ctx.Engine, ctx.BaseHandler, ctx.Config)
+	}
+
+	return fmt.Errorf("unsupported module type %T (must implement RouteModule or RouteModuleV2)", mod)
+}

internal/api/server.go

@@ -21,6 +21,8 @@ import (
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/access"
 	managementHandlers "github.com/router-for-me/CLIProxyAPI/v6/internal/api/handlers/management"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/api/middleware"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/api/modules"
+	ampmodule "github.com/router-for-me/CLIProxyAPI/v6/internal/api/modules/amp"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/logging"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/managementasset"
@@ -148,6 +150,9 @@ type Server struct {
 	// management handler
 	mgmt *managementHandlers.Handler
 
+	// ampModule is the Amp routing module for model mapping hot-reload
+	ampModule *ampmodule.AmpModule
+
 	// managementRoutesRegistered tracks whether the management routes have been attached to the engine.
 	managementRoutesRegistered atomic.Bool
 	// managementRoutesEnabled controls whether management endpoints serve real handlers.
@@ -245,6 +250,9 @@ func NewServer(cfg *config.Config, authManager *auth.Manager, accessManager *sdk
 	// Save initial YAML snapshot
 	s.oldConfigYaml, _ = yaml.Marshal(cfg)
 	s.applyAccessConfig(nil, cfg)
+	if authManager != nil {
+		authManager.SetRetryConfig(cfg.RequestRetry, time.Duration(cfg.MaxRetryInterval)*time.Second)
+	}
 	managementasset.SetCurrentConfig(cfg)
 	auth.SetQuotaCooldownDisabled(cfg.DisableCooling)
 	// Initialize management handler
@@ -261,6 +269,20 @@ func NewServer(cfg *config.Config, authManager *auth.Manager, accessManager *sdk
 
 	// Setup routes
 	s.setupRoutes()
+
+	// Register Amp module using V2 interface with Context
+	s.ampModule = ampmodule.NewLegacy(accessManager, AuthMiddleware(accessManager))
+	ctx := modules.Context{
+		Engine:         engine,
+		BaseHandler:    s.handlers,
+		Config:         cfg,
+		AuthMiddleware: AuthMiddleware(accessManager),
+	}
+	if err := modules.RegisterModule(ctx, s.ampModule); err != nil {
+		log.Errorf("Failed to register Amp module: %v", err)
+	}
+
+	// Apply additional router configurators from options
 	if optionState.routerConfigurator != nil {
 		optionState.routerConfigurator(engine, s.handlers, cfg)
 	}
@@ -278,7 +300,7 @@ func NewServer(cfg *config.Config, authManager *auth.Manager, accessManager *sdk
 
 	// Create HTTP server
 	s.server = &http.Server{
-		Addr:    fmt.Sprintf(":%d", cfg.Port),
+		Addr:    fmt.Sprintf("%s:%d", cfg.Host, cfg.Port),
 		Handler: engine,
 	}
 
@@ -381,6 +403,18 @@ func (s *Server) setupRoutes() {
 		c.String(http.StatusOK, oauthCallbackSuccessHTML)
 	})
 
+	s.engine.GET("/antigravity/callback", func(c *gin.Context) {
+		code := c.Query("code")
+		state := c.Query("state")
+		errStr := c.Query("error")
+		if state != "" {
+			file := fmt.Sprintf("%s/.oauth-antigravity-%s.oauth", s.cfg.AuthDir, state)
+			_ = os.WriteFile(file, []byte(fmt.Sprintf(`{"code":"%s","state":"%s","error":"%s"}`, code, state, errStr)), 0o600)
+		}
+		c.Header("Content-Type", "text/html; charset=utf-8")
+		c.String(http.StatusOK, oauthCallbackSuccessHTML)
+	})
+
 	// Management routes are registered lazily by registerManagementRoutes when a secret is configured.
 }
 
@@ -436,8 +470,9 @@ func (s *Server) registerManagementRoutes() {
 	{
 		mgmt.GET("/usage", s.mgmt.GetUsageStatistics)
 		mgmt.GET("/config", s.mgmt.GetConfig)
+		mgmt.GET("/config.yaml", s.mgmt.GetConfigYAML)
 		mgmt.PUT("/config.yaml", s.mgmt.PutConfigYAML)
-		mgmt.GET("/config.yaml", s.mgmt.GetConfigFile)
+		mgmt.GET("/latest-version", s.mgmt.GetLatestVersion)
 
 		mgmt.GET("/debug", s.mgmt.GetDebug)
 		mgmt.PUT("/debug", s.mgmt.PutDebug)
@@ -469,11 +504,6 @@ func (s *Server) registerManagementRoutes() {
 		mgmt.PATCH("/api-keys", s.mgmt.PatchAPIKeys)
 		mgmt.DELETE("/api-keys", s.mgmt.DeleteAPIKeys)
 
-		mgmt.GET("/generative-language-api-key", s.mgmt.GetGlKeys)
-		mgmt.PUT("/generative-language-api-key", s.mgmt.PutGlKeys)
-		mgmt.PATCH("/generative-language-api-key", s.mgmt.PatchGlKeys)
-		mgmt.DELETE("/generative-language-api-key", s.mgmt.DeleteGlKeys)
-
 		mgmt.GET("/gemini-api-key", s.mgmt.GetGeminiKeys)
 		mgmt.PUT("/gemini-api-key", s.mgmt.PutGeminiKeys)
 		mgmt.PATCH("/gemini-api-key", s.mgmt.PatchGeminiKey)
@@ -481,6 +511,8 @@ func (s *Server) registerManagementRoutes() {
 
 		mgmt.GET("/logs", s.mgmt.GetLogs)
 		mgmt.DELETE("/logs", s.mgmt.DeleteLogs)
+		mgmt.GET("/request-error-logs", s.mgmt.GetRequestErrorLogs)
+		mgmt.GET("/request-error-logs/:name", s.mgmt.DownloadRequestErrorLog)
 		mgmt.GET("/request-log", s.mgmt.GetRequestLog)
 		mgmt.PUT("/request-log", s.mgmt.PutRequestLog)
 		mgmt.PATCH("/request-log", s.mgmt.PutRequestLog)
@@ -488,9 +520,32 @@ func (s *Server) registerManagementRoutes() {
 		mgmt.PUT("/ws-auth", s.mgmt.PutWebsocketAuth)
 		mgmt.PATCH("/ws-auth", s.mgmt.PutWebsocketAuth)
 
+		mgmt.GET("/ampcode", s.mgmt.GetAmpCode)
+		mgmt.GET("/ampcode/upstream-url", s.mgmt.GetAmpUpstreamURL)
+		mgmt.PUT("/ampcode/upstream-url", s.mgmt.PutAmpUpstreamURL)
+		mgmt.PATCH("/ampcode/upstream-url", s.mgmt.PutAmpUpstreamURL)
+		mgmt.DELETE("/ampcode/upstream-url", s.mgmt.DeleteAmpUpstreamURL)
+		mgmt.GET("/ampcode/upstream-api-key", s.mgmt.GetAmpUpstreamAPIKey)
+		mgmt.PUT("/ampcode/upstream-api-key", s.mgmt.PutAmpUpstreamAPIKey)
+		mgmt.PATCH("/ampcode/upstream-api-key", s.mgmt.PutAmpUpstreamAPIKey)
+		mgmt.DELETE("/ampcode/upstream-api-key", s.mgmt.DeleteAmpUpstreamAPIKey)
+		mgmt.GET("/ampcode/restrict-management-to-localhost", s.mgmt.GetAmpRestrictManagementToLocalhost)
+		mgmt.PUT("/ampcode/restrict-management-to-localhost", s.mgmt.PutAmpRestrictManagementToLocalhost)
+		mgmt.PATCH("/ampcode/restrict-management-to-localhost", s.mgmt.PutAmpRestrictManagementToLocalhost)
+		mgmt.GET("/ampcode/model-mappings", s.mgmt.GetAmpModelMappings)
+		mgmt.PUT("/ampcode/model-mappings", s.mgmt.PutAmpModelMappings)
+		mgmt.PATCH("/ampcode/model-mappings", s.mgmt.PatchAmpModelMappings)
+		mgmt.DELETE("/ampcode/model-mappings", s.mgmt.DeleteAmpModelMappings)
+		mgmt.GET("/ampcode/force-model-mappings", s.mgmt.GetAmpForceModelMappings)
+		mgmt.PUT("/ampcode/force-model-mappings", s.mgmt.PutAmpForceModelMappings)
+		mgmt.PATCH("/ampcode/force-model-mappings", s.mgmt.PutAmpForceModelMappings)
+
 		mgmt.GET("/request-retry", s.mgmt.GetRequestRetry)
 		mgmt.PUT("/request-retry", s.mgmt.PutRequestRetry)
 		mgmt.PATCH("/request-retry", s.mgmt.PutRequestRetry)
+		mgmt.GET("/max-retry-interval", s.mgmt.GetMaxRetryInterval)
+		mgmt.PUT("/max-retry-interval", s.mgmt.PutMaxRetryInterval)
+		mgmt.PATCH("/max-retry-interval", s.mgmt.PutMaxRetryInterval)
 
 		mgmt.GET("/claude-api-key", s.mgmt.GetClaudeKeys)
 		mgmt.PUT("/claude-api-key", s.mgmt.PutClaudeKeys)
@@ -507,6 +562,11 @@ func (s *Server) registerManagementRoutes() {
 		mgmt.PATCH("/openai-compatibility", s.mgmt.PatchOpenAICompat)
 		mgmt.DELETE("/openai-compatibility", s.mgmt.DeleteOpenAICompat)
 
+		mgmt.GET("/oauth-excluded-models", s.mgmt.GetOAuthExcludedModels)
+		mgmt.PUT("/oauth-excluded-models", s.mgmt.PutOAuthExcludedModels)
+		mgmt.PATCH("/oauth-excluded-models", s.mgmt.PatchOAuthExcludedModels)
+		mgmt.DELETE("/oauth-excluded-models", s.mgmt.DeleteOAuthExcludedModels)
+
 		mgmt.GET("/auth-files", s.mgmt.ListAuthFiles)
 		mgmt.GET("/auth-files/download", s.mgmt.DownloadAuthFile)
 		mgmt.POST("/auth-files", s.mgmt.UploadAuthFile)
@@ -516,8 +576,10 @@ func (s *Server) registerManagementRoutes() {
 		mgmt.GET("/anthropic-auth-url", s.mgmt.RequestAnthropicToken)
 		mgmt.GET("/codex-auth-url", s.mgmt.RequestCodexToken)
 		mgmt.GET("/gemini-cli-auth-url", s.mgmt.RequestGeminiCLIToken)
+		mgmt.GET("/antigravity-auth-url", s.mgmt.RequestAntigravityToken)
 		mgmt.GET("/qwen-auth-url", s.mgmt.RequestQwenToken)
 		mgmt.GET("/iflow-auth-url", s.mgmt.RequestIFlowToken)
+		mgmt.POST("/iflow-auth-url", s.mgmt.RequestIFlowCookieToken)
 		mgmt.GET("/get-auth-status", s.mgmt.GetAuthStatus)
 	}
 }
@@ -656,17 +718,33 @@ func (s *Server) unifiedModelsHandler(openaiHandler *openai.OpenAIAPIHandler, cl
 	}
 }
 
-// Start begins listening for and serving HTTP requests.
+// Start begins listening for and serving HTTP or HTTPS requests.
 // It's a blocking call and will only return on an unrecoverable error.
 //
 // Returns:
 //   - error: An error if the server fails to start
 func (s *Server) Start() error {
-	log.Debugf("Starting API server on %s", s.server.Addr)
+	if s == nil || s.server == nil {
+		return fmt.Errorf("failed to start HTTP server: server not initialized")
+	}
 
-	// Start the HTTP server.
-	if err := s.server.ListenAndServe(); err != nil && !errors.Is(err, http.ErrServerClosed) {
-		return fmt.Errorf("failed to start HTTP server: %v", err)
+	useTLS := s.cfg != nil && s.cfg.TLS.Enable
+	if useTLS {
+		cert := strings.TrimSpace(s.cfg.TLS.Cert)
+		key := strings.TrimSpace(s.cfg.TLS.Key)
+		if cert == "" || key == "" {
+			return fmt.Errorf("failed to start HTTPS server: tls.cert or tls.key is empty")
+		}
+		log.Debugf("Starting API server on %s with TLS", s.server.Addr)
+		if errServeTLS := s.server.ListenAndServeTLS(cert, key); errServeTLS != nil && !errors.Is(errServeTLS, http.ErrServerClosed) {
+			return fmt.Errorf("failed to start HTTPS server: %v", errServeTLS)
+		}
+		return nil
+	}
+
+	log.Debugf("Starting API server on %s", s.server.Addr)
+	if errServe := s.server.ListenAndServe(); errServe != nil && !errors.Is(errServe, http.ErrServerClosed) {
+		return fmt.Errorf("failed to start HTTP server: %v", errServe)
 	}
 
 	return nil
@@ -784,6 +862,9 @@ func (s *Server) UpdateClients(cfg *config.Config) {
 			log.Debugf("disable_cooling toggled to %t", cfg.DisableCooling)
 		}
 	}
+	if s.handlers != nil && s.handlers.AuthManager != nil {
+		s.handlers.AuthManager.SetRetryConfig(cfg.RequestRetry, time.Duration(cfg.MaxRetryInterval)*time.Second)
+	}
 
 	// Update log level dynamically when debug flag changes
 	if oldCfg == nil || oldCfg.Debug != cfg.Debug {
@@ -854,28 +935,36 @@ func (s *Server) UpdateClients(cfg *config.Config) {
 		s.mgmt.SetAuthManager(s.handlers.AuthManager)
 	}
 
+	// Notify Amp module of config changes (for model mapping hot-reload)
+	if s.ampModule != nil {
+		log.Debugf("triggering amp module config update")
+		if err := s.ampModule.OnConfigUpdated(cfg); err != nil {
+			log.Errorf("failed to update Amp module config: %v", err)
+		}
+	} else {
+		log.Warnf("amp module is nil, skipping config update")
+	}
+
 	// Count client sources from configuration and auth directory
 	authFiles := util.CountAuthFiles(cfg.AuthDir)
 	geminiAPIKeyCount := len(cfg.GeminiKey)
 	claudeAPIKeyCount := len(cfg.ClaudeKey)
 	codexAPIKeyCount := len(cfg.CodexKey)
+	vertexAICompatCount := len(cfg.VertexCompatAPIKey)
 	openAICompatCount := 0
 	for i := range cfg.OpenAICompatibility {
 		entry := cfg.OpenAICompatibility[i]
-		if len(entry.APIKeyEntries) > 0 {
-			openAICompatCount += len(entry.APIKeyEntries)
-			continue
-		}
-		openAICompatCount += len(entry.APIKeys)
+		openAICompatCount += len(entry.APIKeyEntries)
 	}
 
-	total := authFiles + geminiAPIKeyCount + claudeAPIKeyCount + codexAPIKeyCount + openAICompatCount
-	fmt.Printf("server clients and configuration updated: %d clients (%d auth files + %d Gemini API keys + %d Claude API keys + %d Codex keys + %d OpenAI-compat)\n",
+	total := authFiles + geminiAPIKeyCount + claudeAPIKeyCount + codexAPIKeyCount + vertexAICompatCount + openAICompatCount
+	fmt.Printf("server clients and configuration updated: %d clients (%d auth files + %d Gemini API keys + %d Claude API keys + %d Codex keys + %d Vertex-compat + %d OpenAI-compat)\n",
 		total,
 		authFiles,
 		geminiAPIKeyCount,
 		claudeAPIKeyCount,
 		codexAPIKeyCount,
+		vertexAICompatCount,
 		openAICompatCount,
 	)
 }

internal/api/server_test.go

@@ -0,0 +1,111 @@
+package api
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	gin "github.com/gin-gonic/gin"
+	proxyconfig "github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	sdkaccess "github.com/router-for-me/CLIProxyAPI/v6/sdk/access"
+	"github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/auth"
+	sdkconfig "github.com/router-for-me/CLIProxyAPI/v6/sdk/config"
+)
+
+func newTestServer(t *testing.T) *Server {
+	t.Helper()
+
+	gin.SetMode(gin.TestMode)
+
+	tmpDir := t.TempDir()
+	authDir := filepath.Join(tmpDir, "auth")
+	if err := os.MkdirAll(authDir, 0o700); err != nil {
+		t.Fatalf("failed to create auth dir: %v", err)
+	}
+
+	cfg := &proxyconfig.Config{
+		SDKConfig: sdkconfig.SDKConfig{
+			APIKeys: []string{"test-key"},
+		},
+		Port:                   0,
+		AuthDir:                authDir,
+		Debug:                  true,
+		LoggingToFile:          false,
+		UsageStatisticsEnabled: false,
+	}
+
+	authManager := auth.NewManager(nil, nil, nil)
+	accessManager := sdkaccess.NewManager()
+
+	configPath := filepath.Join(tmpDir, "config.yaml")
+	return NewServer(cfg, authManager, accessManager, configPath)
+}
+
+func TestAmpProviderModelRoutes(t *testing.T) {
+	testCases := []struct {
+		name         string
+		path         string
+		wantStatus   int
+		wantContains string
+	}{
+		{
+			name:         "openai root models",
+			path:         "/api/provider/openai/models",
+			wantStatus:   http.StatusOK,
+			wantContains: `"object":"list"`,
+		},
+		{
+			name:         "groq root models",
+			path:         "/api/provider/groq/models",
+			wantStatus:   http.StatusOK,
+			wantContains: `"object":"list"`,
+		},
+		{
+			name:         "openai models",
+			path:         "/api/provider/openai/v1/models",
+			wantStatus:   http.StatusOK,
+			wantContains: `"object":"list"`,
+		},
+		{
+			name:         "anthropic models",
+			path:         "/api/provider/anthropic/v1/models",
+			wantStatus:   http.StatusOK,
+			wantContains: `"data"`,
+		},
+		{
+			name:         "google models v1",
+			path:         "/api/provider/google/v1/models",
+			wantStatus:   http.StatusOK,
+			wantContains: `"models"`,
+		},
+		{
+			name:         "google models v1beta",
+			path:         "/api/provider/google/v1beta/models",
+			wantStatus:   http.StatusOK,
+			wantContains: `"models"`,
+		},
+	}
+
+	for _, tc := range testCases {
+		tc := tc
+		t.Run(tc.name, func(t *testing.T) {
+			server := newTestServer(t)
+
+			req := httptest.NewRequest(http.MethodGet, tc.path, nil)
+			req.Header.Set("Authorization", "Bearer test-key")
+
+			rr := httptest.NewRecorder()
+			server.engine.ServeHTTP(rr, req)
+
+			if rr.Code != tc.wantStatus {
+				t.Fatalf("unexpected status code for %s: got %d want %d; body=%s", tc.path, rr.Code, tc.wantStatus, rr.Body.String())
+			}
+			if body := rr.Body.String(); !strings.Contains(body, tc.wantContains) {
+				t.Fatalf("response body for %s missing %q: %s", tc.path, tc.wantContains, body)
+			}
+		})
+	}
+}

internal/auth/gemini/gemini_auth.go

@@ -76,7 +76,8 @@ func (g *GeminiAuth) GetAuthenticatedClient(ctx context.Context, ts *GeminiToken
 			auth := &proxy.Auth{User: username, Password: password}
 			dialer, errSOCKS5 := proxy.SOCKS5("tcp", proxyURL.Host, auth, proxy.Direct)
 			if errSOCKS5 != nil {
-				log.Fatalf("create SOCKS5 dialer failed: %v", errSOCKS5)
+				log.Errorf("create SOCKS5 dialer failed: %v", errSOCKS5)
+				return nil, fmt.Errorf("create SOCKS5 dialer failed: %w", errSOCKS5)
 			}
 			transport = &http.Transport{
 				DialContext: func(ctx context.Context, network, addr string) (net.Conn, error) {
@@ -238,7 +239,11 @@ func (g *GeminiAuth) getTokenFromWeb(ctx context.Context, config *oauth2.Config,
 	// Start the server in a goroutine.
 	go func() {
 		if err := server.ListenAndServe(); !errors.Is(err, http.ErrServerClosed) {
-			log.Fatalf("ListenAndServe(): %v", err)
+			log.Errorf("ListenAndServe(): %v", err)
+			select {
+			case errChan <- err:
+			default:
+			}
 		}
 	}()
 

internal/auth/iflow/cookie_helpers.go

@@ -0,0 +1,38 @@
+package iflow
+
+import (
+	"fmt"
+	"strings"
+)
+
+// NormalizeCookie normalizes raw cookie strings for iFlow authentication flows.
+func NormalizeCookie(raw string) (string, error) {
+	trimmed := strings.TrimSpace(raw)
+	if trimmed == "" {
+		return "", fmt.Errorf("cookie cannot be empty")
+	}
+
+	combined := strings.Join(strings.Fields(trimmed), " ")
+	if !strings.HasSuffix(combined, ";") {
+		combined += ";"
+	}
+	if !strings.Contains(combined, "BXAuth=") {
+		return "", fmt.Errorf("cookie missing BXAuth field")
+	}
+	return combined, nil
+}
+
+// SanitizeIFlowFileName normalizes user identifiers for safe filename usage.
+func SanitizeIFlowFileName(raw string) string {
+	if raw == "" {
+		return ""
+	}
+	cleanEmail := strings.ReplaceAll(raw, "*", "x")
+	var result strings.Builder
+	for _, r := range cleanEmail {
+		if (r >= 'a' && r <= 'z') || (r >= 'A' && r <= 'Z') || (r >= '0' && r <= '9') || r == '_' || r == '@' || r == '.' || r == '-' {
+			result.WriteRune(r)
+		}
+	}
+	return strings.TrimSpace(result.String())
+}

internal/auth/iflow/iflow_auth.go

@@ -1,6 +1,7 @@
 package iflow
 
 import (
+	"compress/gzip"
 	"context"
 	"encoding/base64"
 	"encoding/json"
@@ -23,6 +24,9 @@ const (
 	iFlowUserInfoEndpoint       = "https://iflow.cn/api/oauth/getUserInfo"
 	iFlowSuccessRedirectURL     = "https://iflow.cn/oauth/success"
 
+	// Cookie authentication endpoints
+	iFlowAPIKeyEndpoint = "https://platform.iflow.cn/api/openapi/apikey"
+
 	// Client credentials provided by iFlow for the Code Assist integration.
 	iFlowOAuthClientID     = "10009311001"
 	iFlowOAuthClientSecret = "4Z3YjXycVsQvyGF1etiNlIBB4RsqSDtW"
@@ -261,6 +265,7 @@ type IFlowTokenData struct {
 	Expire       string
 	APIKey       string
 	Email        string
+	Cookie       string
 }
 
 // userInfoResponse represents the structure returned by the user info endpoint.
@@ -274,3 +279,238 @@ type userInfoData struct {
 	Email  string `json:"email"`
 	Phone  string `json:"phone"`
 }
+
+// iFlowAPIKeyResponse represents the response from the API key endpoint
+type iFlowAPIKeyResponse struct {
+	Success bool         `json:"success"`
+	Code    string       `json:"code"`
+	Message string       `json:"message"`
+	Data    iFlowKeyData `json:"data"`
+	Extra   interface{}  `json:"extra"`
+}
+
+// iFlowKeyData contains the API key information
+type iFlowKeyData struct {
+	HasExpired bool   `json:"hasExpired"`
+	ExpireTime string `json:"expireTime"`
+	Name       string `json:"name"`
+	APIKey     string `json:"apiKey"`
+	APIKeyMask string `json:"apiKeyMask"`
+}
+
+// iFlowRefreshRequest represents the request body for refreshing API key
+type iFlowRefreshRequest struct {
+	Name string `json:"name"`
+}
+
+// AuthenticateWithCookie performs authentication using browser cookies
+func (ia *IFlowAuth) AuthenticateWithCookie(ctx context.Context, cookie string) (*IFlowTokenData, error) {
+	if strings.TrimSpace(cookie) == "" {
+		return nil, fmt.Errorf("iflow cookie authentication: cookie is empty")
+	}
+
+	// First, get initial API key information using GET request to obtain the name
+	keyInfo, err := ia.fetchAPIKeyInfo(ctx, cookie)
+	if err != nil {
+		return nil, fmt.Errorf("iflow cookie authentication: fetch initial API key info failed: %w", err)
+	}
+
+	// Refresh the API key using POST request
+	refreshedKeyInfo, err := ia.RefreshAPIKey(ctx, cookie, keyInfo.Name)
+	if err != nil {
+		return nil, fmt.Errorf("iflow cookie authentication: refresh API key failed: %w", err)
+	}
+
+	// Convert to token data format using refreshed key
+	data := &IFlowTokenData{
+		APIKey: refreshedKeyInfo.APIKey,
+		Expire: refreshedKeyInfo.ExpireTime,
+		Email:  refreshedKeyInfo.Name,
+		Cookie: cookie,
+	}
+
+	return data, nil
+}
+
+// fetchAPIKeyInfo retrieves API key information using GET request with cookie
+func (ia *IFlowAuth) fetchAPIKeyInfo(ctx context.Context, cookie string) (*iFlowKeyData, error) {
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, iFlowAPIKeyEndpoint, nil)
+	if err != nil {
+		return nil, fmt.Errorf("iflow cookie: create GET request failed: %w", err)
+	}
+
+	// Set cookie and other headers to mimic browser
+	req.Header.Set("Cookie", cookie)
+	req.Header.Set("Accept", "application/json, text/plain, */*")
+	req.Header.Set("User-Agent", "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36")
+	req.Header.Set("Accept-Language", "zh-CN,zh;q=0.9,en;q=0.8")
+	req.Header.Set("Accept-Encoding", "gzip, deflate, br")
+	req.Header.Set("Connection", "keep-alive")
+	req.Header.Set("Sec-Fetch-Dest", "empty")
+	req.Header.Set("Sec-Fetch-Mode", "cors")
+	req.Header.Set("Sec-Fetch-Site", "same-origin")
+
+	resp, err := ia.httpClient.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("iflow cookie: GET request failed: %w", err)
+	}
+	defer func() { _ = resp.Body.Close() }()
+
+	// Handle gzip compression
+	var reader io.Reader = resp.Body
+	if resp.Header.Get("Content-Encoding") == "gzip" {
+		gzipReader, err := gzip.NewReader(resp.Body)
+		if err != nil {
+			return nil, fmt.Errorf("iflow cookie: create gzip reader failed: %w", err)
+		}
+		defer func() { _ = gzipReader.Close() }()
+		reader = gzipReader
+	}
+
+	body, err := io.ReadAll(reader)
+	if err != nil {
+		return nil, fmt.Errorf("iflow cookie: read GET response failed: %w", err)
+	}
+
+	if resp.StatusCode != http.StatusOK {
+		log.Debugf("iflow cookie GET request failed: status=%d body=%s", resp.StatusCode, string(body))
+		return nil, fmt.Errorf("iflow cookie: GET request failed with status %d: %s", resp.StatusCode, strings.TrimSpace(string(body)))
+	}
+
+	var keyResp iFlowAPIKeyResponse
+	if err = json.Unmarshal(body, &keyResp); err != nil {
+		return nil, fmt.Errorf("iflow cookie: decode GET response failed: %w", err)
+	}
+
+	if !keyResp.Success {
+		return nil, fmt.Errorf("iflow cookie: GET request not successful: %s", keyResp.Message)
+	}
+
+	// Handle initial response where apiKey field might be apiKeyMask
+	if keyResp.Data.APIKey == "" && keyResp.Data.APIKeyMask != "" {
+		keyResp.Data.APIKey = keyResp.Data.APIKeyMask
+	}
+
+	return &keyResp.Data, nil
+}
+
+// RefreshAPIKey refreshes the API key using POST request
+func (ia *IFlowAuth) RefreshAPIKey(ctx context.Context, cookie, name string) (*iFlowKeyData, error) {
+	if strings.TrimSpace(cookie) == "" {
+		return nil, fmt.Errorf("iflow cookie refresh: cookie is empty")
+	}
+	if strings.TrimSpace(name) == "" {
+		return nil, fmt.Errorf("iflow cookie refresh: name is empty")
+	}
+
+	// Prepare request body
+	refreshReq := iFlowRefreshRequest{
+		Name: name,
+	}
+
+	bodyBytes, err := json.Marshal(refreshReq)
+	if err != nil {
+		return nil, fmt.Errorf("iflow cookie refresh: marshal request failed: %w", err)
+	}
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, iFlowAPIKeyEndpoint, strings.NewReader(string(bodyBytes)))
+	if err != nil {
+		return nil, fmt.Errorf("iflow cookie refresh: create POST request failed: %w", err)
+	}
+
+	// Set cookie and other headers to mimic browser
+	req.Header.Set("Cookie", cookie)
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Accept", "application/json, text/plain, */*")
+	req.Header.Set("User-Agent", "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36")
+	req.Header.Set("Accept-Language", "zh-CN,zh;q=0.9,en;q=0.8")
+	req.Header.Set("Accept-Encoding", "gzip, deflate, br")
+	req.Header.Set("Connection", "keep-alive")
+	req.Header.Set("Origin", "https://platform.iflow.cn")
+	req.Header.Set("Referer", "https://platform.iflow.cn/")
+
+	resp, err := ia.httpClient.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("iflow cookie refresh: POST request failed: %w", err)
+	}
+	defer func() { _ = resp.Body.Close() }()
+
+	// Handle gzip compression
+	var reader io.Reader = resp.Body
+	if resp.Header.Get("Content-Encoding") == "gzip" {
+		gzipReader, err := gzip.NewReader(resp.Body)
+		if err != nil {
+			return nil, fmt.Errorf("iflow cookie refresh: create gzip reader failed: %w", err)
+		}
+		defer func() { _ = gzipReader.Close() }()
+		reader = gzipReader
+	}
+
+	body, err := io.ReadAll(reader)
+	if err != nil {
+		return nil, fmt.Errorf("iflow cookie refresh: read POST response failed: %w", err)
+	}
+
+	if resp.StatusCode != http.StatusOK {
+		log.Debugf("iflow cookie POST request failed: status=%d body=%s", resp.StatusCode, string(body))
+		return nil, fmt.Errorf("iflow cookie refresh: POST request failed with status %d: %s", resp.StatusCode, strings.TrimSpace(string(body)))
+	}
+
+	var keyResp iFlowAPIKeyResponse
+	if err = json.Unmarshal(body, &keyResp); err != nil {
+		return nil, fmt.Errorf("iflow cookie refresh: decode POST response failed: %w", err)
+	}
+
+	if !keyResp.Success {
+		return nil, fmt.Errorf("iflow cookie refresh: POST request not successful: %s", keyResp.Message)
+	}
+
+	return &keyResp.Data, nil
+}
+
+// ShouldRefreshAPIKey checks if the API key needs to be refreshed (within 2 days of expiry)
+func ShouldRefreshAPIKey(expireTime string) (bool, time.Duration, error) {
+	if strings.TrimSpace(expireTime) == "" {
+		return false, 0, fmt.Errorf("iflow cookie: expire time is empty")
+	}
+
+	expire, err := time.Parse("2006-01-02 15:04", expireTime)
+	if err != nil {
+		return false, 0, fmt.Errorf("iflow cookie: parse expire time failed: %w", err)
+	}
+
+	now := time.Now()
+	twoDaysFromNow := now.Add(48 * time.Hour)
+
+	needsRefresh := expire.Before(twoDaysFromNow)
+	timeUntilExpiry := expire.Sub(now)
+
+	return needsRefresh, timeUntilExpiry, nil
+}
+
+// CreateCookieTokenStorage converts cookie-based token data into persistence storage
+func (ia *IFlowAuth) CreateCookieTokenStorage(data *IFlowTokenData) *IFlowTokenStorage {
+	if data == nil {
+		return nil
+	}
+
+	return &IFlowTokenStorage{
+		APIKey:      data.APIKey,
+		Email:       data.Email,
+		Expire:      data.Expire,
+		Cookie:      data.Cookie,
+		LastRefresh: time.Now().Format(time.RFC3339),
+		Type:        "iflow",
+	}
+}
+
+// UpdateCookieTokenStorage updates the persisted token storage with refreshed API key data
+func (ia *IFlowAuth) UpdateCookieTokenStorage(storage *IFlowTokenStorage, keyData *iFlowKeyData) {
+	if storage == nil || keyData == nil {
+		return
+	}
+
+	storage.APIKey = keyData.APIKey
+	storage.Expire = keyData.ExpireTime
+	storage.LastRefresh = time.Now().Format(time.RFC3339)
+}

internal/auth/iflow/iflow_token.go

@@ -19,6 +19,7 @@ type IFlowTokenStorage struct {
 	Email        string `json:"email"`
 	TokenType    string `json:"token_type"`
 	Scope        string `json:"scope"`
+	Cookie       string `json:"cookie"`
 	Type         string `json:"type"`
 }
 

internal/cmd/antigravity_login.go

@@ -0,0 +1,38 @@
+package cmd
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	sdkAuth "github.com/router-for-me/CLIProxyAPI/v6/sdk/auth"
+	log "github.com/sirupsen/logrus"
+)
+
+// DoAntigravityLogin triggers the OAuth flow for the antigravity provider and saves tokens.
+func DoAntigravityLogin(cfg *config.Config, options *LoginOptions) {
+	if options == nil {
+		options = &LoginOptions{}
+	}
+
+	manager := newAuthManager()
+	authOpts := &sdkAuth.LoginOptions{
+		NoBrowser: options.NoBrowser,
+		Metadata:  map[string]string{},
+		Prompt:    options.Prompt,
+	}
+
+	record, savedPath, err := manager.Login(context.Background(), "antigravity", cfg, authOpts)
+	if err != nil {
+		log.Errorf("Antigravity authentication failed: %v", err)
+		return
+	}
+
+	if savedPath != "" {
+		fmt.Printf("Authentication saved to %s\n", savedPath)
+	}
+	if record != nil && record.Label != "" {
+		fmt.Printf("Authenticated as %s\n", record.Label)
+	}
+	fmt.Println("Antigravity authentication successful!")
+}

internal/cmd/auth_manager.go

@@ -18,6 +18,7 @@ func newAuthManager() *sdkAuth.Manager {
 		sdkAuth.NewClaudeAuthenticator(),
 		sdkAuth.NewQwenAuthenticator(),
 		sdkAuth.NewIFlowAuthenticator(),
+		sdkAuth.NewAntigravityAuthenticator(),
 	)
 	return manager
 }

internal/cmd/iflow_cookie.go

@@ -0,0 +1,86 @@
+package cmd
+
+import (
+	"bufio"
+	"context"
+	"fmt"
+	"os"
+	"strings"
+
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/auth/iflow"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+)
+
+// DoIFlowCookieAuth performs the iFlow cookie-based authentication.
+func DoIFlowCookieAuth(cfg *config.Config, options *LoginOptions) {
+	if options == nil {
+		options = &LoginOptions{}
+	}
+
+	promptFn := options.Prompt
+	if promptFn == nil {
+		reader := bufio.NewReader(os.Stdin)
+		promptFn = func(prompt string) (string, error) {
+			fmt.Print(prompt)
+			value, err := reader.ReadString('\n')
+			if err != nil {
+				return "", err
+			}
+			return strings.TrimSpace(value), nil
+		}
+	}
+
+	// Prompt user for cookie
+	cookie, err := promptForCookie(promptFn)
+	if err != nil {
+		fmt.Printf("Failed to get cookie: %v\n", err)
+		return
+	}
+
+	// Authenticate with cookie
+	auth := iflow.NewIFlowAuth(cfg)
+	ctx := context.Background()
+
+	tokenData, err := auth.AuthenticateWithCookie(ctx, cookie)
+	if err != nil {
+		fmt.Printf("iFlow cookie authentication failed: %v\n", err)
+		return
+	}
+
+	// Create token storage
+	tokenStorage := auth.CreateCookieTokenStorage(tokenData)
+
+	// Get auth file path using email in filename
+	authFilePath := getAuthFilePath(cfg, "iflow", tokenData.Email)
+
+	// Save token to file
+	if err := tokenStorage.SaveTokenToFile(authFilePath); err != nil {
+		fmt.Printf("Failed to save authentication: %v\n", err)
+		return
+	}
+
+	fmt.Printf("Authentication successful! API key: %s\n", tokenData.APIKey)
+	fmt.Printf("Expires at: %s\n", tokenData.Expire)
+	fmt.Printf("Authentication saved to: %s\n", authFilePath)
+}
+
+// promptForCookie prompts the user to enter their iFlow cookie
+func promptForCookie(promptFn func(string) (string, error)) (string, error) {
+	line, err := promptFn("Enter iFlow Cookie (from browser cookies): ")
+	if err != nil {
+		return "", fmt.Errorf("failed to read cookie: %w", err)
+	}
+
+	cookie, err := iflow.NormalizeCookie(line)
+	if err != nil {
+		return "", err
+	}
+
+	return cookie, nil
+}
+
+// getAuthFilePath returns the auth file path for the given provider and email
+func getAuthFilePath(cfg *config.Config, provider, email string) string {
+	fileName := iflow.SanitizeIFlowFileName(email)
+	return fmt.Sprintf("%s/%s-%s.json", cfg.AuthDir, provider, fileName)
+}

internal/cmd/login.go

@@ -65,20 +65,20 @@ func DoLogin(cfg *config.Config, projectID string, options *LoginOptions) {
 	authenticator := sdkAuth.NewGeminiAuthenticator()
 	record, errLogin := authenticator.Login(ctx, cfg, loginOpts)
 	if errLogin != nil {
-		log.Fatalf("Gemini authentication failed: %v", errLogin)
+		log.Errorf("Gemini authentication failed: %v", errLogin)
 		return
 	}
 
 	storage, okStorage := record.Storage.(*gemini.GeminiTokenStorage)
 	if !okStorage || storage == nil {
-		log.Fatal("Gemini authentication failed: unsupported token storage")
+		log.Error("Gemini authentication failed: unsupported token storage")
 		return
 	}
 
 	geminiAuth := gemini.NewGeminiAuth()
 	httpClient, errClient := geminiAuth.GetAuthenticatedClient(ctx, storage, cfg, options.NoBrowser)
 	if errClient != nil {
-		log.Fatalf("Gemini authentication failed: %v", errClient)
+		log.Errorf("Gemini authentication failed: %v", errClient)
 		return
 	}
 
@@ -86,7 +86,7 @@ func DoLogin(cfg *config.Config, projectID string, options *LoginOptions) {
 
 	projects, errProjects := fetchGCPProjects(ctx, httpClient)
 	if errProjects != nil {
-		log.Fatalf("Failed to get project list: %v", errProjects)
+		log.Errorf("Failed to get project list: %v", errProjects)
 		return
 	}
 
@@ -98,11 +98,11 @@ func DoLogin(cfg *config.Config, projectID string, options *LoginOptions) {
 	selectedProjectID := promptForProjectSelection(projects, strings.TrimSpace(projectID), promptFn)
 	projectSelections, errSelection := resolveProjectSelections(selectedProjectID, projects)
 	if errSelection != nil {
-		log.Fatalf("Invalid project selection: %v", errSelection)
+		log.Errorf("Invalid project selection: %v", errSelection)
 		return
 	}
 	if len(projectSelections) == 0 {
-		log.Fatal("No project selected; aborting login.")
+		log.Error("No project selected; aborting login.")
 		return
 	}
 
@@ -116,7 +116,7 @@ func DoLogin(cfg *config.Config, projectID string, options *LoginOptions) {
 				showProjectSelectionHelp(storage.Email, projects)
 				return
 			}
-			log.Fatalf("Failed to complete user setup: %v", errSetup)
+			log.Errorf("Failed to complete user setup: %v", errSetup)
 			return
 		}
 		finalID := strings.TrimSpace(storage.ProjectID)
@@ -133,11 +133,11 @@ func DoLogin(cfg *config.Config, projectID string, options *LoginOptions) {
 		for _, pid := range activatedProjects {
 			isChecked, errCheck := checkCloudAPIIsEnabled(ctx, httpClient, pid)
 			if errCheck != nil {
-				log.Fatalf("Failed to check if Cloud AI API is enabled for %s: %v", pid, errCheck)
+				log.Errorf("Failed to check if Cloud AI API is enabled for %s: %v", pid, errCheck)
 				return
 			}
 			if !isChecked {
-				log.Fatalf("Failed to check if Cloud AI API is enabled for project %s. If you encounter an error message, please create an issue.", pid)
+				log.Errorf("Failed to check if Cloud AI API is enabled for project %s. If you encounter an error message, please create an issue.", pid)
 				return
 			}
 		}
@@ -153,7 +153,7 @@ func DoLogin(cfg *config.Config, projectID string, options *LoginOptions) {
 
 	savedPath, errSave := store.Save(ctx, record)
 	if errSave != nil {
-		log.Fatalf("Failed to save token to file: %v", errSave)
+		log.Errorf("Failed to save token to file: %v", errSave)
 		return
 	}
 
@@ -555,6 +555,7 @@ func checkCloudAPIIsEnabled(ctx context.Context, httpClient *http.Client, projec
 				continue
 			}
 		}
+		_ = resp.Body.Close()
 		return false, fmt.Errorf("project activation required: %s", errMessage)
 	}
 	return true, nil

internal/cmd/run.go

@@ -45,12 +45,13 @@ func StartService(cfg *config.Config, configPath string, localPassword string) {
 
 	service, err := builder.Build()
 	if err != nil {
-		log.Fatalf("failed to build proxy service: %v", err)
+		log.Errorf("failed to build proxy service: %v", err)
+		return
 	}
 
 	err = service.Run(runCtx)
 	if err != nil && !errors.Is(err, context.Canceled) {
-		log.Fatalf("proxy service exited with error: %v", err)
+		log.Errorf("proxy service exited with error: %v", err)
 	}
 }
 

internal/cmd/vertex_import.go

@@ -29,30 +29,30 @@ func DoVertexImport(cfg *config.Config, keyPath string) {
 	}
 	rawPath := strings.TrimSpace(keyPath)
 	if rawPath == "" {
-		log.Fatalf("vertex-import: missing service account key path")
+		log.Errorf("vertex-import: missing service account key path")
 		return
 	}
 	data, errRead := os.ReadFile(rawPath)
 	if errRead != nil {
-		log.Fatalf("vertex-import: read file failed: %v", errRead)
+		log.Errorf("vertex-import: read file failed: %v", errRead)
 		return
 	}
 	var sa map[string]any
 	if errUnmarshal := json.Unmarshal(data, &sa); errUnmarshal != nil {
-		log.Fatalf("vertex-import: invalid service account json: %v", errUnmarshal)
+		log.Errorf("vertex-import: invalid service account json: %v", errUnmarshal)
 		return
 	}
 	// Validate and normalize private_key before saving
 	normalizedSA, errFix := vertex.NormalizeServiceAccountMap(sa)
 	if errFix != nil {
-		log.Fatalf("vertex-import: %v", errFix)
+		log.Errorf("vertex-import: %v", errFix)
 		return
 	}
 	sa = normalizedSA
 	email, _ := sa["client_email"].(string)
 	projectID, _ := sa["project_id"].(string)
 	if strings.TrimSpace(projectID) == "" {
-		log.Fatalf("vertex-import: project_id missing in service account json")
+		log.Errorf("vertex-import: project_id missing in service account json")
 		return
 	}
 	if strings.TrimSpace(email) == "" {
@@ -92,7 +92,7 @@ func DoVertexImport(cfg *config.Config, keyPath string) {
 	}
 	path, errSave := store.Save(context.Background(), record)
 	if errSave != nil {
-		log.Fatalf("vertex-import: save credential failed: %v", errSave)
+		log.Errorf("vertex-import: save credential failed: %v", errSave)
 		return
 	}
 	fmt.Printf("Vertex credentials imported: %s\n", path)

internal/config/config.go

@@ -20,9 +20,18 @@ import (
 // Config represents the application's configuration, loaded from a YAML file.
 type Config struct {
 	config.SDKConfig `yaml:",inline"`
+	// Host is the network host/interface on which the API server will bind.
+	// Default is empty ("") to bind all interfaces (IPv4 + IPv6). Use "127.0.0.1" or "localhost" for local-only access.
+	Host string `yaml:"host" json:"-"`
 	// Port is the network port on which the API server will listen.
 	Port int `yaml:"port" json:"-"`
 
+	// TLS config controls HTTPS server settings.
+	TLS TLSConfig `yaml:"tls" json:"tls"`
+
+	// RemoteManagement nests management-related options under 'remote-management'.
+	RemoteManagement RemoteManagement `yaml:"remote-management" json:"-"`
+
 	// AuthDir is the directory where authentication token files are stored.
 	AuthDir string `yaml:"auth-dir" json:"-"`
 
@@ -38,35 +47,53 @@ type Config struct {
 	// DisableCooling disables quota cooldown scheduling when true.
 	DisableCooling bool `yaml:"disable-cooling" json:"disable-cooling"`
 
+	// RequestRetry defines the retry times when the request failed.
+	RequestRetry int `yaml:"request-retry" json:"request-retry"`
+	// MaxRetryInterval defines the maximum wait time in seconds before retrying a cooled-down credential.
+	MaxRetryInterval int `yaml:"max-retry-interval" json:"max-retry-interval"`
+
 	// QuotaExceeded defines the behavior when a quota is exceeded.
 	QuotaExceeded QuotaExceeded `yaml:"quota-exceeded" json:"quota-exceeded"`
 
 	// WebsocketAuth enables or disables authentication for the WebSocket API.
 	WebsocketAuth bool `yaml:"ws-auth" json:"ws-auth"`
 
-	// GlAPIKey exposes the legacy generative language API key list for backward compatibility.
-	GlAPIKey []string `yaml:"generative-language-api-key" json:"generative-language-api-key"`
-
 	// GeminiKey defines Gemini API key configurations with optional routing overrides.
 	GeminiKey []GeminiKey `yaml:"gemini-api-key" json:"gemini-api-key"`
 
-	// RequestRetry defines the retry times when the request failed.
-	RequestRetry int `yaml:"request-retry" json:"request-retry"`
-
-	// ClaudeKey defines a list of Claude API key configurations as specified in the YAML configuration file.
-	ClaudeKey []ClaudeKey `yaml:"claude-api-key" json:"claude-api-key"`
-
 	// Codex defines a list of Codex API key configurations as specified in the YAML configuration file.
 	CodexKey []CodexKey `yaml:"codex-api-key" json:"codex-api-key"`
 
+	// ClaudeKey defines a list of Claude API key configurations as specified in the YAML configuration file.
+	ClaudeKey []ClaudeKey `yaml:"claude-api-key" json:"claude-api-key"`
+
 	// OpenAICompatibility defines OpenAI API compatibility configurations for external providers.
 	OpenAICompatibility []OpenAICompatibility `yaml:"openai-compatibility" json:"openai-compatibility"`
 
-	// RemoteManagement nests management-related options under 'remote-management'.
-	RemoteManagement RemoteManagement `yaml:"remote-management" json:"-"`
+	// VertexCompatAPIKey defines Vertex AI-compatible API key configurations for third-party providers.
+	// Used for services that use Vertex AI-style paths but with simple API key authentication.
+	VertexCompatAPIKey []VertexCompatKey `yaml:"vertex-api-key" json:"vertex-api-key"`
+
+	// AmpCode contains Amp CLI upstream configuration, management restrictions, and model mappings.
+	AmpCode AmpCode `yaml:"ampcode" json:"ampcode"`
+
+	// OAuthExcludedModels defines per-provider global model exclusions applied to OAuth/file-backed auth entries.
+	OAuthExcludedModels map[string][]string `yaml:"oauth-excluded-models,omitempty" json:"oauth-excluded-models,omitempty"`
 
 	// Payload defines default and override rules for provider payload parameters.
 	Payload PayloadConfig `yaml:"payload" json:"payload"`
+
+	legacyMigrationPending bool `yaml:"-" json:"-"`
+}
+
+// TLSConfig holds HTTPS server settings.
+type TLSConfig struct {
+	// Enable toggles HTTPS server mode.
+	Enable bool `yaml:"enable" json:"enable"`
+	// Cert is the path to the TLS certificate file.
+	Cert string `yaml:"cert" json:"cert"`
+	// Key is the path to the TLS private key file.
+	Key string `yaml:"key" json:"key"`
 }
 
 // RemoteManagement holds management API configuration under 'remote-management'.
@@ -89,6 +116,42 @@ type QuotaExceeded struct {
 	SwitchPreviewModel bool `yaml:"switch-preview-model" json:"switch-preview-model"`
 }
 
+// AmpModelMapping defines a model name mapping for Amp CLI requests.
+// When Amp requests a model that isn't available locally, this mapping
+// allows routing to an alternative model that IS available.
+type AmpModelMapping struct {
+	// From is the model name that Amp CLI requests (e.g., "claude-opus-4.5").
+	From string `yaml:"from" json:"from"`
+
+	// To is the target model name to route to (e.g., "claude-sonnet-4").
+	// The target model must have available providers in the registry.
+	To string `yaml:"to" json:"to"`
+}
+
+// AmpCode groups Amp CLI integration settings including upstream routing,
+// optional overrides, management route restrictions, and model fallback mappings.
+type AmpCode struct {
+	// UpstreamURL defines the upstream Amp control plane used for non-provider calls.
+	UpstreamURL string `yaml:"upstream-url" json:"upstream-url"`
+
+	// UpstreamAPIKey optionally overrides the Authorization header when proxying Amp upstream calls.
+	UpstreamAPIKey string `yaml:"upstream-api-key" json:"upstream-api-key"`
+
+	// RestrictManagementToLocalhost restricts Amp management routes (/api/user, /api/threads, etc.)
+	// to only accept connections from localhost (127.0.0.1, ::1). When true, prevents drive-by
+	// browser attacks and remote access to management endpoints. Default: true (recommended).
+	RestrictManagementToLocalhost bool `yaml:"restrict-management-to-localhost" json:"restrict-management-to-localhost"`
+
+	// ModelMappings defines model name mappings for Amp CLI requests.
+	// When Amp requests a model that isn't available locally, these mappings
+	// allow routing to an alternative model that IS available.
+	ModelMappings []AmpModelMapping `yaml:"model-mappings" json:"model-mappings"`
+
+	// ForceModelMappings when true, model mappings take precedence over local API keys.
+	// When false (default), local API keys are used first if available.
+	ForceModelMappings bool `yaml:"force-model-mappings" json:"force-model-mappings"`
+}
+
 // PayloadConfig defines default and override parameter rules applied to provider payloads.
 type PayloadConfig struct {
 	// Default defines rules that only set parameters when they are missing in the payload.
@@ -131,6 +194,9 @@ type ClaudeKey struct {
 
 	// Headers optionally adds extra HTTP headers for requests sent with this key.
 	Headers map[string]string `yaml:"headers,omitempty" json:"headers,omitempty"`
+
+	// ExcludedModels lists model IDs that should be excluded for this provider.
+	ExcludedModels []string `yaml:"excluded-models,omitempty" json:"excluded-models,omitempty"`
 }
 
 // ClaudeModel describes a mapping between an alias and the actual upstream model name.
@@ -157,6 +223,9 @@ type CodexKey struct {
 
 	// Headers optionally adds extra HTTP headers for requests sent with this key.
 	Headers map[string]string `yaml:"headers,omitempty" json:"headers,omitempty"`
+
+	// ExcludedModels lists model IDs that should be excluded for this provider.
+	ExcludedModels []string `yaml:"excluded-models,omitempty" json:"excluded-models,omitempty"`
 }
 
 // GeminiKey represents the configuration for a Gemini API key,
@@ -173,6 +242,9 @@ type GeminiKey struct {
 
 	// Headers optionally adds extra HTTP headers for requests sent with this key.
 	Headers map[string]string `yaml:"headers,omitempty" json:"headers,omitempty"`
+
+	// ExcludedModels lists model IDs that should be excluded for this provider.
+	ExcludedModels []string `yaml:"excluded-models,omitempty" json:"excluded-models,omitempty"`
 }
 
 // OpenAICompatibility represents the configuration for OpenAI API compatibility
@@ -184,10 +256,6 @@ type OpenAICompatibility struct {
 	// BaseURL is the base URL for the external OpenAI-compatible API endpoint.
 	BaseURL string `yaml:"base-url" json:"base-url"`
 
-	// APIKeys are the authentication keys for accessing the external API services.
-	// Deprecated: Use APIKeyEntries instead to support per-key proxy configuration.
-	APIKeys []string `yaml:"api-keys,omitempty" json:"api-keys,omitempty"`
-
 	// APIKeyEntries defines API keys with optional per-key proxy configuration.
 	APIKeyEntries []OpenAICompatibilityAPIKey `yaml:"api-key-entries,omitempty" json:"api-key-entries,omitempty"`
 
@@ -255,9 +323,11 @@ func LoadConfigOptional(configFile string, optional bool) (*Config, error) {
 	// Unmarshal the YAML data into the Config struct.
 	var cfg Config
 	// Set defaults before unmarshal so that absent keys keep defaults.
+	cfg.Host = "" // Default empty: binds to all interfaces (IPv4 + IPv6)
 	cfg.LoggingToFile = false
 	cfg.UsageStatisticsEnabled = false
 	cfg.DisableCooling = false
+	cfg.AmpCode.RestrictManagementToLocalhost = true // Default to secure: only localhost access
 	if err = yaml.Unmarshal(data, &cfg); err != nil {
 		if optional {
 			// In cloud deploy mode, if YAML parsing fails, return empty config instead of error.
@@ -266,6 +336,19 @@ func LoadConfigOptional(configFile string, optional bool) (*Config, error) {
 		return nil, fmt.Errorf("failed to parse config file: %w", err)
 	}
 
+	var legacy legacyConfigData
+	if errLegacy := yaml.Unmarshal(data, &legacy); errLegacy == nil {
+		if cfg.migrateLegacyGeminiKeys(legacy.LegacyGeminiKeys) {
+			cfg.legacyMigrationPending = true
+		}
+		if cfg.migrateLegacyOpenAICompatibilityKeys(legacy.OpenAICompat) {
+			cfg.legacyMigrationPending = true
+		}
+		if cfg.migrateLegacyAmpConfig(&legacy) {
+			cfg.legacyMigrationPending = true
+		}
+	}
+
 	// Hash remote management key if plaintext is detected (nested)
 	// We consider a value to be already hashed if it looks like a bcrypt hash ($2a$, $2b$, or $2y$ prefix).
 	if cfg.RemoteManagement.SecretKey != "" && !looksLikeBcrypt(cfg.RemoteManagement.SecretKey) {
@@ -286,6 +369,9 @@ func LoadConfigOptional(configFile string, optional bool) (*Config, error) {
 	// Sanitize Gemini API key configuration and migrate legacy entries.
 	cfg.SanitizeGeminiKeys()
 
+	// Sanitize Vertex-compatible API keys: drop entries without base-url
+	cfg.SanitizeVertexCompatKeys()
+
 	// Sanitize Codex keys: drop entries without base-url
 	cfg.SanitizeCodexKeys()
 
@@ -295,6 +381,21 @@ func LoadConfigOptional(configFile string, optional bool) (*Config, error) {
 	// Sanitize OpenAI compatibility providers: drop entries without base-url
 	cfg.SanitizeOpenAICompatibility()
 
+	// Normalize OAuth provider model exclusion map.
+	cfg.OAuthExcludedModels = NormalizeOAuthExcludedModels(cfg.OAuthExcludedModels)
+
+	if cfg.legacyMigrationPending {
+		fmt.Println("Detected legacy configuration keys, attempting to persist the normalized config...")
+		if !optional && configFile != "" {
+			if err := SaveConfigPreserveComments(configFile, &cfg); err != nil {
+				return nil, fmt.Errorf("failed to persist migrated legacy config: %w", err)
+			}
+			fmt.Println("Legacy configuration normalized and persisted.")
+		} else {
+			fmt.Println("Legacy configuration normalized in memory; persistence skipped.")
+		}
+	}
+
 	// Return the populated configuration struct.
 	return &cfg, nil
 }
@@ -332,6 +433,7 @@ func (cfg *Config) SanitizeCodexKeys() {
 		e := cfg.CodexKey[i]
 		e.BaseURL = strings.TrimSpace(e.BaseURL)
 		e.Headers = NormalizeHeaders(e.Headers)
+		e.ExcludedModels = NormalizeExcludedModels(e.ExcludedModels)
 		if e.BaseURL == "" {
 			continue
 		}
@@ -348,6 +450,7 @@ func (cfg *Config) SanitizeClaudeKeys() {
 	for i := range cfg.ClaudeKey {
 		entry := &cfg.ClaudeKey[i]
 		entry.Headers = NormalizeHeaders(entry.Headers)
+		entry.ExcludedModels = NormalizeExcludedModels(entry.ExcludedModels)
 	}
 }
 
@@ -368,6 +471,7 @@ func (cfg *Config) SanitizeGeminiKeys() {
 		entry.BaseURL = strings.TrimSpace(entry.BaseURL)
 		entry.ProxyURL = strings.TrimSpace(entry.ProxyURL)
 		entry.Headers = NormalizeHeaders(entry.Headers)
+		entry.ExcludedModels = NormalizeExcludedModels(entry.ExcludedModels)
 		if _, exists := seen[entry.APIKey]; exists {
 			continue
 		}
@@ -375,22 +479,6 @@ func (cfg *Config) SanitizeGeminiKeys() {
 		out = append(out, entry)
 	}
 	cfg.GeminiKey = out
-
-	if len(cfg.GlAPIKey) > 0 {
-		for _, raw := range cfg.GlAPIKey {
-			key := strings.TrimSpace(raw)
-			if key == "" {
-				continue
-			}
-			if _, exists := seen[key]; exists {
-				continue
-			}
-			cfg.GeminiKey = append(cfg.GeminiKey, GeminiKey{APIKey: key})
-			seen[key] = struct{}{}
-		}
-	}
-
-	cfg.GlAPIKey = nil
 }
 
 func syncInlineAccessProvider(cfg *Config) {
@@ -430,6 +518,55 @@ func NormalizeHeaders(headers map[string]string) map[string]string {
 	return clean
 }
 
+// NormalizeExcludedModels trims, lowercases, and deduplicates model exclusion patterns.
+// It preserves the order of first occurrences and drops empty entries.
+func NormalizeExcludedModels(models []string) []string {
+	if len(models) == 0 {
+		return nil
+	}
+	seen := make(map[string]struct{}, len(models))
+	out := make([]string, 0, len(models))
+	for _, raw := range models {
+		trimmed := strings.ToLower(strings.TrimSpace(raw))
+		if trimmed == "" {
+			continue
+		}
+		if _, exists := seen[trimmed]; exists {
+			continue
+		}
+		seen[trimmed] = struct{}{}
+		out = append(out, trimmed)
+	}
+	if len(out) == 0 {
+		return nil
+	}
+	return out
+}
+
+// NormalizeOAuthExcludedModels cleans provider -> excluded models mappings by normalizing provider keys
+// and applying model exclusion normalization to each entry.
+func NormalizeOAuthExcludedModels(entries map[string][]string) map[string][]string {
+	if len(entries) == 0 {
+		return nil
+	}
+	out := make(map[string][]string, len(entries))
+	for provider, models := range entries {
+		key := strings.ToLower(strings.TrimSpace(provider))
+		if key == "" {
+			continue
+		}
+		normalized := NormalizeExcludedModels(models)
+		if len(normalized) == 0 {
+			continue
+		}
+		out[key] = normalized
+	}
+	if len(out) == 0 {
+		return nil
+	}
+	return out
+}
+
 // hashSecret hashes the given secret using bcrypt.
 func hashSecret(secret string) (string, error) {
 	// Use default cost for simplicity.
@@ -477,8 +614,13 @@ func SaveConfigPreserveComments(configFile string, cfg *Config) error {
 		return fmt.Errorf("expected generated root mapping node")
 	}
 
-	// Remove deprecated auth block before merging to avoid persisting it again.
-	removeMapKey(original.Content[0], "auth")
+	// Remove deprecated sections before merging back the sanitized config.
+	removeLegacyAuthBlock(original.Content[0])
+	removeLegacyOpenAICompatAPIKeys(original.Content[0])
+	removeLegacyAmpKeys(original.Content[0])
+	removeLegacyGenerativeLanguageKeys(original.Content[0])
+
+	pruneMappingToGeneratedKeys(original.Content[0], generated.Content[0], "oauth-excluded-models")
 
 	// Merge generated into original in-place, preserving comments/order of existing nodes.
 	mergeMappingPreserve(original.Content[0], generated.Content[0])
@@ -677,6 +819,10 @@ func mergeNodePreserve(dst, src *yaml.Node) {
 				continue
 			}
 			mergeNodePreserve(dst.Content[i], src.Content[i])
+			if dst.Content[i] != nil && src.Content[i] != nil &&
+				dst.Content[i].Kind == yaml.MappingNode && src.Content[i].Kind == yaml.MappingNode {
+				pruneMissingMapKeys(dst.Content[i], src.Content[i])
+			}
 		}
 		// Append any extra items from src
 		for i := len(dst.Content); i < len(src.Content); i++ {
@@ -718,6 +864,7 @@ func shouldSkipEmptyCollectionOnPersist(key string, node *yaml.Node) bool {
 	switch key {
 	case "generative-language-api-key",
 		"gemini-api-key",
+		"vertex-api-key",
 		"claude-api-key",
 		"codex-api-key",
 		"openai-compatibility":
@@ -935,6 +1082,73 @@ func removeMapKey(mapNode *yaml.Node, key string) {
 	}
 }
 
+func pruneMappingToGeneratedKeys(dstRoot, srcRoot *yaml.Node, key string) {
+	if key == "" || dstRoot == nil || srcRoot == nil {
+		return
+	}
+	if dstRoot.Kind != yaml.MappingNode || srcRoot.Kind != yaml.MappingNode {
+		return
+	}
+	dstIdx := findMapKeyIndex(dstRoot, key)
+	if dstIdx < 0 || dstIdx+1 >= len(dstRoot.Content) {
+		return
+	}
+	srcIdx := findMapKeyIndex(srcRoot, key)
+	if srcIdx < 0 {
+		removeMapKey(dstRoot, key)
+		return
+	}
+	if srcIdx+1 >= len(srcRoot.Content) {
+		return
+	}
+	srcVal := srcRoot.Content[srcIdx+1]
+	dstVal := dstRoot.Content[dstIdx+1]
+	if srcVal == nil {
+		dstRoot.Content[dstIdx+1] = nil
+		return
+	}
+	if srcVal.Kind != yaml.MappingNode {
+		dstRoot.Content[dstIdx+1] = deepCopyNode(srcVal)
+		return
+	}
+	if dstVal == nil || dstVal.Kind != yaml.MappingNode {
+		dstRoot.Content[dstIdx+1] = deepCopyNode(srcVal)
+		return
+	}
+	pruneMissingMapKeys(dstVal, srcVal)
+}
+
+func pruneMissingMapKeys(dstMap, srcMap *yaml.Node) {
+	if dstMap == nil || srcMap == nil || dstMap.Kind != yaml.MappingNode || srcMap.Kind != yaml.MappingNode {
+		return
+	}
+	keep := make(map[string]struct{}, len(srcMap.Content)/2)
+	for i := 0; i+1 < len(srcMap.Content); i += 2 {
+		keyNode := srcMap.Content[i]
+		if keyNode == nil {
+			continue
+		}
+		key := strings.TrimSpace(keyNode.Value)
+		if key == "" {
+			continue
+		}
+		keep[key] = struct{}{}
+	}
+	for i := 0; i+1 < len(dstMap.Content); {
+		keyNode := dstMap.Content[i]
+		if keyNode == nil {
+			i += 2
+			continue
+		}
+		key := strings.TrimSpace(keyNode.Value)
+		if _, ok := keep[key]; !ok {
+			dstMap.Content = append(dstMap.Content[:i], dstMap.Content[i+2:]...)
+			continue
+		}
+		i += 2
+	}
+}
+
 // normalizeCollectionNodeStyles forces YAML collections to use block notation, keeping
 // lists and maps readable. Empty sequences retain flow style ([]) so empty list markers
 // remain compact.
@@ -961,3 +1175,194 @@ func normalizeCollectionNodeStyles(node *yaml.Node) {
 		// Scalars keep their existing style to preserve quoting
 	}
 }
+
+// Legacy migration helpers (move deprecated config keys into structured fields).
+type legacyConfigData struct {
+	LegacyGeminiKeys      []string                    `yaml:"generative-language-api-key"`
+	OpenAICompat          []legacyOpenAICompatibility `yaml:"openai-compatibility"`
+	AmpUpstreamURL        string                      `yaml:"amp-upstream-url"`
+	AmpUpstreamAPIKey     string                      `yaml:"amp-upstream-api-key"`
+	AmpRestrictManagement *bool                       `yaml:"amp-restrict-management-to-localhost"`
+	AmpModelMappings      []AmpModelMapping           `yaml:"amp-model-mappings"`
+}
+
+type legacyOpenAICompatibility struct {
+	Name    string   `yaml:"name"`
+	BaseURL string   `yaml:"base-url"`
+	APIKeys []string `yaml:"api-keys"`
+}
+
+func (cfg *Config) migrateLegacyGeminiKeys(legacy []string) bool {
+	if cfg == nil || len(legacy) == 0 {
+		return false
+	}
+	changed := false
+	seen := make(map[string]struct{}, len(cfg.GeminiKey))
+	for i := range cfg.GeminiKey {
+		key := strings.TrimSpace(cfg.GeminiKey[i].APIKey)
+		if key == "" {
+			continue
+		}
+		seen[key] = struct{}{}
+	}
+	for _, raw := range legacy {
+		key := strings.TrimSpace(raw)
+		if key == "" {
+			continue
+		}
+		if _, exists := seen[key]; exists {
+			continue
+		}
+		cfg.GeminiKey = append(cfg.GeminiKey, GeminiKey{APIKey: key})
+		seen[key] = struct{}{}
+		changed = true
+	}
+	return changed
+}
+
+func (cfg *Config) migrateLegacyOpenAICompatibilityKeys(legacy []legacyOpenAICompatibility) bool {
+	if cfg == nil || len(cfg.OpenAICompatibility) == 0 || len(legacy) == 0 {
+		return false
+	}
+	changed := false
+	for _, legacyEntry := range legacy {
+		if len(legacyEntry.APIKeys) == 0 {
+			continue
+		}
+		target := findOpenAICompatTarget(cfg.OpenAICompatibility, legacyEntry.Name, legacyEntry.BaseURL)
+		if target == nil {
+			continue
+		}
+		if mergeLegacyOpenAICompatAPIKeys(target, legacyEntry.APIKeys) {
+			changed = true
+		}
+	}
+	return changed
+}
+
+func mergeLegacyOpenAICompatAPIKeys(entry *OpenAICompatibility, keys []string) bool {
+	if entry == nil || len(keys) == 0 {
+		return false
+	}
+	changed := false
+	existing := make(map[string]struct{}, len(entry.APIKeyEntries))
+	for i := range entry.APIKeyEntries {
+		key := strings.TrimSpace(entry.APIKeyEntries[i].APIKey)
+		if key == "" {
+			continue
+		}
+		existing[key] = struct{}{}
+	}
+	for _, raw := range keys {
+		key := strings.TrimSpace(raw)
+		if key == "" {
+			continue
+		}
+		if _, ok := existing[key]; ok {
+			continue
+		}
+		entry.APIKeyEntries = append(entry.APIKeyEntries, OpenAICompatibilityAPIKey{APIKey: key})
+		existing[key] = struct{}{}
+		changed = true
+	}
+	return changed
+}
+
+func findOpenAICompatTarget(entries []OpenAICompatibility, legacyName, legacyBase string) *OpenAICompatibility {
+	nameKey := strings.ToLower(strings.TrimSpace(legacyName))
+	baseKey := strings.ToLower(strings.TrimSpace(legacyBase))
+	if nameKey != "" && baseKey != "" {
+		for i := range entries {
+			if strings.ToLower(strings.TrimSpace(entries[i].Name)) == nameKey &&
+				strings.ToLower(strings.TrimSpace(entries[i].BaseURL)) == baseKey {
+				return &entries[i]
+			}
+		}
+	}
+	if baseKey != "" {
+		for i := range entries {
+			if strings.ToLower(strings.TrimSpace(entries[i].BaseURL)) == baseKey {
+				return &entries[i]
+			}
+		}
+	}
+	if nameKey != "" {
+		for i := range entries {
+			if strings.ToLower(strings.TrimSpace(entries[i].Name)) == nameKey {
+				return &entries[i]
+			}
+		}
+	}
+	return nil
+}
+
+func (cfg *Config) migrateLegacyAmpConfig(legacy *legacyConfigData) bool {
+	if cfg == nil || legacy == nil {
+		return false
+	}
+	changed := false
+	if cfg.AmpCode.UpstreamURL == "" {
+		if val := strings.TrimSpace(legacy.AmpUpstreamURL); val != "" {
+			cfg.AmpCode.UpstreamURL = val
+			changed = true
+		}
+	}
+	if cfg.AmpCode.UpstreamAPIKey == "" {
+		if val := strings.TrimSpace(legacy.AmpUpstreamAPIKey); val != "" {
+			cfg.AmpCode.UpstreamAPIKey = val
+			changed = true
+		}
+	}
+	if legacy.AmpRestrictManagement != nil {
+		cfg.AmpCode.RestrictManagementToLocalhost = *legacy.AmpRestrictManagement
+		changed = true
+	}
+	if len(cfg.AmpCode.ModelMappings) == 0 && len(legacy.AmpModelMappings) > 0 {
+		cfg.AmpCode.ModelMappings = append([]AmpModelMapping(nil), legacy.AmpModelMappings...)
+		changed = true
+	}
+	return changed
+}
+
+func removeLegacyOpenAICompatAPIKeys(root *yaml.Node) {
+	if root == nil || root.Kind != yaml.MappingNode {
+		return
+	}
+	idx := findMapKeyIndex(root, "openai-compatibility")
+	if idx < 0 || idx+1 >= len(root.Content) {
+		return
+	}
+	seq := root.Content[idx+1]
+	if seq == nil || seq.Kind != yaml.SequenceNode {
+		return
+	}
+	for i := range seq.Content {
+		if seq.Content[i] != nil && seq.Content[i].Kind == yaml.MappingNode {
+			removeMapKey(seq.Content[i], "api-keys")
+		}
+	}
+}
+
+func removeLegacyAmpKeys(root *yaml.Node) {
+	if root == nil || root.Kind != yaml.MappingNode {
+		return
+	}
+	removeMapKey(root, "amp-upstream-url")
+	removeMapKey(root, "amp-upstream-api-key")
+	removeMapKey(root, "amp-restrict-management-to-localhost")
+	removeMapKey(root, "amp-model-mappings")
+}
+
+func removeLegacyGenerativeLanguageKeys(root *yaml.Node) {
+	if root == nil || root.Kind != yaml.MappingNode {
+		return
+	}
+	removeMapKey(root, "generative-language-api-key")
+}
+
+func removeLegacyAuthBlock(root *yaml.Node) {
+	if root == nil || root.Kind != yaml.MappingNode {
+		return
+	}
+	removeMapKey(root, "auth")
+}

internal/config/vertex_compat.go

@@ -0,0 +1,84 @@
+package config
+
+import "strings"
+
+// VertexCompatKey represents the configuration for Vertex AI-compatible API keys.
+// This supports third-party services that use Vertex AI-style endpoint paths
+// (/publishers/google/models/{model}:streamGenerateContent) but authenticate
+// with simple API keys instead of Google Cloud service account credentials.
+//
+// Example services: zenmux.ai and similar Vertex-compatible providers.
+type VertexCompatKey struct {
+	// APIKey is the authentication key for accessing the Vertex-compatible API.
+	// Maps to the x-goog-api-key header.
+	APIKey string `yaml:"api-key" json:"api-key"`
+
+	// BaseURL is the base URL for the Vertex-compatible API endpoint.
+	// The executor will append "/v1/publishers/google/models/{model}:action" to this.
+	// Example: "https://zenmux.ai/api" becomes "https://zenmux.ai/api/v1/publishers/google/models/..."
+	BaseURL string `yaml:"base-url,omitempty" json:"base-url,omitempty"`
+
+	// ProxyURL optionally overrides the global proxy for this API key.
+	ProxyURL string `yaml:"proxy-url,omitempty" json:"proxy-url,omitempty"`
+
+	// Headers optionally adds extra HTTP headers for requests sent with this key.
+	// Commonly used for cookies, user-agent, and other authentication headers.
+	Headers map[string]string `yaml:"headers,omitempty" json:"headers,omitempty"`
+
+	// Models defines the model configurations including aliases for routing.
+	Models []VertexCompatModel `yaml:"models,omitempty" json:"models,omitempty"`
+}
+
+// VertexCompatModel represents a model configuration for Vertex compatibility,
+// including the actual model name and its alias for API routing.
+type VertexCompatModel struct {
+	// Name is the actual model name used by the external provider.
+	Name string `yaml:"name" json:"name"`
+
+	// Alias is the model name alias that clients will use to reference this model.
+	Alias string `yaml:"alias" json:"alias"`
+}
+
+// SanitizeVertexCompatKeys deduplicates and normalizes Vertex-compatible API key credentials.
+func (cfg *Config) SanitizeVertexCompatKeys() {
+	if cfg == nil {
+		return
+	}
+
+	seen := make(map[string]struct{}, len(cfg.VertexCompatAPIKey))
+	out := cfg.VertexCompatAPIKey[:0]
+	for i := range cfg.VertexCompatAPIKey {
+		entry := cfg.VertexCompatAPIKey[i]
+		entry.APIKey = strings.TrimSpace(entry.APIKey)
+		if entry.APIKey == "" {
+			continue
+		}
+		entry.BaseURL = strings.TrimSpace(entry.BaseURL)
+		if entry.BaseURL == "" {
+			// BaseURL is required for Vertex API key entries
+			continue
+		}
+		entry.ProxyURL = strings.TrimSpace(entry.ProxyURL)
+		entry.Headers = NormalizeHeaders(entry.Headers)
+
+		// Sanitize models: remove entries without valid alias
+		sanitizedModels := make([]VertexCompatModel, 0, len(entry.Models))
+		for _, model := range entry.Models {
+			model.Alias = strings.TrimSpace(model.Alias)
+			model.Name = strings.TrimSpace(model.Name)
+			if model.Alias != "" && model.Name != "" {
+				sanitizedModels = append(sanitizedModels, model)
+			}
+		}
+		entry.Models = sanitizedModels
+
+		// Use API key + base URL as uniqueness key
+		uniqueKey := entry.APIKey + "|" + entry.BaseURL
+		if _, exists := seen[uniqueKey]; exists {
+			continue
+		}
+		seen[uniqueKey] = struct{}{}
+		out = append(out, entry)
+	}
+	cfg.VertexCompatAPIKey = out
+}

internal/constant/constant.go

@@ -21,4 +21,7 @@ const (
 
 	// OpenaiResponse represents the OpenAI response format identifier.
 	OpenaiResponse = "openai-response"
+
+	// Antigravity represents the Antigravity response format identifier.
+	Antigravity = "antigravity"
 )

internal/interfaces/client_models.go

@@ -56,12 +56,17 @@ type Content struct {
 // Part represents a distinct piece of content within a message.
 // A part can be text, inline data (like an image), a function call, or a function response.
 type Part struct {
+	Thought bool `json:"thought,omitempty"`
+
 	// Text contains plain text content.
 	Text string `json:"text,omitempty"`
 
 	// InlineData contains base64-encoded data with its MIME type (e.g., images).
 	InlineData *InlineData `json:"inlineData,omitempty"`
 
+	// ThoughtSignature is a provider-required signature that accompanies certain parts.
+	ThoughtSignature string `json:"thoughtSignature,omitempty"`
+
 	// FunctionCall represents a tool call requested by the model.
 	FunctionCall *FunctionCall `json:"functionCall,omitempty"`
 
@@ -82,6 +87,9 @@ type InlineData struct {
 // FunctionCall represents a tool call requested by the model.
 // It includes the function name and its arguments that the model wants to execute.
 type FunctionCall struct {
+	// ID is the identifier of the function to be called.
+	ID string `json:"id,omitempty"`
+
 	// Name is the identifier of the function to be called.
 	Name string `json:"name"`
 
@@ -92,6 +100,9 @@ type FunctionCall struct {
 // FunctionResponse represents the result of a tool execution.
 // This is sent back to the model after a tool call has been processed.
 type FunctionResponse struct {
+	// ID is the identifier of the function to be called.
+	ID string `json:"id,omitempty"`
+
 	// Name is the identifier of the function that was called.
 	Name string `json:"name"`
 

internal/logging/gin_logger.go

@@ -14,6 +14,8 @@ import (
 	log "github.com/sirupsen/logrus"
 )
 
+const skipGinLogKey = "__gin_skip_request_logging__"
+
 // GinLogrusLogger returns a Gin middleware handler that logs HTTP requests and responses
 // using logrus. It captures request details including method, path, status code, latency,
 // client IP, and any error messages, formatting them in a Gin-style log format.
@@ -28,6 +30,10 @@ func GinLogrusLogger() gin.HandlerFunc {
 
 		c.Next()
 
+		if shouldSkipGinRequestLogging(c) {
+			return
+		}
+
 		if raw != "" {
 			path = path + "?" + raw
 		}
@@ -77,3 +83,24 @@ func GinLogrusRecovery() gin.HandlerFunc {
 		c.AbortWithStatus(http.StatusInternalServerError)
 	})
 }
+
+// SkipGinRequestLogging marks the provided Gin context so that GinLogrusLogger
+// will skip emitting a log line for the associated request.
+func SkipGinRequestLogging(c *gin.Context) {
+	if c == nil {
+		return
+	}
+	c.Set(skipGinLogKey, true)
+}
+
+func shouldSkipGinRequestLogging(c *gin.Context) bool {
+	if c == nil {
+		return false
+	}
+	val, exists := c.Get(skipGinLogKey)
+	if !exists {
+		return false
+	}
+	flag, ok := val.(bool)
+	return ok && flag
+}

internal/logging/global_logger.go

@@ -38,7 +38,13 @@ func (m *LogFormatter) Format(entry *log.Entry) ([]byte, error) {
 
 	timestamp := entry.Time.Format("2006-01-02 15:04:05")
 	message := strings.TrimRight(entry.Message, "\r\n")
-	formatted := fmt.Sprintf("[%s] [%s] [%s:%d] %s\n", timestamp, entry.Level, filepath.Base(entry.Caller.File), entry.Caller.Line, message)
+
+	var formatted string
+	if entry.Caller != nil {
+		formatted = fmt.Sprintf("[%s] [%s] [%s:%d] %s\n", timestamp, entry.Level, filepath.Base(entry.Caller.File), entry.Caller.Line, message)
+	} else {
+		formatted = fmt.Sprintf("[%s] [%s] %s\n", timestamp, entry.Level, message)
+	}
 	buffer.WriteString(formatted)
 
 	return buffer.Bytes(), nil

internal/logging/request_logger.go

@@ -12,6 +12,7 @@ import (
 	"os"
 	"path/filepath"
 	"regexp"
+	"sort"
 	"strings"
 	"time"
 
@@ -19,6 +20,7 @@ import (
 	"github.com/klauspost/compress/zstd"
 	log "github.com/sirupsen/logrus"
 
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/buildinfo"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/interfaces"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 )
@@ -82,6 +84,26 @@ type StreamingLogWriter interface {
 	//   - error: An error if writing fails, nil otherwise
 	WriteStatus(status int, headers map[string][]string) error
 
+	// WriteAPIRequest writes the upstream API request details to the log.
+	// This should be called before WriteStatus to maintain proper log ordering.
+	//
+	// Parameters:
+	//   - apiRequest: The API request data (typically includes URL, headers, body sent upstream)
+	//
+	// Returns:
+	//   - error: An error if writing fails, nil otherwise
+	WriteAPIRequest(apiRequest []byte) error
+
+	// WriteAPIResponse writes the upstream API response details to the log.
+	// This should be called after the streaming response is complete.
+	//
+	// Parameters:
+	//   - apiResponse: The API response data
+	//
+	// Returns:
+	//   - error: An error if writing fails, nil otherwise
+	WriteAPIResponse(apiResponse []byte) error
+
 	// Close finalizes the log file and cleans up resources.
 	//
 	// Returns:
@@ -156,17 +178,30 @@ func (l *FileRequestLogger) SetEnabled(enabled bool) {
 // Returns:
 //   - error: An error if logging fails, nil otherwise
 func (l *FileRequestLogger) LogRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage) error {
-	if !l.enabled {
+	return l.logRequest(url, method, requestHeaders, body, statusCode, responseHeaders, response, apiRequest, apiResponse, apiResponseErrors, false)
+}
+
+// LogRequestWithOptions logs a request with optional forced logging behavior.
+// The force flag allows writing error logs even when regular request logging is disabled.
+func (l *FileRequestLogger) LogRequestWithOptions(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, force bool) error {
+	return l.logRequest(url, method, requestHeaders, body, statusCode, responseHeaders, response, apiRequest, apiResponse, apiResponseErrors, force)
+}
+
+func (l *FileRequestLogger) logRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, force bool) error {
+	if !l.enabled && !force {
 		return nil
 	}
 
 	// Ensure logs directory exists
-	if err := l.ensureLogsDir(); err != nil {
-		return fmt.Errorf("failed to create logs directory: %w", err)
+	if errEnsure := l.ensureLogsDir(); errEnsure != nil {
+		return fmt.Errorf("failed to create logs directory: %w", errEnsure)
 	}
 
 	// Generate filename
 	filename := l.generateFilename(url)
+	if force && !l.enabled {
+		filename = l.generateErrorFilename(url)
+	}
 	filePath := filepath.Join(l.logsDir, filename)
 
 	// Decompress response if needed
@@ -184,6 +219,12 @@ func (l *FileRequestLogger) LogRequest(url, method string, requestHeaders map[st
 		return fmt.Errorf("failed to write log file: %w", err)
 	}
 
+	if force && !l.enabled {
+		if errCleanup := l.cleanupOldErrorLogs(); errCleanup != nil {
+			log.WithError(errCleanup).Warn("failed to clean up old error logs")
+		}
+	}
+
 	return nil
 }
 
@@ -227,10 +268,11 @@ func (l *FileRequestLogger) LogStreamingRequest(url, method string, headers map[
 
 	// Create streaming writer
 	writer := &FileStreamingLogWriter{
-		file:      file,
-		chunkChan: make(chan []byte, 100), // Buffered channel for async writes
-		closeChan: make(chan struct{}),
-		errorChan: make(chan error, 1),
+		file:           file,
+		chunkChan:      make(chan []byte, 100), // Buffered channel for async writes
+		closeChan:      make(chan struct{}),
+		errorChan:      make(chan error, 1),
+		bufferedChunks: &bytes.Buffer{},
 	}
 
 	// Start async writer goroutine
@@ -239,6 +281,11 @@ func (l *FileRequestLogger) LogStreamingRequest(url, method string, headers map[
 	return writer, nil
 }
 
+// generateErrorFilename creates a filename with an error prefix to differentiate forced error logs.
+func (l *FileRequestLogger) generateErrorFilename(url string) string {
+	return fmt.Sprintf("error-%s", l.generateFilename(url))
+}
+
 // ensureLogsDir creates the logs directory if it doesn't exist.
 //
 // Returns:
@@ -312,6 +359,52 @@ func (l *FileRequestLogger) sanitizeForFilename(path string) string {
 	return sanitized
 }
 
+// cleanupOldErrorLogs keeps only the newest 10 forced error log files.
+func (l *FileRequestLogger) cleanupOldErrorLogs() error {
+	entries, errRead := os.ReadDir(l.logsDir)
+	if errRead != nil {
+		return errRead
+	}
+
+	type logFile struct {
+		name    string
+		modTime time.Time
+	}
+
+	var files []logFile
+	for _, entry := range entries {
+		if entry.IsDir() {
+			continue
+		}
+		name := entry.Name()
+		if !strings.HasPrefix(name, "error-") || !strings.HasSuffix(name, ".log") {
+			continue
+		}
+		info, errInfo := entry.Info()
+		if errInfo != nil {
+			log.WithError(errInfo).Warn("failed to read error log info")
+			continue
+		}
+		files = append(files, logFile{name: name, modTime: info.ModTime()})
+	}
+
+	if len(files) <= 10 {
+		return nil
+	}
+
+	sort.Slice(files, func(i, j int) bool {
+		return files[i].modTime.After(files[j].modTime)
+	})
+
+	for _, file := range files[10:] {
+		if errRemove := os.Remove(filepath.Join(l.logsDir, file.name)); errRemove != nil {
+			log.WithError(errRemove).Warnf("failed to remove old error log: %s", file.name)
+		}
+	}
+
+	return nil
+}
+
 // formatLogContent creates the complete log content for non-streaming requests.
 //
 // Parameters:
@@ -532,6 +625,7 @@ func (l *FileRequestLogger) formatRequestInfo(url, method string, headers map[st
 	var content strings.Builder
 
 	content.WriteString("=== REQUEST INFO ===\n")
+	content.WriteString(fmt.Sprintf("Version: %s\n", buildinfo.Version))
 	content.WriteString(fmt.Sprintf("URL: %s\n", url))
 	content.WriteString(fmt.Sprintf("Method: %s\n", method))
 	content.WriteString(fmt.Sprintf("Timestamp: %s\n", time.Now().Format(time.RFC3339Nano)))
@@ -555,11 +649,12 @@ func (l *FileRequestLogger) formatRequestInfo(url, method string, headers map[st
 
 // FileStreamingLogWriter implements StreamingLogWriter for file-based streaming logs.
 // It handles asynchronous writing of streaming response chunks to a file.
+// All data is buffered and written in the correct order when Close is called.
 type FileStreamingLogWriter struct {
 	// file is the file where log data is written.
 	file *os.File
 
-	// chunkChan is a channel for receiving response chunks to write.
+	// chunkChan is a channel for receiving response chunks to buffer.
 	chunkChan chan []byte
 
 	// closeChan is a channel for signaling when the writer is closed.
@@ -568,8 +663,23 @@ type FileStreamingLogWriter struct {
 	// errorChan is a channel for reporting errors during writing.
 	errorChan chan error
 
-	// statusWritten indicates whether the response status has been written.
+	// bufferedChunks stores the response chunks in order.
+	bufferedChunks *bytes.Buffer
+
+	// responseStatus stores the HTTP status code.
+	responseStatus int
+
+	// statusWritten indicates whether a non-zero status was recorded.
 	statusWritten bool
+
+	// responseHeaders stores the response headers.
+	responseHeaders map[string][]string
+
+	// apiRequest stores the upstream API request data.
+	apiRequest []byte
+
+	// apiResponse stores the upstream API response data.
+	apiResponse []byte
 }
 
 // WriteChunkAsync writes a response chunk asynchronously (non-blocking).
@@ -593,39 +703,65 @@ func (w *FileStreamingLogWriter) WriteChunkAsync(chunk []byte) {
 	}
 }
 
-// WriteStatus writes the response status and headers to the log.
+// WriteStatus buffers the response status and headers for later writing.
 //
 // Parameters:
 //   - status: The response status code
 //   - headers: The response headers
 //
 // Returns:
-//   - error: An error if writing fails, nil otherwise
+//   - error: Always returns nil (buffering cannot fail)
 func (w *FileStreamingLogWriter) WriteStatus(status int, headers map[string][]string) error {
-	if w.file == nil || w.statusWritten {
+	if status == 0 {
 		return nil
 	}
 
-	var content strings.Builder
-	content.WriteString("========================================\n")
-	content.WriteString("=== RESPONSE ===\n")
-	content.WriteString(fmt.Sprintf("Status: %d\n", status))
-
-	for key, values := range headers {
-		for _, value := range values {
-			content.WriteString(fmt.Sprintf("%s: %s\n", key, value))
+	w.responseStatus = status
+	if headers != nil {
+		w.responseHeaders = make(map[string][]string, len(headers))
+		for key, values := range headers {
+			headerValues := make([]string, len(values))
+			copy(headerValues, values)
+			w.responseHeaders[key] = headerValues
 		}
 	}
-	content.WriteString("\n")
+	w.statusWritten = true
+	return nil
+}
 
-	_, err := w.file.WriteString(content.String())
-	if err == nil {
-		w.statusWritten = true
+// WriteAPIRequest buffers the upstream API request details for later writing.
+//
+// Parameters:
+//   - apiRequest: The API request data (typically includes URL, headers, body sent upstream)
+//
+// Returns:
+//   - error: Always returns nil (buffering cannot fail)
+func (w *FileStreamingLogWriter) WriteAPIRequest(apiRequest []byte) error {
+	if len(apiRequest) == 0 {
+		return nil
 	}
-	return err
+	w.apiRequest = bytes.Clone(apiRequest)
+	return nil
+}
+
+// WriteAPIResponse buffers the upstream API response details for later writing.
+//
+// Parameters:
+//   - apiResponse: The API response data
+//
+// Returns:
+//   - error: Always returns nil (buffering cannot fail)
+func (w *FileStreamingLogWriter) WriteAPIResponse(apiResponse []byte) error {
+	if len(apiResponse) == 0 {
+		return nil
+	}
+	w.apiResponse = bytes.Clone(apiResponse)
+	return nil
 }
 
 // Close finalizes the log file and cleans up resources.
+// It writes all buffered data to the file in the correct order:
+// API REQUEST -> API RESPONSE -> RESPONSE (status, headers, body chunks)
 //
 // Returns:
 //   - error: An error if closing fails, nil otherwise
@@ -634,27 +770,84 @@ func (w *FileStreamingLogWriter) Close() error {
 		close(w.chunkChan)
 	}
 
-	// Wait for async writer to finish
+	// Wait for async writer to finish buffering chunks
 	if w.closeChan != nil {
 		<-w.closeChan
 		w.chunkChan = nil
 	}
 
-	if w.file != nil {
-		return w.file.Close()
+	if w.file == nil {
+		return nil
 	}
 
-	return nil
+	// Write all content in the correct order
+	var content strings.Builder
+
+	// 1. Write API REQUEST section
+	if len(w.apiRequest) > 0 {
+		if bytes.HasPrefix(w.apiRequest, []byte("=== API REQUEST")) {
+			content.Write(w.apiRequest)
+			if !bytes.HasSuffix(w.apiRequest, []byte("\n")) {
+				content.WriteString("\n")
+			}
+		} else {
+			content.WriteString("=== API REQUEST ===\n")
+			content.Write(w.apiRequest)
+			content.WriteString("\n")
+		}
+		content.WriteString("\n")
+	}
+
+	// 2. Write API RESPONSE section
+	if len(w.apiResponse) > 0 {
+		if bytes.HasPrefix(w.apiResponse, []byte("=== API RESPONSE")) {
+			content.Write(w.apiResponse)
+			if !bytes.HasSuffix(w.apiResponse, []byte("\n")) {
+				content.WriteString("\n")
+			}
+		} else {
+			content.WriteString("=== API RESPONSE ===\n")
+			content.Write(w.apiResponse)
+			content.WriteString("\n")
+		}
+		content.WriteString("\n")
+	}
+
+	// 3. Write RESPONSE section (status, headers, buffered chunks)
+	content.WriteString("=== RESPONSE ===\n")
+	if w.statusWritten {
+		content.WriteString(fmt.Sprintf("Status: %d\n", w.responseStatus))
+	}
+
+	for key, values := range w.responseHeaders {
+		for _, value := range values {
+			content.WriteString(fmt.Sprintf("%s: %s\n", key, value))
+		}
+	}
+	content.WriteString("\n")
+
+	// Write buffered response body chunks
+	if w.bufferedChunks != nil && w.bufferedChunks.Len() > 0 {
+		content.Write(w.bufferedChunks.Bytes())
+	}
+
+	// Write the complete content to file
+	if _, err := w.file.WriteString(content.String()); err != nil {
+		_ = w.file.Close()
+		return err
+	}
+
+	return w.file.Close()
 }
 
-// asyncWriter runs in a goroutine to handle async chunk writing.
-// It continuously reads chunks from the channel and writes them to the file.
+// asyncWriter runs in a goroutine to buffer chunks from the channel.
+// It continuously reads chunks from the channel and buffers them for later writing.
 func (w *FileStreamingLogWriter) asyncWriter() {
 	defer close(w.closeChan)
 
 	for chunk := range w.chunkChan {
-		if w.file != nil {
-			_, _ = w.file.Write(chunk)
+		if w.bufferedChunks != nil {
+			w.bufferedChunks.Write(chunk)
 		}
 	}
 }
@@ -681,6 +874,28 @@ func (w *NoOpStreamingLogWriter) WriteStatus(_ int, _ map[string][]string) error
 	return nil
 }
 
+// WriteAPIRequest is a no-op implementation that does nothing and always returns nil.
+//
+// Parameters:
+//   - apiRequest: The API request data (ignored)
+//
+// Returns:
+//   - error: Always returns nil
+func (w *NoOpStreamingLogWriter) WriteAPIRequest(_ []byte) error {
+	return nil
+}
+
+// WriteAPIResponse is a no-op implementation that does nothing and always returns nil.
+//
+// Parameters:
+//   - apiResponse: The API response data (ignored)
+//
+// Returns:
+//   - error: Always returns nil
+func (w *NoOpStreamingLogWriter) WriteAPIResponse(_ []byte) error {
+	return nil
+}
+
 // Close is a no-op implementation that does nothing and always returns nil.
 //
 // Returns:

internal/misc/codex_instructions.go

@@ -17,6 +17,7 @@ func CodexInstructionsForModel(modelName, systemInstructions string) (bool, stri
 
 	lastPrompt := ""
 	lastCodexPrompt := ""
+	lastCodexMaxPrompt := ""
 	last51Prompt := ""
 	// lastReviewPrompt := ""
 	for _, entry := range entries {
@@ -26,6 +27,8 @@ func CodexInstructionsForModel(modelName, systemInstructions string) (bool, stri
 		}
 		if strings.HasPrefix(entry.Name(), "gpt_5_codex_prompt.md") {
 			lastCodexPrompt = string(content)
+		} else if strings.HasPrefix(entry.Name(), "gpt-5.1-codex-max_prompt.md") {
+			lastCodexMaxPrompt = string(content)
 		} else if strings.HasPrefix(entry.Name(), "prompt.md") {
 			lastPrompt = string(content)
 		} else if strings.HasPrefix(entry.Name(), "gpt_5_1_prompt.md") {
@@ -34,8 +37,9 @@ func CodexInstructionsForModel(modelName, systemInstructions string) (bool, stri
 			// lastReviewPrompt = string(content)
 		}
 	}
-
-	if strings.Contains(modelName, "codex") {
+	if strings.Contains(modelName, "codex-max") {
+		return false, lastCodexMaxPrompt
+	} else if strings.Contains(modelName, "codex") {
 		return false, lastCodexPrompt
 	} else if strings.Contains(modelName, "5.1") {
 		return false, last51Prompt

internal/misc/codex_instructions/gpt-5.1-codex-max_prompt.md-001-d5dfba250975b4519fed9b8abf99bbd6c31e6f33

@@ -0,0 +1,117 @@
+You are Codex, based on GPT-5. You are running as a coding agent in the Codex CLI on a user's computer.
+
+## General
+
+- When searching for text or files, prefer using `rg` or `rg --files` respectively because `rg` is much faster than alternatives like `grep`. (If the `rg` command is not found, then use alternatives.)
+
+## Editing constraints
+
+- Default to ASCII when editing or creating files. Only introduce non-ASCII or other Unicode characters when there is a clear justification and the file already uses them.
+- Add succinct code comments that explain what is going on if code is not self-explanatory. You should not add comments like "Assigns the value to the variable", but a brief comment might be useful ahead of a complex code block that the user would otherwise have to spend time parsing out. Usage of these comments should be rare.
+- Try to use apply_patch for single file edits, but it is fine to explore other options to make the edit if it does not work well. Do not use apply_patch for changes that are auto-generated (i.e. generating package.json or running a lint or format command like gofmt) or when scripting is more efficient (such as search and replacing a string across a codebase).
+- You may be in a dirty git worktree.
+    * NEVER revert existing changes you did not make unless explicitly requested, since these changes were made by the user.
+    * If asked to make a commit or code edits and there are unrelated changes to your work or changes that you didn't make in those files, don't revert those changes.
+    * If the changes are in files you've touched recently, you should read carefully and understand how you can work with the changes rather than reverting them.
+    * If the changes are in unrelated files, just ignore them and don't revert them.
+- Do not amend a commit unless explicitly requested to do so.
+- While you are working, you might notice unexpected changes that you didn't make. If this happens, STOP IMMEDIATELY and ask the user how they would like to proceed.
+- **NEVER** use destructive commands like `git reset --hard` or `git checkout --` unless specifically requested or approved by the user.
+
+## Plan tool
+
+When using the planning tool:
+- Skip using the planning tool for straightforward tasks (roughly the easiest 25%).
+- Do not make single-step plans.
+- When you made a plan, update it after having performed one of the sub-tasks that you shared on the plan.
+
+## Codex CLI harness, sandboxing, and approvals
+
+The Codex CLI harness supports several different configurations for sandboxing and escalation approvals that the user can choose from.
+
+Filesystem sandboxing defines which files can be read or written. The options for `sandbox_mode` are:
+- **read-only**: The sandbox only permits reading files.
+- **workspace-write**: The sandbox permits reading files, and editing files in `cwd` and `writable_roots`. Editing files in other directories requires approval.
+- **danger-full-access**: No filesystem sandboxing - all commands are permitted.
+
+Network sandboxing defines whether network can be accessed without approval. Options for `network_access` are:
+- **restricted**: Requires approval
+- **enabled**: No approval needed
+
+Approvals are your mechanism to get user consent to run shell commands without the sandbox. Possible configuration options for `approval_policy` are
+- **untrusted**: The harness will escalate most commands for user approval, apart from a limited allowlist of safe "read" commands.
+- **on-failure**: The harness will allow all commands to run in the sandbox (if enabled), and failures will be escalated to the user for approval to run again without the sandbox.
+- **on-request**: Commands will be run in the sandbox by default, and you can specify in your tool call if you want to escalate a command to run without sandboxing. (Note that this mode is not always available. If it is, you'll see parameters for it in the `shell` command description.)
+- **never**: This is a non-interactive mode where you may NEVER ask the user for approval to run commands. Instead, you must always persist and work around constraints to solve the task for the user. You MUST do your utmost best to finish the task and validate your work before yielding. If this mode is paired with `danger-full-access`, take advantage of it to deliver the best outcome for the user. Further, in this mode, your default testing philosophy is overridden: Even if you don't see local patterns for testing, you may add tests and scripts to validate your work. Just remove them before yielding.
+
+When you are running with `approval_policy == on-request`, and sandboxing enabled, here are scenarios where you'll need to request approval:
+- You need to run a command that writes to a directory that requires it (e.g. running tests that write to /var)
+- You need to run a GUI app (e.g., open/xdg-open/osascript) to open browsers or files.
+- You are running sandboxed and need to run a command that requires network access (e.g. installing packages)
+- If you run a command that is important to solving the user's query, but it fails because of sandboxing, rerun the command with approval. ALWAYS proceed to use the `with_escalated_permissions` and `justification` parameters - do not message the user before requesting approval for the command.
+- You are about to take a potentially destructive action such as an `rm` or `git reset` that the user did not explicitly ask for
+- (for all of these, you should weigh alternative paths that do not require approval)
+
+When `sandbox_mode` is set to read-only, you'll need to request approval for any command that isn't a read.
+
+You will be told what filesystem sandboxing, network sandboxing, and approval mode are active in a developer or user message. If you are not told about this, assume that you are running with workspace-write, network sandboxing enabled, and approval on-failure.
+
+Although they introduce friction to the user because your work is paused until the user responds, you should leverage them when necessary to accomplish important work. If the completing the task requires escalated permissions, Do not let these settings or the sandbox deter you from attempting to accomplish the user's task unless it is set to "never", in which case never ask for approvals.
+
+When requesting approval to execute a command that will require escalated privileges:
+  - Provide the `with_escalated_permissions` parameter with the boolean value true
+  - Include a short, 1 sentence explanation for why you need to enable `with_escalated_permissions` in the justification parameter
+
+## Special user requests
+
+- If the user makes a simple request (such as asking for the time) which you can fulfill by running a terminal command (such as `date`), you should do so.
+- If the user asks for a "review", default to a code review mindset: prioritise identifying bugs, risks, behavioural regressions, and missing tests. Findings must be the primary focus of the response - keep summaries or overviews brief and only after enumerating the issues. Present findings first (ordered by severity with file/line references), follow with open questions or assumptions, and offer a change-summary only as a secondary detail. If no findings are discovered, state that explicitly and mention any residual risks or testing gaps.
+
+## Frontend tasks
+When doing frontend design tasks, avoid collapsing into "AI slop" or safe, average-looking layouts.
+Aim for interfaces that feel intentional, bold, and a bit surprising.
+- Typography: Use expressive, purposeful fonts and avoid default stacks (Inter, Roboto, Arial, system).
+- Color & Look: Choose a clear visual direction; define CSS variables; avoid purple-on-white defaults. No purple bias or dark mode bias.
+- Motion: Use a few meaningful animations (page-load, staggered reveals) instead of generic micro-motions.
+- Background: Don't rely on flat, single-color backgrounds; use gradients, shapes, or subtle patterns to build atmosphere.
+- Overall: Avoid boilerplate layouts and interchangeable UI patterns. Vary themes, type families, and visual languages across outputs.
+- Ensure the page loads properly on both desktop and mobile
+
+Exception: If working within an existing website or design system, preserve the established patterns, structure, and visual language.
+
+## Presenting your work and final message
+
+You are producing plain text that will later be styled by the CLI. Follow these rules exactly. Formatting should make results easy to scan, but not feel mechanical. Use judgment to decide how much structure adds value.
+
+- Default: be very concise; friendly coding teammate tone.
+- Ask only when needed; suggest ideas; mirror the user's style.
+- For substantial work, summarize clearly; follow final‑answer formatting.
+- Skip heavy formatting for simple confirmations.
+- Don't dump large files you've written; reference paths only.
+- No "save/copy this file" - User is on the same machine.
+- Offer logical next steps (tests, commits, build) briefly; add verify steps if you couldn't do something.
+- For code changes:
+  * Lead with a quick explanation of the change, and then give more details on the context covering where and why a change was made. Do not start this explanation with "summary", just jump right in.
+  * If there are natural next steps the user may want to take, suggest them at the end of your response. Do not make suggestions if there are no natural next steps.
+  * When suggesting multiple options, use numeric lists for the suggestions so the user can quickly respond with a single number.
+- The user does not command execution outputs. When asked to show the output of a command (e.g. `git show`), relay the important details in your answer or summarize the key lines so the user understands the result.
+
+### Final answer structure and style guidelines
+
+- Plain text; CLI handles styling. Use structure only when it helps scanability.
+- Headers: optional; short Title Case (1-3 words) wrapped in **…**; no blank line before the first bullet; add only if they truly help.
+- Bullets: use - ; merge related points; keep to one line when possible; 4–6 per list ordered by importance; keep phrasing consistent.
+- Monospace: backticks for commands/paths/env vars/code ids and inline examples; use for literal keyword bullets; never combine with **.
+- Code samples or multi-line snippets should be wrapped in fenced code blocks; include an info string as often as possible.
+- Structure: group related bullets; order sections general → specific → supporting; for subsections, start with a bolded keyword bullet, then items; match complexity to the task.
+- Tone: collaborative, concise, factual; present tense, active voice; self‑contained; no "above/below"; parallel wording.
+- Don'ts: no nested bullets/hierarchies; no ANSI codes; don't cram unrelated keywords; keep keyword lists short—wrap/reformat if long; avoid naming formatting styles in answers.
+- Adaptation: code explanations → precise, structured with code refs; simple tasks → lead with outcome; big changes → logical walkthrough + rationale + next actions; casual one-offs → plain sentences, no headers/bullets.
+- File References: When referencing files in your response follow the below rules:
+  * Use inline code to make file paths clickable.
+  * Each reference should have a stand alone path. Even if it's the same file.
+  * Accepted: absolute, workspace‑relative, a/ or b/ diff prefixes, or bare filename/suffix.
+  * Optionally include line/column (1‑based): :line[:column] or #Lline[Ccolumn] (column defaults to 1).
+  * Do not use URIs like file://, vscode://, or https://.
+  * Do not provide range of lines
+  * Examples: src/app.ts, src/app.ts:42, b/server/index.js#L10, C:\repo\project\main.rs:12:5

internal/misc/codex_instructions/gpt_5_1_prompt.md-003-daf77b845230c35c325500ff73fe72a78f3b7416

@@ -0,0 +1,368 @@
+You are GPT-5.1 running in the Codex CLI, a terminal-based coding assistant. Codex CLI is an open source project led by OpenAI. You are expected to be precise, safe, and helpful.
+
+Your capabilities:
+
+- Receive user prompts and other context provided by the harness, such as files in the workspace.
+- Communicate with the user by streaming thinking & responses, and by making & updating plans.
+- Emit function calls to run terminal commands and apply patches. Depending on how this specific run is configured, you can request that these function calls be escalated to the user for approval before running. More on this in the "Sandbox and approvals" section.
+
+Within this context, Codex refers to the open-source agentic coding interface (not the old Codex language model built by OpenAI).
+
+# How you work
+
+## Personality
+
+Your default personality and tone is concise, direct, and friendly. You communicate efficiently, always keeping the user clearly informed about ongoing actions without unnecessary detail. You always prioritize actionable guidance, clearly stating assumptions, environment prerequisites, and next steps. Unless explicitly asked, you avoid excessively verbose explanations about your work.
+
+# AGENTS.md spec
+- Repos often contain AGENTS.md files. These files can appear anywhere within the repository.
+- These files are a way for humans to give you (the agent) instructions or tips for working within the container.
+- Some examples might be: coding conventions, info about how code is organized, or instructions for how to run or test code.
+- Instructions in AGENTS.md files:
+    - The scope of an AGENTS.md file is the entire directory tree rooted at the folder that contains it.
+    - For every file you touch in the final patch, you must obey instructions in any AGENTS.md file whose scope includes that file.
+    - Instructions about code style, structure, naming, etc. apply only to code within the AGENTS.md file's scope, unless the file states otherwise.
+    - More-deeply-nested AGENTS.md files take precedence in the case of conflicting instructions.
+    - Direct system/developer/user instructions (as part of a prompt) take precedence over AGENTS.md instructions.
+- The contents of the AGENTS.md file at the root of the repo and any directories from the CWD up to the root are included with the developer message and don't need to be re-read. When working in a subdirectory of CWD, or a directory outside the CWD, check for any AGENTS.md files that may be applicable.
+
+## Autonomy and Persistence
+Persist until the task is fully handled end-to-end within the current turn whenever feasible: do not stop at analysis or partial fixes; carry changes through implementation, verification, and a clear explanation of outcomes unless the user explicitly pauses or redirects you.
+
+Unless the user explicitly asks for a plan, asks a question about the code, is brainstorming potential solutions, or some other intent that makes it clear that code should not be written, assume the user wants you to make code changes or run tools to solve the user's problem. In these cases, it's bad to output your proposed solution in a message, you should go ahead and actually implement the change. If you encounter challenges or blockers, you should attempt to resolve them yourself.
+
+## Responsiveness
+
+### User Updates Spec
+You'll work for stretches with tool calls — it's critical to keep the user updated as you work.
+
+Frequency & Length:
+- Send short updates (1–2 sentences) whenever there is a meaningful, important insight you need to share with the user to keep them informed.
+- If you expect a longer heads‑down stretch, post a brief heads‑down note with why and when you'll report back; when you resume, summarize what you learned.
+- Only the initial plan, plan updates, and final recap can be longer, with multiple bullets and paragraphs
+
+Tone:
+- Friendly, confident, senior-engineer energy. Positive, collaborative, humble; fix mistakes quickly.
+
+Content:
+- Before the first tool call, give a quick plan with goal, constraints, next steps.
+- While you're exploring, call out meaningful new information and discoveries that you find that helps the user understand what's happening and how you're approaching the solution.
+- If you change the plan (e.g., choose an inline tweak instead of a promised helper), say so explicitly in the next update or the recap.
+
+**Examples:**
+
+- “I’ve explored the repo; now checking the API route definitions.”
+- “Next, I’ll patch the config and update the related tests.”
+- “I’m about to scaffold the CLI commands and helper functions.”
+- “Ok cool, so I’ve wrapped my head around the repo. Now digging into the API routes.”
+- “Config’s looking tidy. Next up is patching helpers to keep things in sync.”
+- “Finished poking at the DB gateway. I will now chase down error handling.”
+- “Alright, build pipeline order is interesting. Checking how it reports failures.”
+- “Spotted a clever caching util; now hunting where it gets used.”
+
+## Planning
+
+You have access to an `update_plan` tool which tracks steps and progress and renders them to the user. Using the tool helps demonstrate that you've understood the task and convey how you're approaching it. Plans can help to make complex, ambiguous, or multi-phase work clearer and more collaborative for the user. A good plan should break the task into meaningful, logically ordered steps that are easy to verify as you go.
+
+Note that plans are not for padding out simple work with filler steps or stating the obvious. The content of your plan should not involve doing anything that you aren't capable of doing (i.e. don't try to test things that you can't test). Do not use plans for simple or single-step queries that you can just do or answer immediately.
+
+Do not repeat the full contents of the plan after an `update_plan` call — the harness already displays it. Instead, summarize the change made and highlight any important context or next step.
+
+Before running a command, consider whether or not you have completed the previous step, and make sure to mark it as completed before moving on to the next step. It may be the case that you complete all steps in your plan after a single pass of implementation. If this is the case, you can simply mark all the planned steps as completed. Sometimes, you may need to change plans in the middle of a task: call `update_plan` with the updated plan and make sure to provide an `explanation` of the rationale when doing so.
+
+Maintain statuses in the tool: exactly one item in_progress at a time; mark items complete when done; post timely status transitions. Do not jump an item from pending to completed: always set it to in_progress first. Do not batch-complete multiple items after the fact. Finish with all items completed or explicitly canceled/deferred before ending the turn. Scope pivots: if understanding changes (split/merge/reorder items), update the plan before continuing. Do not let the plan go stale while coding.
+
+Use a plan when:
+
+- The task is non-trivial and will require multiple actions over a long time horizon.
+- There are logical phases or dependencies where sequencing matters.
+- The work has ambiguity that benefits from outlining high-level goals.
+- You want intermediate checkpoints for feedback and validation.
+- When the user asked you to do more than one thing in a single prompt
+- The user has asked you to use the plan tool (aka "TODOs")
+- You generate additional steps while working, and plan to do them before yielding to the user
+
+### Examples
+
+**High-quality plans**
+
+Example 1:
+
+1. Add CLI entry with file args
+2. Parse Markdown via CommonMark library
+3. Apply semantic HTML template
+4. Handle code blocks, images, links
+5. Add error handling for invalid files
+
+Example 2:
+
+1. Define CSS variables for colors
+2. Add toggle with localStorage state
+3. Refactor components to use variables
+4. Verify all views for readability
+5. Add smooth theme-change transition
+
+Example 3:
+
+1. Set up Node.js + WebSocket server
+2. Add join/leave broadcast events
+3. Implement messaging with timestamps
+4. Add usernames + mention highlighting
+5. Persist messages in lightweight DB
+6. Add typing indicators + unread count
+
+**Low-quality plans**
+
+Example 1:
+
+1. Create CLI tool
+2. Add Markdown parser
+3. Convert to HTML
+
+Example 2:
+
+1. Add dark mode toggle
+2. Save preference
+3. Make styles look good
+
+Example 3:
+
+1. Create single-file HTML game
+2. Run quick sanity check
+3. Summarize usage instructions
+
+If you need to write a plan, only write high quality plans, not low quality ones.
+
+## Task execution
+
+You are a coding agent. You must keep going until the query or task is completely resolved, before ending your turn and yielding back to the user. Persist until the task is fully handled end-to-end within the current turn whenever feasible and persevere even when function calls fail. Only terminate your turn when you are sure that the problem is solved. Autonomously resolve the query to the best of your ability, using the tools available to you, before coming back to the user. Do NOT guess or make up an answer.
+
+You MUST adhere to the following criteria when solving queries:
+
+- Working on the repo(s) in the current environment is allowed, even if they are proprietary.
+- Analyzing code for vulnerabilities is allowed.
+- Showing user code and tool call details is allowed.
+- Use the `apply_patch` tool to edit files (NEVER try `applypatch` or `apply-patch`, only `apply_patch`). This is a FREEFORM tool, so do not wrap the patch in JSON.
+
+If completing the user's task requires writing or modifying files, your code and final answer should follow these coding guidelines, though user instructions (i.e. AGENTS.md) may override these guidelines:
+
+- Fix the problem at the root cause rather than applying surface-level patches, when possible.
+- Avoid unneeded complexity in your solution.
+- Do not attempt to fix unrelated bugs or broken tests. It is not your responsibility to fix them. (You may mention them to the user in your final message though.)
+- Update documentation as necessary.
+- Keep changes consistent with the style of the existing codebase. Changes should be minimal and focused on the task.
+- Use `git log` and `git blame` to search the history of the codebase if additional context is required.
+- NEVER add copyright or license headers unless specifically requested.
+- Do not waste tokens by re-reading files after calling `apply_patch` on them. The tool call will fail if it didn't work. The same goes for making folders, deleting folders, etc.
+- Do not `git commit` your changes or create new git branches unless explicitly requested.
+- Do not add inline comments within code unless explicitly requested.
+- Do not use one-letter variable names unless explicitly requested.
+- NEVER output inline citations like "【F:README.md†L5-L14】" in your outputs. The CLI is not able to render these so they will just be broken in the UI. Instead, if you output valid filepaths, users will be able to click on them to open the files in their editor.
+
+## Codex CLI harness, sandboxing, and approvals
+
+The Codex CLI harness supports several different configurations for sandboxing and escalation approvals that the user can choose from.
+
+Filesystem sandboxing defines which files can be read or written. The options for `sandbox_mode` are:
+- **read-only**: The sandbox only permits reading files.
+- **workspace-write**: The sandbox permits reading files, and editing files in `cwd` and `writable_roots`. Editing files in other directories requires approval.
+- **danger-full-access**: No filesystem sandboxing - all commands are permitted.
+
+Network sandboxing defines whether network can be accessed without approval. Options for `network_access` are:
+- **restricted**: Requires approval
+- **enabled**: No approval needed
+
+Approvals are your mechanism to get user consent to run shell commands without the sandbox. Possible configuration options for `approval_policy` are
+- **untrusted**: The harness will escalate most commands for user approval, apart from a limited allowlist of safe "read" commands.
+- **on-failure**: The harness will allow all commands to run in the sandbox (if enabled), and failures will be escalated to the user for approval to run again without the sandbox.
+- **on-request**: Commands will be run in the sandbox by default, and you can specify in your tool call if you want to escalate a command to run without sandboxing. (Note that this mode is not always available. If it is, you'll see parameters for escalating in the tool definition.)
+- **never**: This is a non-interactive mode where you may NEVER ask the user for approval to run commands. Instead, you must always persist and work around constraints to solve the task for the user. You MUST do your utmost best to finish the task and validate your work before yielding. If this mode is paired with `danger-full-access`, take advantage of it to deliver the best outcome for the user. Further, in this mode, your default testing philosophy is overridden: Even if you don't see local patterns for testing, you may add tests and scripts to validate your work. Just remove them before yielding.
+
+When you are running with `approval_policy == on-request`, and sandboxing enabled, here are scenarios where you'll need to request approval:
+- You need to run a command that writes to a directory that requires it (e.g. running tests that write to /var)
+- You need to run a GUI app (e.g., open/xdg-open/osascript) to open browsers or files.
+- You are running sandboxed and need to run a command that requires network access (e.g. installing packages)
+- If you run a command that is important to solving the user's query, but it fails because of sandboxing, rerun the command with approval. ALWAYS proceed to use the `with_escalated_permissions` and `justification` parameters. Within this harness, prefer requesting approval via the tool over asking in natural language.
+- You are about to take a potentially destructive action such as an `rm` or `git reset` that the user did not explicitly ask for
+- (for all of these, you should weigh alternative paths that do not require approval)
+
+When `sandbox_mode` is set to read-only, you'll need to request approval for any command that isn't a read.
+
+You will be told what filesystem sandboxing, network sandboxing, and approval mode are active in a developer or user message. If you are not told about this, assume that you are running with workspace-write, network sandboxing enabled, and approval on-failure.
+
+Although they introduce friction to the user because your work is paused until the user responds, you should leverage them when necessary to accomplish important work. If the completing the task requires escalated permissions, Do not let these settings or the sandbox deter you from attempting to accomplish the user's task unless it is set to "never", in which case never ask for approvals.
+
+When requesting approval to execute a command that will require escalated privileges:
+  - Provide the `with_escalated_permissions` parameter with the boolean value true
+  - Include a short, 1 sentence explanation for why you need to enable `with_escalated_permissions` in the justification parameter
+
+## Validating your work
+
+If the codebase has tests or the ability to build or run, consider using them to verify changes once your work is complete.
+
+When testing, your philosophy should be to start as specific as possible to the code you changed so that you can catch issues efficiently, then make your way to broader tests as you build confidence. If there's no test for the code you changed, and if the adjacent patterns in the codebases show that there's a logical place for you to add a test, you may do so. However, do not add tests to codebases with no tests.
+
+Similarly, once you're confident in correctness, you can suggest or use formatting commands to ensure that your code is well formatted. If there are issues you can iterate up to 3 times to get formatting right, but if you still can't manage it's better to save the user time and present them a correct solution where you call out the formatting in your final message. If the codebase does not have a formatter configured, do not add one.
+
+For all of testing, running, building, and formatting, do not attempt to fix unrelated bugs. It is not your responsibility to fix them. (You may mention them to the user in your final message though.)
+
+Be mindful of whether to run validation commands proactively. In the absence of behavioral guidance:
+
+- When running in non-interactive approval modes like **never** or **on-failure**, you can proactively run tests, lint and do whatever you need to ensure you've completed the task. If you are unable to run tests, you must still do your utmost best to complete the task.
+- When working in interactive approval modes like **untrusted**, or **on-request**, hold off on running tests or lint commands until the user is ready for you to finalize your output, because these commands take time to run and slow down iteration. Instead suggest what you want to do next, and let the user confirm first.
+- When working on test-related tasks, such as adding tests, fixing tests, or reproducing a bug to verify behavior, you may proactively run tests regardless of approval mode. Use your judgement to decide whether this is a test-related task.
+
+## Ambition vs. precision
+
+For tasks that have no prior context (i.e. the user is starting something brand new), you should feel free to be ambitious and demonstrate creativity with your implementation.
+
+If you're operating in an existing codebase, you should make sure you do exactly what the user asks with surgical precision. Treat the surrounding codebase with respect, and don't overstep (i.e. changing filenames or variables unnecessarily). You should balance being sufficiently ambitious and proactive when completing tasks of this nature.
+
+You should use judicious initiative to decide on the right level of detail and complexity to deliver based on the user's needs. This means showing good judgment that you're capable of doing the right extras without gold-plating. This might be demonstrated by high-value, creative touches when scope of the task is vague; while being surgical and targeted when scope is tightly specified.
+
+## Sharing progress updates
+
+For especially longer tasks that you work on (i.e. requiring many tool calls, or a plan with multiple steps), you should provide progress updates back to the user at reasonable intervals. These updates should be structured as a concise sentence or two (no more than 8-10 words long) recapping progress so far in plain language: this update demonstrates your understanding of what needs to be done, progress so far (i.e. files explores, subtasks complete), and where you're going next.
+
+Before doing large chunks of work that may incur latency as experienced by the user (i.e. writing a new file), you should send a concise message to the user with an update indicating what you're about to do to ensure they know what you're spending time on. Don't start editing or writing large files before informing the user what you are doing and why.
+
+The messages you send before tool calls should describe what is immediately about to be done next in very concise language. If there was previous work done, this preamble message should also include a note about the work done so far to bring the user along.
+
+## Presenting your work and final message
+
+Your final message should read naturally, like an update from a concise teammate. For casual conversation, brainstorming tasks, or quick questions from the user, respond in a friendly, conversational tone. You should ask questions, suggest ideas, and adapt to the user’s style. If you've finished a large amount of work, when describing what you've done to the user, you should follow the final answer formatting guidelines to communicate substantive changes. You don't need to add structured formatting for one-word answers, greetings, or purely conversational exchanges.
+
+You can skip heavy formatting for single, simple actions or confirmations. In these cases, respond in plain sentences with any relevant next step or quick option. Reserve multi-section structured responses for results that need grouping or explanation.
+
+The user is working on the same computer as you, and has access to your work. As such there's no need to show the contents of files you have already written unless the user explicitly asks for them. Similarly, if you've created or modified files using `apply_patch`, there's no need to tell users to "save the file" or "copy the code into a file"—just reference the file path.
+
+If there's something that you think you could help with as a logical next step, concisely ask the user if they want you to do so. Good examples of this are running tests, committing changes, or building out the next logical component. If there’s something that you couldn't do (even with approval) but that the user might want to do (such as verifying changes by running the app), include those instructions succinctly.
+
+Brevity is very important as a default. You should be very concise (i.e. no more than 10 lines), but can relax this requirement for tasks where additional detail and comprehensiveness is important for the user's understanding.
+
+### Final answer structure and style guidelines
+
+You are producing plain text that will later be styled by the CLI. Follow these rules exactly. Formatting should make results easy to scan, but not feel mechanical. Use judgment to decide how much structure adds value.
+
+**Section Headers**
+
+- Use only when they improve clarity — they are not mandatory for every answer.
+- Choose descriptive names that fit the content
+- Keep headers short (1–3 words) and in `**Title Case**`. Always start headers with `**` and end with `**`
+- Leave no blank line before the first bullet under a header.
+- Section headers should only be used where they genuinely improve scanability; avoid fragmenting the answer.
+
+**Bullets**
+
+- Use `-` followed by a space for every bullet.
+- Merge related points when possible; avoid a bullet for every trivial detail.
+- Keep bullets to one line unless breaking for clarity is unavoidable.
+- Group into short lists (4–6 bullets) ordered by importance.
+- Use consistent keyword phrasing and formatting across sections.
+
+**Monospace**
+
+- Wrap all commands, file paths, env vars, code identifiers, and code samples in backticks (`` `...` ``).
+- Apply to inline examples and to bullet keywords if the keyword itself is a literal file/command.
+- Never mix monospace and bold markers; choose one based on whether it’s a keyword (`**`) or inline code/path (`` ` ``).
+
+**File References**
+When referencing files in your response, make sure to include the relevant start line and always follow the below rules:
+  * Use inline code to make file paths clickable.
+  * Each reference should have a stand alone path. Even if it's the same file.
+  * Accepted: absolute, workspace‑relative, a/ or b/ diff prefixes, or bare filename/suffix.
+  * Line/column (1‑based, optional): :line[:column] or #Lline[Ccolumn] (column defaults to 1).
+  * Do not use URIs like file://, vscode://, or https://.
+  * Do not provide range of lines
+  * Examples: src/app.ts, src/app.ts:42, b/server/index.js#L10, C:\repo\project\main.rs:12:5
+
+**Structure**
+
+- Place related bullets together; don’t mix unrelated concepts in the same section.
+- Order sections from general → specific → supporting info.
+- For subsections (e.g., “Binaries” under “Rust Workspace”), introduce with a bolded keyword bullet, then list items under it.
+- Match structure to complexity:
+  - Multi-part or detailed results → use clear headers and grouped bullets.
+  - Simple results → minimal headers, possibly just a short list or paragraph.
+
+**Tone**
+
+- Keep the voice collaborative and natural, like a coding partner handing off work.
+- Be concise and factual — no filler or conversational commentary and avoid unnecessary repetition
+- Use present tense and active voice (e.g., “Runs tests” not “This will run tests”).
+- Keep descriptions self-contained; don’t refer to “above” or “below”.
+- Use parallel structure in lists for consistency.
+
+**Verbosity**
+- Final answer compactness rules (enforced):
+  - Tiny/small single-file change (≤ ~10 lines): 2–5 sentences or ≤3 bullets. No headings. 0–1 short snippet (≤3 lines) only if essential.
+  - Medium change (single area or a few files): ≤6 bullets or 6–10 sentences. At most 1–2 short snippets total (≤8 lines each).
+  - Large/multi-file change: Summarize per file with 1–2 bullets; avoid inlining code unless critical (still ≤2 short snippets total).
+  - Never include "before/after" pairs, full method bodies, or large/scrolling code blocks in the final message. Prefer referencing file/symbol names instead.
+
+**Don’t**
+
+- Don’t use literal words “bold” or “monospace” in the content.
+- Don’t nest bullets or create deep hierarchies.
+- Don’t output ANSI escape codes directly — the CLI renderer applies them.
+- Don’t cram unrelated keywords into a single bullet; split for clarity.
+- Don’t let keyword lists run long — wrap or reformat for scanability.
+
+Generally, ensure your final answers adapt their shape and depth to the request. For example, answers to code explanations should have a precise, structured explanation with code references that answer the question directly. For tasks with a simple implementation, lead with the outcome and supplement only with what’s needed for clarity. Larger changes can be presented as a logical walkthrough of your approach, grouping related steps, explaining rationale where it adds value, and highlighting next actions to accelerate the user. Your answers should provide the right level of detail while being easily scannable.
+
+For casual greetings, acknowledgements, or other one-off conversational messages that are not delivering substantive information or structured results, respond naturally without section headers or bullet formatting.
+
+# Tool Guidelines
+
+## Shell commands
+
+When using the shell, you must adhere to the following guidelines:
+
+- When searching for text or files, prefer using `rg` or `rg --files` respectively because `rg` is much faster than alternatives like `grep`. (If the `rg` command is not found, then use alternatives.)
+- Read files in chunks with a max chunk size of 250 lines. Do not use python scripts to attempt to output larger chunks of a file. Command line output will be truncated after 10 kilobytes or 256 lines of output, regardless of the command used.
+
+## apply_patch
+
+Use the `apply_patch` tool to edit files. Your patch language is a stripped‑down, file‑oriented diff format designed to be easy to parse and safe to apply. You can think of it as a high‑level envelope:
+
+*** Begin Patch
+[ one or more file sections ]
+*** End Patch
+
+Within that envelope, you get a sequence of file operations.
+You MUST include a header to specify the action you are taking.
+Each operation starts with one of three headers:
+
+*** Add File: <path> - create a new file. Every following line is a + line (the initial contents).
+*** Delete File: <path> - remove an existing file. Nothing follows.
+*** Update File: <path> - patch an existing file in place (optionally with a rename).
+
+Example patch:
+
+```
+*** Begin Patch
+*** Add File: hello.txt
++Hello world
+*** Update File: src/app.py
+*** Move to: src/main.py
+@@ def greet():
+-print("Hi")
++print("Hello, world!")
+*** Delete File: obsolete.txt
+*** End Patch
+```
+
+It is important to remember:
+
+- You must include a header with your intended action (Add/Delete/Update)
+- You must prefix new lines with `+` even when creating a new file
+
+## `update_plan`
+
+A tool named `update_plan` is available to you. You can use it to keep an up‑to‑date, step‑by‑step plan for the task.
+
+To create a new plan, call `update_plan` with a short list of 1‑sentence steps (no more than 5-7 words each) with a `status` for each step (`pending`, `in_progress`, or `completed`).
+
+When steps have been completed, use `update_plan` to mark each finished step as `completed` and the next step you are working on as `in_progress`. There should always be exactly one `in_progress` step until everything is done. You can mark multiple items as complete in a single `update_plan` call.
+
+If all steps are complete, ensure you call `update_plan` to mark all steps as `completed`.

internal/misc/codex_instructions/gpt_5_codex_prompt.md-008-daf77b845230c35c325500ff73fe72a78f3b7416

@@ -0,0 +1,105 @@
+You are Codex, based on GPT-5. You are running as a coding agent in the Codex CLI on a user's computer.
+
+## General
+
+- When searching for text or files, prefer using `rg` or `rg --files` respectively because `rg` is much faster than alternatives like `grep`. (If the `rg` command is not found, then use alternatives.)
+
+## Editing constraints
+
+- Default to ASCII when editing or creating files. Only introduce non-ASCII or other Unicode characters when there is a clear justification and the file already uses them.
+- Add succinct code comments that explain what is going on if code is not self-explanatory. You should not add comments like "Assigns the value to the variable", but a brief comment might be useful ahead of a complex code block that the user would otherwise have to spend time parsing out. Usage of these comments should be rare.
+- Try to use apply_patch for single file edits, but it is fine to explore other options to make the edit if it does not work well. Do not use apply_patch for changes that are auto-generated (i.e. generating package.json or running a lint or format command like gofmt) or when scripting is more efficient (such as search and replacing a string across a codebase).
+- You may be in a dirty git worktree.
+    * NEVER revert existing changes you did not make unless explicitly requested, since these changes were made by the user.
+    * If asked to make a commit or code edits and there are unrelated changes to your work or changes that you didn't make in those files, don't revert those changes.
+    * If the changes are in files you've touched recently, you should read carefully and understand how you can work with the changes rather than reverting them.
+    * If the changes are in unrelated files, just ignore them and don't revert them.
+- Do not amend a commit unless explicitly requested to do so.
+- While you are working, you might notice unexpected changes that you didn't make. If this happens, STOP IMMEDIATELY and ask the user how they would like to proceed.
+- **NEVER** use destructive commands like `git reset --hard` or `git checkout --` unless specifically requested or approved by the user.
+
+## Plan tool
+
+When using the planning tool:
+- Skip using the planning tool for straightforward tasks (roughly the easiest 25%).
+- Do not make single-step plans.
+- When you made a plan, update it after having performed one of the sub-tasks that you shared on the plan.
+
+## Codex CLI harness, sandboxing, and approvals
+
+The Codex CLI harness supports several different configurations for sandboxing and escalation approvals that the user can choose from.
+
+Filesystem sandboxing defines which files can be read or written. The options for `sandbox_mode` are:
+- **read-only**: The sandbox only permits reading files.
+- **workspace-write**: The sandbox permits reading files, and editing files in `cwd` and `writable_roots`. Editing files in other directories requires approval.
+- **danger-full-access**: No filesystem sandboxing - all commands are permitted.
+
+Network sandboxing defines whether network can be accessed without approval. Options for `network_access` are:
+- **restricted**: Requires approval
+- **enabled**: No approval needed
+
+Approvals are your mechanism to get user consent to run shell commands without the sandbox. Possible configuration options for `approval_policy` are
+- **untrusted**: The harness will escalate most commands for user approval, apart from a limited allowlist of safe "read" commands.
+- **on-failure**: The harness will allow all commands to run in the sandbox (if enabled), and failures will be escalated to the user for approval to run again without the sandbox.
+- **on-request**: Commands will be run in the sandbox by default, and you can specify in your tool call if you want to escalate a command to run without sandboxing. (Note that this mode is not always available. If it is, you'll see parameters for it in the `shell` command description.)
+- **never**: This is a non-interactive mode where you may NEVER ask the user for approval to run commands. Instead, you must always persist and work around constraints to solve the task for the user. You MUST do your utmost best to finish the task and validate your work before yielding. If this mode is paired with `danger-full-access`, take advantage of it to deliver the best outcome for the user. Further, in this mode, your default testing philosophy is overridden: Even if you don't see local patterns for testing, you may add tests and scripts to validate your work. Just remove them before yielding.
+
+When you are running with `approval_policy == on-request`, and sandboxing enabled, here are scenarios where you'll need to request approval:
+- You need to run a command that writes to a directory that requires it (e.g. running tests that write to /var)
+- You need to run a GUI app (e.g., open/xdg-open/osascript) to open browsers or files.
+- You are running sandboxed and need to run a command that requires network access (e.g. installing packages)
+- If you run a command that is important to solving the user's query, but it fails because of sandboxing, rerun the command with approval. ALWAYS proceed to use the `with_escalated_permissions` and `justification` parameters - do not message the user before requesting approval for the command.
+- You are about to take a potentially destructive action such as an `rm` or `git reset` that the user did not explicitly ask for
+- (for all of these, you should weigh alternative paths that do not require approval)
+
+When `sandbox_mode` is set to read-only, you'll need to request approval for any command that isn't a read.
+
+You will be told what filesystem sandboxing, network sandboxing, and approval mode are active in a developer or user message. If you are not told about this, assume that you are running with workspace-write, network sandboxing enabled, and approval on-failure.
+
+Although they introduce friction to the user because your work is paused until the user responds, you should leverage them when necessary to accomplish important work. If the completing the task requires escalated permissions, Do not let these settings or the sandbox deter you from attempting to accomplish the user's task unless it is set to "never", in which case never ask for approvals.
+
+When requesting approval to execute a command that will require escalated privileges:
+  - Provide the `with_escalated_permissions` parameter with the boolean value true
+  - Include a short, 1 sentence explanation for why you need to enable `with_escalated_permissions` in the justification parameter
+
+## Special user requests
+
+- If the user makes a simple request (such as asking for the time) which you can fulfill by running a terminal command (such as `date`), you should do so.
+- If the user asks for a "review", default to a code review mindset: prioritise identifying bugs, risks, behavioural regressions, and missing tests. Findings must be the primary focus of the response - keep summaries or overviews brief and only after enumerating the issues. Present findings first (ordered by severity with file/line references), follow with open questions or assumptions, and offer a change-summary only as a secondary detail. If no findings are discovered, state that explicitly and mention any residual risks or testing gaps.
+
+## Presenting your work and final message
+
+You are producing plain text that will later be styled by the CLI. Follow these rules exactly. Formatting should make results easy to scan, but not feel mechanical. Use judgment to decide how much structure adds value.
+
+- Default: be very concise; friendly coding teammate tone.
+- Ask only when needed; suggest ideas; mirror the user's style.
+- For substantial work, summarize clearly; follow final‑answer formatting.
+- Skip heavy formatting for simple confirmations.
+- Don't dump large files you've written; reference paths only.
+- No "save/copy this file" - User is on the same machine.
+- Offer logical next steps (tests, commits, build) briefly; add verify steps if you couldn't do something.
+- For code changes:
+  * Lead with a quick explanation of the change, and then give more details on the context covering where and why a change was made. Do not start this explanation with "summary", just jump right in.
+  * If there are natural next steps the user may want to take, suggest them at the end of your response. Do not make suggestions if there are no natural next steps.
+  * When suggesting multiple options, use numeric lists for the suggestions so the user can quickly respond with a single number.
+- The user does not command execution outputs. When asked to show the output of a command (e.g. `git show`), relay the important details in your answer or summarize the key lines so the user understands the result.
+
+### Final answer structure and style guidelines
+
+- Plain text; CLI handles styling. Use structure only when it helps scanability.
+- Headers: optional; short Title Case (1-3 words) wrapped in **…**; no blank line before the first bullet; add only if they truly help.
+- Bullets: use - ; merge related points; keep to one line when possible; 4–6 per list ordered by importance; keep phrasing consistent.
+- Monospace: backticks for commands/paths/env vars/code ids and inline examples; use for literal keyword bullets; never combine with **.
+- Code samples or multi-line snippets should be wrapped in fenced code blocks; include an info string as often as possible.
+- Structure: group related bullets; order sections general → specific → supporting; for subsections, start with a bolded keyword bullet, then items; match complexity to the task.
+- Tone: collaborative, concise, factual; present tense, active voice; self‑contained; no "above/below"; parallel wording.
+- Don'ts: no nested bullets/hierarchies; no ANSI codes; don't cram unrelated keywords; keep keyword lists short—wrap/reformat if long; avoid naming formatting styles in answers.
+- Adaptation: code explanations → precise, structured with code refs; simple tasks → lead with outcome; big changes → logical walkthrough + rationale + next actions; casual one-offs → plain sentences, no headers/bullets.
+- File References: When referencing files in your response, make sure to include the relevant start line and always follow the below rules:
+  * Use inline code to make file paths clickable.
+  * Each reference should have a stand alone path. Even if it's the same file.
+  * Accepted: absolute, workspace‑relative, a/ or b/ diff prefixes, or bare filename/suffix.
+  * Line/column (1‑based, optional): :line[:column] or #Lline[Ccolumn] (column defaults to 1).
+  * Do not use URIs like file://, vscode://, or https://.
+  * Do not provide range of lines
+  * Examples: src/app.ts, src/app.ts:42, b/server/index.js#L10, C:\repo\project\main.rs:12:5

internal/registry/model_registry.go

@@ -63,6 +63,9 @@ type ThinkingSupport struct {
 	ZeroAllowed bool `json:"zero_allowed,omitempty"`
 	// DynamicAllowed indicates whether -1 is a valid value (dynamic thinking budget).
 	DynamicAllowed bool `json:"dynamic_allowed,omitempty"`
+	// Levels defines discrete reasoning effort levels (e.g., "low", "medium", "high").
+	// When set, the model uses level-based reasoning instead of token budgets.
+	Levels []string `json:"levels,omitempty"`
 }
 
 // ModelRegistration tracks a model's availability
@@ -826,7 +829,6 @@ func (r *ModelRegistry) CleanupExpiredQuotas() {
 	}
 }
 
-
 // GetFirstAvailableModel returns the first available model for the given handler type.
 // It prioritizes models by their creation timestamp (newest first) and checks if they have
 // available clients that are not suspended or over quota.

internal/runtime/executor/aistudio_executor.go

@@ -1,3 +1,6 @@
+// Package executor provides runtime execution capabilities for various AI service providers.
+// This file implements the AI Studio executor that routes requests through a websocket-backed
+// transport for the AI Studio provider.
 package executor
 
 import (
@@ -26,19 +29,28 @@ type AIStudioExecutor struct {
 	cfg      *config.Config
 }
 
-// NewAIStudioExecutor constructs a websocket executor for the provider name.
+// NewAIStudioExecutor creates a new AI Studio executor instance.
+//
+// Parameters:
+//   - cfg: The application configuration
+//   - provider: The provider name
+//   - relay: The websocket relay manager
+//
+// Returns:
+//   - *AIStudioExecutor: A new AI Studio executor instance
 func NewAIStudioExecutor(cfg *config.Config, provider string, relay *wsrelay.Manager) *AIStudioExecutor {
 	return &AIStudioExecutor{provider: strings.ToLower(provider), relay: relay, cfg: cfg}
 }
 
-// Identifier returns the logical provider key for routing.
+// Identifier returns the executor identifier.
 func (e *AIStudioExecutor) Identifier() string { return "aistudio" }
 
-// PrepareRequest is a no-op because websocket transport already injects headers.
+// PrepareRequest prepares the HTTP request for execution (no-op for AI Studio).
 func (e *AIStudioExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error {
 	return nil
 }
 
+// Execute performs a non-streaming request to the AI Studio API.
 func (e *AIStudioExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
@@ -92,6 +104,7 @@ func (e *AIStudioExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth,
 	return resp, nil
 }
 
+// ExecuteStream performs a streaming request to the AI Studio API.
 func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
@@ -129,18 +142,60 @@ func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth
 		recordAPIResponseError(ctx, e.cfg, err)
 		return nil, err
 	}
+	firstEvent, ok := <-wsStream
+	if !ok {
+		err = fmt.Errorf("wsrelay: stream closed before start")
+		recordAPIResponseError(ctx, e.cfg, err)
+		return nil, err
+	}
+	if firstEvent.Status > 0 && firstEvent.Status != http.StatusOK {
+		metadataLogged := false
+		if firstEvent.Status > 0 {
+			recordAPIResponseMetadata(ctx, e.cfg, firstEvent.Status, firstEvent.Headers.Clone())
+			metadataLogged = true
+		}
+		var body bytes.Buffer
+		if len(firstEvent.Payload) > 0 {
+			appendAPIResponseChunk(ctx, e.cfg, bytes.Clone(firstEvent.Payload))
+			body.Write(firstEvent.Payload)
+		}
+		if firstEvent.Type == wsrelay.MessageTypeStreamEnd {
+			return nil, statusErr{code: firstEvent.Status, msg: body.String()}
+		}
+		for event := range wsStream {
+			if event.Err != nil {
+				recordAPIResponseError(ctx, e.cfg, event.Err)
+				if body.Len() == 0 {
+					body.WriteString(event.Err.Error())
+				}
+				break
+			}
+			if !metadataLogged && event.Status > 0 {
+				recordAPIResponseMetadata(ctx, e.cfg, event.Status, event.Headers.Clone())
+				metadataLogged = true
+			}
+			if len(event.Payload) > 0 {
+				appendAPIResponseChunk(ctx, e.cfg, bytes.Clone(event.Payload))
+				body.Write(event.Payload)
+			}
+			if event.Type == wsrelay.MessageTypeStreamEnd {
+				break
+			}
+		}
+		return nil, statusErr{code: firstEvent.Status, msg: body.String()}
+	}
 	out := make(chan cliproxyexecutor.StreamChunk)
 	stream = out
-	go func() {
+	go func(first wsrelay.StreamEvent) {
 		defer close(out)
 		var param any
 		metadataLogged := false
-		for event := range wsStream {
+		processEvent := func(event wsrelay.StreamEvent) bool {
 			if event.Err != nil {
 				recordAPIResponseError(ctx, e.cfg, event.Err)
 				reporter.publishFailure(ctx)
 				out <- cliproxyexecutor.StreamChunk{Err: fmt.Errorf("wsrelay: %v", event.Err)}
-				return
+				return false
 			}
 			switch event.Type {
 			case wsrelay.MessageTypeStreamStart:
@@ -151,7 +206,7 @@ func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth
 			case wsrelay.MessageTypeStreamChunk:
 				if len(event.Payload) > 0 {
 					appendAPIResponseChunk(ctx, e.cfg, bytes.Clone(event.Payload))
-					filtered := filterAIStudioUsageMetadata(event.Payload)
+					filtered := FilterSSEUsageMetadata(event.Payload)
 					if detail, ok := parseGeminiStreamUsage(filtered); ok {
 						reporter.publish(ctx, detail)
 					}
@@ -162,7 +217,7 @@ func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth
 					break
 				}
 			case wsrelay.MessageTypeStreamEnd:
-				return
+				return false
 			case wsrelay.MessageTypeHTTPResp:
 				if !metadataLogged && event.Status > 0 {
 					recordAPIResponseMetadata(ctx, e.cfg, event.Status, event.Headers.Clone())
@@ -176,18 +231,28 @@ func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth
 					out <- cliproxyexecutor.StreamChunk{Payload: ensureColonSpacedJSON([]byte(lines[i]))}
 				}
 				reporter.publish(ctx, parseGeminiUsage(event.Payload))
-				return
+				return false
 			case wsrelay.MessageTypeError:
 				recordAPIResponseError(ctx, e.cfg, event.Err)
 				reporter.publishFailure(ctx)
 				out <- cliproxyexecutor.StreamChunk{Err: fmt.Errorf("wsrelay: %v", event.Err)}
+				return false
+			}
+			return true
+		}
+		if !processEvent(first) {
+			return
+		}
+		for event := range wsStream {
+			if !processEvent(event) {
 				return
 			}
 		}
-	}()
+	}(firstEvent)
 	return stream, nil
 }
 
+// CountTokens counts tokens for the given request using the AI Studio API.
 func (e *AIStudioExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
 	_, body, err := e.translateRequest(req, opts, false)
 	if err != nil {
@@ -242,8 +307,8 @@ func (e *AIStudioExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.A
 	return cliproxyexecutor.Response{Payload: []byte(translated)}, nil
 }
 
-func (e *AIStudioExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
-	_ = ctx
+// Refresh refreshes the authentication credentials (no-op for AI Studio).
+func (e *AIStudioExecutor) Refresh(_ context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
 	return auth, nil
 }
 
@@ -257,16 +322,16 @@ func (e *AIStudioExecutor) translateRequest(req cliproxyexecutor.Request, opts c
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
 	payload := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), stream)
-	if budgetOverride, includeOverride, ok := util.GeminiThinkingFromMetadata(req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
-		if budgetOverride != nil {
-			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
-			budgetOverride = &norm
-		}
-		payload = util.ApplyGeminiThinkingConfig(payload, budgetOverride, includeOverride)
-	}
+	payload = applyThinkingMetadata(payload, req.Metadata, req.Model)
+	payload = util.ApplyDefaultThinkingIfNeeded(req.Model, payload)
+	payload = util.ConvertThinkingLevelToBudget(payload)
+	payload = util.NormalizeGeminiThinkingBudget(req.Model, payload)
 	payload = util.StripThinkingConfigIfUnsupported(req.Model, payload)
 	payload = fixGeminiImageAspectRatio(req.Model, payload)
 	payload = applyPayloadConfig(e.cfg, req.Model, payload)
+	payload, _ = sjson.DeleteBytes(payload, "generationConfig.maxOutputTokens")
+	payload, _ = sjson.DeleteBytes(payload, "generationConfig.responseMimeType")
+	payload, _ = sjson.DeleteBytes(payload, "generationConfig.responseJsonSchema")
 	metadataAction := "generateContent"
 	if req.Metadata != nil {
 		if action, _ := req.Metadata["action"].(string); action == "countTokens" {
@@ -295,65 +360,6 @@ func (e *AIStudioExecutor) buildEndpoint(model, action, alt string) string {
 	return base
 }
 
-// filterAIStudioUsageMetadata removes usageMetadata from intermediate SSE events so that
-// only the terminal chunk retains token statistics.
-func filterAIStudioUsageMetadata(payload []byte) []byte {
-	if len(payload) == 0 {
-		return payload
-	}
-
-	lines := bytes.Split(payload, []byte("\n"))
-	modified := false
-	for idx, line := range lines {
-		trimmed := bytes.TrimSpace(line)
-		if len(trimmed) == 0 || !bytes.HasPrefix(trimmed, []byte("data:")) {
-			continue
-		}
-		dataIdx := bytes.Index(line, []byte("data:"))
-		if dataIdx < 0 {
-			continue
-		}
-		rawJSON := bytes.TrimSpace(line[dataIdx+5:])
-		cleaned, changed := stripUsageMetadataFromJSON(rawJSON)
-		if !changed {
-			continue
-		}
-		var rebuilt []byte
-		rebuilt = append(rebuilt, line[:dataIdx]...)
-		rebuilt = append(rebuilt, []byte("data:")...)
-		if len(cleaned) > 0 {
-			rebuilt = append(rebuilt, ' ')
-			rebuilt = append(rebuilt, cleaned...)
-		}
-		lines[idx] = rebuilt
-		modified = true
-	}
-	if !modified {
-		return payload
-	}
-	return bytes.Join(lines, []byte("\n"))
-}
-
-// stripUsageMetadataFromJSON drops usageMetadata when no finishReason is present.
-func stripUsageMetadataFromJSON(rawJSON []byte) ([]byte, bool) {
-	jsonBytes := bytes.TrimSpace(rawJSON)
-	if len(jsonBytes) == 0 || !gjson.ValidBytes(jsonBytes) {
-		return rawJSON, false
-	}
-	finishReason := gjson.GetBytes(jsonBytes, "candidates.0.finishReason")
-	if finishReason.Exists() && finishReason.String() != "" {
-		return rawJSON, false
-	}
-	if !gjson.GetBytes(jsonBytes, "usageMetadata").Exists() {
-		return rawJSON, false
-	}
-	cleaned, err := sjson.DeleteBytes(jsonBytes, "usageMetadata")
-	if err != nil {
-		return rawJSON, false
-	}
-	return cleaned, true
-}
-
 // ensureColonSpacedJSON normalizes JSON objects so that colons are followed by a single space while
 // keeping the payload otherwise compact. Non-JSON inputs are returned unchanged.
 func ensureColonSpacedJSON(payload []byte) []byte {

internal/runtime/executor/antigravity_executor.go

@@ -0,0 +1,893 @@
+// Package executor provides runtime execution capabilities for various AI service providers.
+// This file implements the Antigravity executor that proxies requests to the antigravity
+// upstream using OAuth credentials.
+package executor
+
+import (
+	"bufio"
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"math/rand"
+	"net/http"
+	"net/url"
+	"strconv"
+	"strings"
+	"time"
+
+	"github.com/google/uuid"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
+	cliproxyauth "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/auth"
+	cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/executor"
+	sdktranslator "github.com/router-for-me/CLIProxyAPI/v6/sdk/translator"
+	log "github.com/sirupsen/logrus"
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+const (
+	antigravityBaseURLDaily = "https://daily-cloudcode-pa.sandbox.googleapis.com"
+	// antigravityBaseURLAutopush     = "https://autopush-cloudcode-pa.sandbox.googleapis.com"
+	antigravityBaseURLProd  = "https://cloudcode-pa.googleapis.com"
+	antigravityStreamPath   = "/v1internal:streamGenerateContent"
+	antigravityGeneratePath = "/v1internal:generateContent"
+	antigravityModelsPath   = "/v1internal:fetchAvailableModels"
+	antigravityClientID     = "1071006060591-tmhssin2h21lcre235vtolojh4g403ep.apps.googleusercontent.com"
+	antigravityClientSecret = "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"
+	defaultAntigravityAgent = "antigravity/1.11.5 windows/amd64"
+	antigravityAuthType     = "antigravity"
+	refreshSkew             = 3000 * time.Second
+)
+
+var randSource = rand.New(rand.NewSource(time.Now().UnixNano()))
+
+// AntigravityExecutor proxies requests to the antigravity upstream.
+type AntigravityExecutor struct {
+	cfg *config.Config
+}
+
+// NewAntigravityExecutor creates a new Antigravity executor instance.
+//
+// Parameters:
+//   - cfg: The application configuration
+//
+// Returns:
+//   - *AntigravityExecutor: A new Antigravity executor instance
+func NewAntigravityExecutor(cfg *config.Config) *AntigravityExecutor {
+	return &AntigravityExecutor{cfg: cfg}
+}
+
+// Identifier returns the executor identifier.
+func (e *AntigravityExecutor) Identifier() string { return antigravityAuthType }
+
+// PrepareRequest prepares the HTTP request for execution (no-op for Antigravity).
+func (e *AntigravityExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error { return nil }
+
+// Execute performs a non-streaming request to the Antigravity API.
+func (e *AntigravityExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
+	token, updatedAuth, errToken := e.ensureAccessToken(ctx, auth)
+	if errToken != nil {
+		return resp, errToken
+	}
+	if updatedAuth != nil {
+		auth = updatedAuth
+	}
+
+	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
+	defer reporter.trackFailure(ctx, &err)
+
+	from := opts.SourceFormat
+	to := sdktranslator.FromString("antigravity")
+	translated := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
+
+	translated = applyThinkingMetadataCLI(translated, req.Metadata, req.Model)
+	translated = util.ApplyDefaultThinkingIfNeededCLI(req.Model, translated)
+	translated = normalizeAntigravityThinking(req.Model, translated)
+
+	baseURLs := antigravityBaseURLFallbackOrder(auth)
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+
+	var lastStatus int
+	var lastBody []byte
+	var lastErr error
+
+	for idx, baseURL := range baseURLs {
+		httpReq, errReq := e.buildRequest(ctx, auth, token, req.Model, translated, false, opts.Alt, baseURL)
+		if errReq != nil {
+			err = errReq
+			return resp, err
+		}
+
+		httpResp, errDo := httpClient.Do(httpReq)
+		if errDo != nil {
+			recordAPIResponseError(ctx, e.cfg, errDo)
+			lastStatus = 0
+			lastBody = nil
+			lastErr = errDo
+			if idx+1 < len(baseURLs) {
+				log.Debugf("antigravity executor: request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+				continue
+			}
+			err = errDo
+			return resp, err
+		}
+
+		recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
+		bodyBytes, errRead := io.ReadAll(httpResp.Body)
+		if errClose := httpResp.Body.Close(); errClose != nil {
+			log.Errorf("antigravity executor: close response body error: %v", errClose)
+		}
+		if errRead != nil {
+			recordAPIResponseError(ctx, e.cfg, errRead)
+			err = errRead
+			return resp, err
+		}
+		appendAPIResponseChunk(ctx, e.cfg, bodyBytes)
+
+		if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
+			log.Debugf("antigravity executor: upstream error status: %d, body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), bodyBytes))
+			lastStatus = httpResp.StatusCode
+			lastBody = append([]byte(nil), bodyBytes...)
+			lastErr = nil
+			if httpResp.StatusCode == http.StatusTooManyRequests && idx+1 < len(baseURLs) {
+				log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+				continue
+			}
+			err = statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+			return resp, err
+		}
+
+		reporter.publish(ctx, parseAntigravityUsage(bodyBytes))
+		var param any
+		converted := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, bodyBytes, &param)
+		resp = cliproxyexecutor.Response{Payload: []byte(converted)}
+		reporter.ensurePublished(ctx)
+		return resp, nil
+	}
+
+	switch {
+	case lastStatus != 0:
+		err = statusErr{code: lastStatus, msg: string(lastBody)}
+	case lastErr != nil:
+		err = lastErr
+	default:
+		err = statusErr{code: http.StatusServiceUnavailable, msg: "antigravity executor: no base url available"}
+	}
+	return resp, err
+}
+
+// ExecuteStream performs a streaming request to the Antigravity API.
+func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+	ctx = context.WithValue(ctx, "alt", "")
+
+	token, updatedAuth, errToken := e.ensureAccessToken(ctx, auth)
+	if errToken != nil {
+		return nil, errToken
+	}
+	if updatedAuth != nil {
+		auth = updatedAuth
+	}
+
+	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
+	defer reporter.trackFailure(ctx, &err)
+
+	from := opts.SourceFormat
+	to := sdktranslator.FromString("antigravity")
+	translated := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
+
+	translated = applyThinkingMetadataCLI(translated, req.Metadata, req.Model)
+	translated = util.ApplyDefaultThinkingIfNeededCLI(req.Model, translated)
+	translated = normalizeAntigravityThinking(req.Model, translated)
+
+	baseURLs := antigravityBaseURLFallbackOrder(auth)
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+
+	var lastStatus int
+	var lastBody []byte
+	var lastErr error
+
+	for idx, baseURL := range baseURLs {
+		httpReq, errReq := e.buildRequest(ctx, auth, token, req.Model, translated, true, opts.Alt, baseURL)
+		if errReq != nil {
+			err = errReq
+			return nil, err
+		}
+
+		httpResp, errDo := httpClient.Do(httpReq)
+		if errDo != nil {
+			recordAPIResponseError(ctx, e.cfg, errDo)
+			lastStatus = 0
+			lastBody = nil
+			lastErr = errDo
+			if idx+1 < len(baseURLs) {
+				log.Debugf("antigravity executor: request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+				continue
+			}
+			err = errDo
+			return nil, err
+		}
+		recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
+		if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
+			bodyBytes, errRead := io.ReadAll(httpResp.Body)
+			if errClose := httpResp.Body.Close(); errClose != nil {
+				log.Errorf("antigravity executor: close response body error: %v", errClose)
+			}
+			if errRead != nil {
+				recordAPIResponseError(ctx, e.cfg, errRead)
+				lastStatus = 0
+				lastBody = nil
+				lastErr = errRead
+				if idx+1 < len(baseURLs) {
+					log.Debugf("antigravity executor: read error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+					continue
+				}
+				err = errRead
+				return nil, err
+			}
+			appendAPIResponseChunk(ctx, e.cfg, bodyBytes)
+			lastStatus = httpResp.StatusCode
+			lastBody = append([]byte(nil), bodyBytes...)
+			lastErr = nil
+			if httpResp.StatusCode == http.StatusTooManyRequests && idx+1 < len(baseURLs) {
+				log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+				continue
+			}
+			err = statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+			return nil, err
+		}
+
+		out := make(chan cliproxyexecutor.StreamChunk)
+		stream = out
+		go func(resp *http.Response) {
+			defer close(out)
+			defer func() {
+				if errClose := resp.Body.Close(); errClose != nil {
+					log.Errorf("antigravity executor: close response body error: %v", errClose)
+				}
+			}()
+			scanner := bufio.NewScanner(resp.Body)
+			scanner.Buffer(nil, streamScannerBuffer)
+			var param any
+			for scanner.Scan() {
+				line := scanner.Bytes()
+				appendAPIResponseChunk(ctx, e.cfg, line)
+
+				// Filter usage metadata for all models
+				// Only retain usage statistics in the terminal chunk
+				line = FilterSSEUsageMetadata(line)
+
+				payload := jsonPayload(line)
+				if payload == nil {
+					continue
+				}
+
+				if detail, ok := parseAntigravityStreamUsage(payload); ok {
+					reporter.publish(ctx, detail)
+				}
+
+				chunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, bytes.Clone(payload), &param)
+				for i := range chunks {
+					out <- cliproxyexecutor.StreamChunk{Payload: []byte(chunks[i])}
+				}
+			}
+			tail := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, []byte("[DONE]"), &param)
+			for i := range tail {
+				out <- cliproxyexecutor.StreamChunk{Payload: []byte(tail[i])}
+			}
+			if errScan := scanner.Err(); errScan != nil {
+				recordAPIResponseError(ctx, e.cfg, errScan)
+				reporter.publishFailure(ctx)
+				out <- cliproxyexecutor.StreamChunk{Err: errScan}
+			} else {
+				reporter.ensurePublished(ctx)
+			}
+		}(httpResp)
+		return stream, nil
+	}
+
+	switch {
+	case lastStatus != 0:
+		err = statusErr{code: lastStatus, msg: string(lastBody)}
+	case lastErr != nil:
+		err = lastErr
+	default:
+		err = statusErr{code: http.StatusServiceUnavailable, msg: "antigravity executor: no base url available"}
+	}
+	return nil, err
+}
+
+// Refresh refreshes the authentication credentials using the refresh token.
+func (e *AntigravityExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
+	if auth == nil {
+		return auth, nil
+	}
+	updated, errRefresh := e.refreshToken(ctx, auth.Clone())
+	if errRefresh != nil {
+		return nil, errRefresh
+	}
+	return updated, nil
+}
+
+// CountTokens counts tokens for the given request (not supported for Antigravity).
+func (e *AntigravityExecutor) CountTokens(context.Context, *cliproxyauth.Auth, cliproxyexecutor.Request, cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
+	return cliproxyexecutor.Response{}, statusErr{code: http.StatusNotImplemented, msg: "count tokens not supported"}
+}
+
+// FetchAntigravityModels retrieves available models using the supplied auth.
+func FetchAntigravityModels(ctx context.Context, auth *cliproxyauth.Auth, cfg *config.Config) []*registry.ModelInfo {
+	exec := &AntigravityExecutor{cfg: cfg}
+	token, updatedAuth, errToken := exec.ensureAccessToken(ctx, auth)
+	if errToken != nil || token == "" {
+		return nil
+	}
+	if updatedAuth != nil {
+		auth = updatedAuth
+	}
+
+	baseURLs := antigravityBaseURLFallbackOrder(auth)
+	httpClient := newProxyAwareHTTPClient(ctx, cfg, auth, 0)
+
+	for idx, baseURL := range baseURLs {
+		modelsURL := baseURL + antigravityModelsPath
+		httpReq, errReq := http.NewRequestWithContext(ctx, http.MethodPost, modelsURL, bytes.NewReader([]byte(`{}`)))
+		if errReq != nil {
+			return nil
+		}
+		httpReq.Header.Set("Content-Type", "application/json")
+		httpReq.Header.Set("Authorization", "Bearer "+token)
+		httpReq.Header.Set("User-Agent", resolveUserAgent(auth))
+		if host := resolveHost(baseURL); host != "" {
+			httpReq.Host = host
+		}
+
+		httpResp, errDo := httpClient.Do(httpReq)
+		if errDo != nil {
+			if idx+1 < len(baseURLs) {
+				log.Debugf("antigravity executor: models request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+				continue
+			}
+			return nil
+		}
+
+		bodyBytes, errRead := io.ReadAll(httpResp.Body)
+		if errClose := httpResp.Body.Close(); errClose != nil {
+			log.Errorf("antigravity executor: close response body error: %v", errClose)
+		}
+		if errRead != nil {
+			if idx+1 < len(baseURLs) {
+				log.Debugf("antigravity executor: models read error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+				continue
+			}
+			return nil
+		}
+		if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
+			if httpResp.StatusCode == http.StatusTooManyRequests && idx+1 < len(baseURLs) {
+				log.Debugf("antigravity executor: models request rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+				continue
+			}
+			return nil
+		}
+
+		result := gjson.GetBytes(bodyBytes, "models")
+		if !result.Exists() {
+			return nil
+		}
+
+		now := time.Now().Unix()
+		modelConfig := registry.GetAntigravityModelConfig()
+		models := make([]*registry.ModelInfo, 0, len(result.Map()))
+		for originalName := range result.Map() {
+			aliasName := modelName2Alias(originalName)
+			if aliasName != "" {
+				cfg := modelConfig[aliasName]
+				modelName := aliasName
+				if cfg != nil && cfg.Name != "" {
+					modelName = cfg.Name
+				}
+				modelInfo := &registry.ModelInfo{
+					ID:          aliasName,
+					Name:        modelName,
+					Description: aliasName,
+					DisplayName: aliasName,
+					Version:     aliasName,
+					Object:      "model",
+					Created:     now,
+					OwnedBy:     antigravityAuthType,
+					Type:        antigravityAuthType,
+				}
+				// Look up Thinking support from static config using alias name
+				if cfg != nil {
+					if cfg.Thinking != nil {
+						modelInfo.Thinking = cfg.Thinking
+					}
+					if cfg.MaxCompletionTokens > 0 {
+						modelInfo.MaxCompletionTokens = cfg.MaxCompletionTokens
+					}
+				}
+				models = append(models, modelInfo)
+			}
+		}
+		return models
+	}
+	return nil
+}
+
+func (e *AntigravityExecutor) ensureAccessToken(ctx context.Context, auth *cliproxyauth.Auth) (string, *cliproxyauth.Auth, error) {
+	if auth == nil {
+		return "", nil, statusErr{code: http.StatusUnauthorized, msg: "missing auth"}
+	}
+	accessToken := metaStringValue(auth.Metadata, "access_token")
+	expiry := tokenExpiry(auth.Metadata)
+	if accessToken != "" && expiry.After(time.Now().Add(refreshSkew)) {
+		return accessToken, nil, nil
+	}
+	updated, errRefresh := e.refreshToken(ctx, auth.Clone())
+	if errRefresh != nil {
+		return "", nil, errRefresh
+	}
+	return metaStringValue(updated.Metadata, "access_token"), updated, nil
+}
+
+func (e *AntigravityExecutor) refreshToken(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
+	if auth == nil {
+		return nil, statusErr{code: http.StatusUnauthorized, msg: "missing auth"}
+	}
+	refreshToken := metaStringValue(auth.Metadata, "refresh_token")
+	if refreshToken == "" {
+		return auth, statusErr{code: http.StatusUnauthorized, msg: "missing refresh token"}
+	}
+
+	form := url.Values{}
+	form.Set("client_id", antigravityClientID)
+	form.Set("client_secret", antigravityClientSecret)
+	form.Set("grant_type", "refresh_token")
+	form.Set("refresh_token", refreshToken)
+
+	httpReq, errReq := http.NewRequestWithContext(ctx, http.MethodPost, "https://oauth2.googleapis.com/token", strings.NewReader(form.Encode()))
+	if errReq != nil {
+		return auth, errReq
+	}
+	httpReq.Header.Set("Host", "oauth2.googleapis.com")
+	httpReq.Header.Set("User-Agent", defaultAntigravityAgent)
+	httpReq.Header.Set("Content-Type", "application/x-www-form-urlencoded")
+
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	httpResp, errDo := httpClient.Do(httpReq)
+	if errDo != nil {
+		return auth, errDo
+	}
+	defer func() {
+		if errClose := httpResp.Body.Close(); errClose != nil {
+			log.Errorf("antigravity executor: close response body error: %v", errClose)
+		}
+	}()
+
+	bodyBytes, errRead := io.ReadAll(httpResp.Body)
+	if errRead != nil {
+		return auth, errRead
+	}
+
+	if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
+		return auth, statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+	}
+
+	var tokenResp struct {
+		AccessToken  string `json:"access_token"`
+		RefreshToken string `json:"refresh_token"`
+		ExpiresIn    int64  `json:"expires_in"`
+		TokenType    string `json:"token_type"`
+	}
+	if errUnmarshal := json.Unmarshal(bodyBytes, &tokenResp); errUnmarshal != nil {
+		return auth, errUnmarshal
+	}
+
+	if auth.Metadata == nil {
+		auth.Metadata = make(map[string]any)
+	}
+	auth.Metadata["access_token"] = tokenResp.AccessToken
+	if tokenResp.RefreshToken != "" {
+		auth.Metadata["refresh_token"] = tokenResp.RefreshToken
+	}
+	auth.Metadata["expires_in"] = tokenResp.ExpiresIn
+	auth.Metadata["timestamp"] = time.Now().UnixMilli()
+	auth.Metadata["expired"] = time.Now().Add(time.Duration(tokenResp.ExpiresIn) * time.Second).Format(time.RFC3339)
+	auth.Metadata["type"] = antigravityAuthType
+	return auth, nil
+}
+
+func (e *AntigravityExecutor) buildRequest(ctx context.Context, auth *cliproxyauth.Auth, token, modelName string, payload []byte, stream bool, alt, baseURL string) (*http.Request, error) {
+	if token == "" {
+		return nil, statusErr{code: http.StatusUnauthorized, msg: "missing access token"}
+	}
+
+	base := strings.TrimSuffix(baseURL, "/")
+	if base == "" {
+		base = buildBaseURL(auth)
+	}
+	path := antigravityGeneratePath
+	if stream {
+		path = antigravityStreamPath
+	}
+	var requestURL strings.Builder
+	requestURL.WriteString(base)
+	requestURL.WriteString(path)
+	if stream {
+		if alt != "" {
+			requestURL.WriteString("?$alt=")
+			requestURL.WriteString(url.QueryEscape(alt))
+		} else {
+			requestURL.WriteString("?alt=sse")
+		}
+	} else if alt != "" {
+		requestURL.WriteString("?$alt=")
+		requestURL.WriteString(url.QueryEscape(alt))
+	}
+
+	// Extract project_id from auth metadata if available
+	projectID := ""
+	if auth != nil && auth.Metadata != nil {
+		if pid, ok := auth.Metadata["project_id"].(string); ok {
+			projectID = strings.TrimSpace(pid)
+		}
+	}
+	payload = geminiToAntigravity(modelName, payload, projectID)
+	payload, _ = sjson.SetBytes(payload, "model", alias2ModelName(modelName))
+
+	if strings.Contains(modelName, "claude") {
+		strJSON := string(payload)
+		paths := make([]string, 0)
+		util.Walk(gjson.ParseBytes(payload), "", "parametersJsonSchema", &paths)
+		for _, p := range paths {
+			strJSON, _ = util.RenameKey(strJSON, p, p[:len(p)-len("parametersJsonSchema")]+"parameters")
+		}
+
+		strJSON = util.DeleteKey(strJSON, "$schema")
+		strJSON = util.DeleteKey(strJSON, "maxItems")
+		strJSON = util.DeleteKey(strJSON, "minItems")
+		strJSON = util.DeleteKey(strJSON, "minLength")
+		strJSON = util.DeleteKey(strJSON, "maxLength")
+		strJSON = util.DeleteKey(strJSON, "exclusiveMinimum")
+		strJSON = util.DeleteKey(strJSON, "exclusiveMaximum")
+		strJSON = util.DeleteKey(strJSON, "$ref")
+		strJSON = util.DeleteKey(strJSON, "$defs")
+
+		paths = make([]string, 0)
+		util.Walk(gjson.Parse(strJSON), "", "anyOf", &paths)
+		for _, p := range paths {
+			anyOf := gjson.Get(strJSON, p)
+			if anyOf.IsArray() {
+				anyOfItems := anyOf.Array()
+				if len(anyOfItems) > 0 {
+					strJSON, _ = sjson.SetRaw(strJSON, p[:len(p)-len(".anyOf")], anyOfItems[0].Raw)
+				}
+			}
+		}
+
+		payload = []byte(strJSON)
+	}
+
+	httpReq, errReq := http.NewRequestWithContext(ctx, http.MethodPost, requestURL.String(), bytes.NewReader(payload))
+	if errReq != nil {
+		return nil, errReq
+	}
+	httpReq.Header.Set("Content-Type", "application/json")
+	httpReq.Header.Set("Authorization", "Bearer "+token)
+	httpReq.Header.Set("User-Agent", resolveUserAgent(auth))
+	if stream {
+		httpReq.Header.Set("Accept", "text/event-stream")
+	} else {
+		httpReq.Header.Set("Accept", "application/json")
+	}
+	if host := resolveHost(base); host != "" {
+		httpReq.Host = host
+	}
+
+	var authID, authLabel, authType, authValue string
+	if auth != nil {
+		authID = auth.ID
+		authLabel = auth.Label
+		authType, authValue = auth.AccountInfo()
+	}
+	recordAPIRequest(ctx, e.cfg, upstreamRequestLog{
+		URL:       requestURL.String(),
+		Method:    http.MethodPost,
+		Headers:   httpReq.Header.Clone(),
+		Body:      payload,
+		Provider:  e.Identifier(),
+		AuthID:    authID,
+		AuthLabel: authLabel,
+		AuthType:  authType,
+		AuthValue: authValue,
+	})
+
+	return httpReq, nil
+}
+
+func tokenExpiry(metadata map[string]any) time.Time {
+	if metadata == nil {
+		return time.Time{}
+	}
+	if expStr, ok := metadata["expired"].(string); ok {
+		expStr = strings.TrimSpace(expStr)
+		if expStr != "" {
+			if parsed, errParse := time.Parse(time.RFC3339, expStr); errParse == nil {
+				return parsed
+			}
+		}
+	}
+	expiresIn, hasExpires := int64Value(metadata["expires_in"])
+	tsMs, hasTimestamp := int64Value(metadata["timestamp"])
+	if hasExpires && hasTimestamp {
+		return time.Unix(0, tsMs*int64(time.Millisecond)).Add(time.Duration(expiresIn) * time.Second)
+	}
+	return time.Time{}
+}
+
+func metaStringValue(metadata map[string]any, key string) string {
+	if metadata == nil {
+		return ""
+	}
+	if v, ok := metadata[key]; ok {
+		switch typed := v.(type) {
+		case string:
+			return strings.TrimSpace(typed)
+		case []byte:
+			return strings.TrimSpace(string(typed))
+		}
+	}
+	return ""
+}
+
+func int64Value(value any) (int64, bool) {
+	switch typed := value.(type) {
+	case int:
+		return int64(typed), true
+	case int64:
+		return typed, true
+	case float64:
+		return int64(typed), true
+	case json.Number:
+		if i, errParse := typed.Int64(); errParse == nil {
+			return i, true
+		}
+	case string:
+		if strings.TrimSpace(typed) == "" {
+			return 0, false
+		}
+		if i, errParse := strconv.ParseInt(strings.TrimSpace(typed), 10, 64); errParse == nil {
+			return i, true
+		}
+	}
+	return 0, false
+}
+
+func buildBaseURL(auth *cliproxyauth.Auth) string {
+	if baseURLs := antigravityBaseURLFallbackOrder(auth); len(baseURLs) > 0 {
+		return baseURLs[0]
+	}
+	return antigravityBaseURLDaily
+}
+
+func resolveHost(base string) string {
+	parsed, errParse := url.Parse(base)
+	if errParse != nil {
+		return ""
+	}
+	if parsed.Host != "" {
+		return parsed.Host
+	}
+	return strings.TrimPrefix(strings.TrimPrefix(base, "https://"), "http://")
+}
+
+func resolveUserAgent(auth *cliproxyauth.Auth) string {
+	if auth != nil {
+		if auth.Attributes != nil {
+			if ua := strings.TrimSpace(auth.Attributes["user_agent"]); ua != "" {
+				return ua
+			}
+		}
+		if auth.Metadata != nil {
+			if ua, ok := auth.Metadata["user_agent"].(string); ok && strings.TrimSpace(ua) != "" {
+				return strings.TrimSpace(ua)
+			}
+		}
+	}
+	return defaultAntigravityAgent
+}
+
+func antigravityBaseURLFallbackOrder(auth *cliproxyauth.Auth) []string {
+	if base := resolveCustomAntigravityBaseURL(auth); base != "" {
+		return []string{base}
+	}
+	return []string{
+		antigravityBaseURLDaily,
+		// antigravityBaseURLAutopush,
+		antigravityBaseURLProd,
+	}
+}
+
+func resolveCustomAntigravityBaseURL(auth *cliproxyauth.Auth) string {
+	if auth == nil {
+		return ""
+	}
+	if auth.Attributes != nil {
+		if v := strings.TrimSpace(auth.Attributes["base_url"]); v != "" {
+			return strings.TrimSuffix(v, "/")
+		}
+	}
+	if auth.Metadata != nil {
+		if v, ok := auth.Metadata["base_url"].(string); ok {
+			v = strings.TrimSpace(v)
+			if v != "" {
+				return strings.TrimSuffix(v, "/")
+			}
+		}
+	}
+	return ""
+}
+
+func geminiToAntigravity(modelName string, payload []byte, projectID string) []byte {
+	template, _ := sjson.Set(string(payload), "model", modelName)
+	template, _ = sjson.Set(template, "userAgent", "antigravity")
+
+	// Use real project ID from auth if available, otherwise generate random (legacy fallback)
+	if projectID != "" {
+		template, _ = sjson.Set(template, "project", projectID)
+	} else {
+		template, _ = sjson.Set(template, "project", generateProjectID())
+	}
+	template, _ = sjson.Set(template, "requestId", generateRequestID())
+	template, _ = sjson.Set(template, "request.sessionId", generateSessionID())
+
+	template, _ = sjson.Delete(template, "request.safetySettings")
+	template, _ = sjson.Set(template, "request.toolConfig.functionCallingConfig.mode", "VALIDATED")
+
+	if !strings.HasPrefix(modelName, "gemini-3-") {
+		if thinkingLevel := gjson.Get(template, "request.generationConfig.thinkingConfig.thinkingLevel"); thinkingLevel.Exists() {
+			template, _ = sjson.Delete(template, "request.generationConfig.thinkingConfig.thinkingLevel")
+			template, _ = sjson.Set(template, "request.generationConfig.thinkingConfig.thinkingBudget", -1)
+		}
+	}
+
+	if strings.Contains(modelName, "claude") {
+		gjson.Get(template, "request.tools").ForEach(func(key, tool gjson.Result) bool {
+			tool.Get("functionDeclarations").ForEach(func(funKey, funcDecl gjson.Result) bool {
+				if funcDecl.Get("parametersJsonSchema").Exists() {
+					template, _ = sjson.SetRaw(template, fmt.Sprintf("request.tools.%d.functionDeclarations.%d.parameters", key.Int(), funKey.Int()), funcDecl.Get("parametersJsonSchema").Raw)
+					template, _ = sjson.Delete(template, fmt.Sprintf("request.tools.%d.functionDeclarations.%d.parameters.$schema", key.Int(), funKey.Int()))
+					template, _ = sjson.Delete(template, fmt.Sprintf("request.tools.%d.functionDeclarations.%d.parametersJsonSchema", key.Int(), funKey.Int()))
+				}
+				return true
+			})
+			return true
+		})
+	} else {
+		template, _ = sjson.Delete(template, "request.generationConfig.maxOutputTokens")
+	}
+
+	return []byte(template)
+}
+
+func generateRequestID() string {
+	return "agent-" + uuid.NewString()
+}
+
+func generateSessionID() string {
+	n := randSource.Int63n(9_000_000_000_000_000_000)
+	return "-" + strconv.FormatInt(n, 10)
+}
+
+func generateProjectID() string {
+	adjectives := []string{"useful", "bright", "swift", "calm", "bold"}
+	nouns := []string{"fuze", "wave", "spark", "flow", "core"}
+	adj := adjectives[randSource.Intn(len(adjectives))]
+	noun := nouns[randSource.Intn(len(nouns))]
+	randomPart := strings.ToLower(uuid.NewString())[:5]
+	return adj + "-" + noun + "-" + randomPart
+}
+
+func modelName2Alias(modelName string) string {
+	switch modelName {
+	case "rev19-uic3-1p":
+		return "gemini-2.5-computer-use-preview-10-2025"
+	case "gemini-3-pro-image":
+		return "gemini-3-pro-image-preview"
+	case "gemini-3-pro-high":
+		return "gemini-3-pro-preview"
+	case "claude-sonnet-4-5":
+		return "gemini-claude-sonnet-4-5"
+	case "claude-sonnet-4-5-thinking":
+		return "gemini-claude-sonnet-4-5-thinking"
+	case "claude-opus-4-5-thinking":
+		return "gemini-claude-opus-4-5-thinking"
+	case "chat_20706", "chat_23310", "gemini-2.5-flash-thinking", "gemini-3-pro-low", "gemini-2.5-pro":
+		return ""
+	default:
+		return modelName
+	}
+}
+
+func alias2ModelName(modelName string) string {
+	switch modelName {
+	case "gemini-2.5-computer-use-preview-10-2025":
+		return "rev19-uic3-1p"
+	case "gemini-3-pro-image-preview":
+		return "gemini-3-pro-image"
+	case "gemini-3-pro-preview":
+		return "gemini-3-pro-high"
+	case "gemini-claude-sonnet-4-5":
+		return "claude-sonnet-4-5"
+	case "gemini-claude-sonnet-4-5-thinking":
+		return "claude-sonnet-4-5-thinking"
+	case "gemini-claude-opus-4-5-thinking":
+		return "claude-opus-4-5-thinking"
+	default:
+		return modelName
+	}
+}
+
+// normalizeAntigravityThinking clamps or removes thinking config based on model support.
+// For Claude models, it additionally ensures thinking budget < max_tokens.
+func normalizeAntigravityThinking(model string, payload []byte) []byte {
+	payload = util.StripThinkingConfigIfUnsupported(model, payload)
+	if !util.ModelSupportsThinking(model) {
+		return payload
+	}
+	budget := gjson.GetBytes(payload, "request.generationConfig.thinkingConfig.thinkingBudget")
+	if !budget.Exists() {
+		return payload
+	}
+	raw := int(budget.Int())
+	normalized := util.NormalizeThinkingBudget(model, raw)
+
+	isClaude := strings.Contains(strings.ToLower(model), "claude")
+	if isClaude {
+		effectiveMax, setDefaultMax := antigravityEffectiveMaxTokens(model, payload)
+		if effectiveMax > 0 && normalized >= effectiveMax {
+			normalized = effectiveMax - 1
+		}
+		minBudget := antigravityMinThinkingBudget(model)
+		if minBudget > 0 && normalized >= 0 && normalized < minBudget {
+			// Budget is below minimum, remove thinking config entirely
+			payload, _ = sjson.DeleteBytes(payload, "request.generationConfig.thinkingConfig")
+			return payload
+		}
+		if setDefaultMax {
+			if res, errSet := sjson.SetBytes(payload, "request.generationConfig.maxOutputTokens", effectiveMax); errSet == nil {
+				payload = res
+			}
+		}
+	}
+
+	updated, err := sjson.SetBytes(payload, "request.generationConfig.thinkingConfig.thinkingBudget", normalized)
+	if err != nil {
+		return payload
+	}
+	return updated
+}
+
+// antigravityEffectiveMaxTokens returns the max tokens to cap thinking:
+// prefer request-provided maxOutputTokens; otherwise fall back to model default.
+// The boolean indicates whether the value came from the model default (and thus should be written back).
+func antigravityEffectiveMaxTokens(model string, payload []byte) (max int, fromModel bool) {
+	if maxTok := gjson.GetBytes(payload, "request.generationConfig.maxOutputTokens"); maxTok.Exists() && maxTok.Int() > 0 {
+		return int(maxTok.Int()), false
+	}
+	if modelInfo := registry.GetGlobalRegistry().GetModelInfo(model); modelInfo != nil && modelInfo.MaxCompletionTokens > 0 {
+		return modelInfo.MaxCompletionTokens, true
+	}
+	return 0, false
+}
+
+// antigravityMinThinkingBudget returns the minimum thinking budget for a model.
+// Falls back to -1 if no model info is found.
+func antigravityMinThinkingBudget(model string) int {
+	if modelInfo := registry.GetGlobalRegistry().GetModelInfo(model); modelInfo != nil && modelInfo.Thinking != nil {
+		return modelInfo.Thinking.Min
+	}
+	return -1
+}

internal/runtime/executor/claude_executor.go

@@ -17,6 +17,7 @@ import (
 	claudeauth "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/claude"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/misc"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 	cliproxyauth "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/auth"
 	cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/executor"
@@ -53,23 +54,39 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	// Use streaming translation to preserve function calling, except for claude.
 	stream := from != to
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), stream)
-	modelForUpstream := req.Model
-	if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {
-		body, _ = sjson.SetBytes(body, "model", modelOverride)
-		modelForUpstream = modelOverride
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel == "" {
+		upstreamModel = req.Model
 	}
+	if modelOverride := e.resolveUpstreamModel(upstreamModel, auth); modelOverride != "" {
+		upstreamModel = modelOverride
+	} else if !strings.EqualFold(upstreamModel, req.Model) {
+		if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {
+			upstreamModel = modelOverride
+		}
+	}
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
+	// Inject thinking config based on model metadata for thinking variants
+	body = e.injectThinkingConfig(req.Model, req.Metadata, body)
 
-	if !strings.HasPrefix(modelForUpstream, "claude-3-5-haiku") {
-		body, _ = sjson.SetRawBytes(body, "system", []byte(misc.ClaudeCodeInstructions))
+	if !strings.HasPrefix(upstreamModel, "claude-3-5-haiku") {
+		body = checkSystemInstructions(body)
 	}
 	body = applyPayloadConfig(e.cfg, req.Model, body)
 
+	// Ensure max_tokens > thinking.budget_tokens when thinking is enabled
+	body = ensureMaxTokensForThinking(req.Model, body)
+
+	// Extract betas from body and convert to header
+	var extraBetas []string
+	extraBetas, body = extractAndRemoveBetas(body)
+
 	url := fmt.Sprintf("%s/v1/messages?beta=true", baseURL)
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
 	if err != nil {
 		return resp, err
 	}
-	applyClaudeHeaders(httpReq, auth, apiKey, false)
+	applyClaudeHeaders(httpReq, auth, apiKey, false, extraBetas)
 	var authID, authLabel, authType, authValue string
 	if auth != nil {
 		authID = auth.ID
@@ -151,18 +168,36 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("claude")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
-	if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {
-		body, _ = sjson.SetBytes(body, "model", modelOverride)
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel == "" {
+		upstreamModel = req.Model
 	}
-	body, _ = sjson.SetRawBytes(body, "system", []byte(misc.ClaudeCodeInstructions))
+	if modelOverride := e.resolveUpstreamModel(upstreamModel, auth); modelOverride != "" {
+		upstreamModel = modelOverride
+	} else if !strings.EqualFold(upstreamModel, req.Model) {
+		if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {
+			upstreamModel = modelOverride
+		}
+	}
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
+	// Inject thinking config based on model metadata for thinking variants
+	body = e.injectThinkingConfig(req.Model, req.Metadata, body)
+	body = checkSystemInstructions(body)
 	body = applyPayloadConfig(e.cfg, req.Model, body)
 
+	// Ensure max_tokens > thinking.budget_tokens when thinking is enabled
+	body = ensureMaxTokensForThinking(req.Model, body)
+
+	// Extract betas from body and convert to header
+	var extraBetas []string
+	extraBetas, body = extractAndRemoveBetas(body)
+
 	url := fmt.Sprintf("%s/v1/messages?beta=true", baseURL)
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
 	if err != nil {
 		return nil, err
 	}
-	applyClaudeHeaders(httpReq, auth, apiKey, true)
+	applyClaudeHeaders(httpReq, auth, apiKey, true, extraBetas)
 	var authID, authLabel, authType, authValue string
 	if auth != nil {
 		authID = auth.ID
@@ -219,8 +254,7 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 		// If from == to (Claude → Claude), directly forward the SSE stream without translation
 		if from == to {
 			scanner := bufio.NewScanner(decodedBody)
-			buf := make([]byte, 20_971_520)
-			scanner.Buffer(buf, 20_971_520)
+			scanner.Buffer(nil, 52_428_800) // 50MB
 			for scanner.Scan() {
 				line := scanner.Bytes()
 				appendAPIResponseChunk(ctx, e.cfg, line)
@@ -243,8 +277,7 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 
 		// For other formats, use translation
 		scanner := bufio.NewScanner(decodedBody)
-		buf := make([]byte, 20_971_520)
-		scanner.Buffer(buf, 20_971_520)
+		scanner.Buffer(nil, 52_428_800) // 50MB
 		var param any
 		for scanner.Scan() {
 			line := scanner.Bytes()
@@ -278,22 +311,33 @@ func (e *ClaudeExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
 	// Use streaming translation to preserve function calling, except for claude.
 	stream := from != to
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), stream)
-	modelForUpstream := req.Model
-	if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {
-		body, _ = sjson.SetBytes(body, "model", modelOverride)
-		modelForUpstream = modelOverride
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel == "" {
+		upstreamModel = req.Model
+	}
+	if modelOverride := e.resolveUpstreamModel(upstreamModel, auth); modelOverride != "" {
+		upstreamModel = modelOverride
+	} else if !strings.EqualFold(upstreamModel, req.Model) {
+		if modelOverride := e.resolveUpstreamModel(req.Model, auth); modelOverride != "" {
+			upstreamModel = modelOverride
+		}
+	}
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
+
+	if !strings.HasPrefix(upstreamModel, "claude-3-5-haiku") {
+		body = checkSystemInstructions(body)
 	}
 
-	if !strings.HasPrefix(modelForUpstream, "claude-3-5-haiku") {
-		body, _ = sjson.SetRawBytes(body, "system", []byte(misc.ClaudeCodeInstructions))
-	}
+	// Extract betas from body and convert to header (for count_tokens too)
+	var extraBetas []string
+	extraBetas, body = extractAndRemoveBetas(body)
 
 	url := fmt.Sprintf("%s/v1/messages/count_tokens?beta=true", baseURL)
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
 	if err != nil {
 		return cliproxyexecutor.Response{}, err
 	}
-	applyClaudeHeaders(httpReq, auth, apiKey, false)
+	applyClaudeHeaders(httpReq, auth, apiKey, false, extraBetas)
 	var authID, authLabel, authType, authValue string
 	if auth != nil {
 		authID = auth.ID
@@ -385,29 +429,114 @@ func (e *ClaudeExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (
 	return auth, nil
 }
 
+// extractAndRemoveBetas extracts the "betas" array from the body and removes it.
+// Returns the extracted betas as a string slice and the modified body.
+func extractAndRemoveBetas(body []byte) ([]string, []byte) {
+	betasResult := gjson.GetBytes(body, "betas")
+	if !betasResult.Exists() {
+		return nil, body
+	}
+	var betas []string
+	if betasResult.IsArray() {
+		for _, item := range betasResult.Array() {
+			if s := strings.TrimSpace(item.String()); s != "" {
+				betas = append(betas, s)
+			}
+		}
+	} else if s := strings.TrimSpace(betasResult.String()); s != "" {
+		betas = append(betas, s)
+	}
+	body, _ = sjson.DeleteBytes(body, "betas")
+	return betas, body
+}
+
+// injectThinkingConfig adds thinking configuration based on metadata using the unified flow.
+// It uses util.ResolveClaudeThinkingConfig which internally calls ResolveThinkingConfigFromMetadata
+// and NormalizeThinkingBudget, ensuring consistency with other executors like Gemini.
+func (e *ClaudeExecutor) injectThinkingConfig(modelName string, metadata map[string]any, body []byte) []byte {
+	budget, ok := util.ResolveClaudeThinkingConfig(modelName, metadata)
+	if !ok {
+		return body
+	}
+	return util.ApplyClaudeThinkingConfig(body, budget)
+}
+
+// ensureMaxTokensForThinking ensures max_tokens > thinking.budget_tokens when thinking is enabled.
+// Anthropic API requires this constraint; violating it returns a 400 error.
+// This function should be called after all thinking configuration is finalized.
+// It looks up the model's MaxCompletionTokens from the registry to use as the cap.
+func ensureMaxTokensForThinking(modelName string, body []byte) []byte {
+	thinkingType := gjson.GetBytes(body, "thinking.type").String()
+	if thinkingType != "enabled" {
+		return body
+	}
+
+	budgetTokens := gjson.GetBytes(body, "thinking.budget_tokens").Int()
+	if budgetTokens <= 0 {
+		return body
+	}
+
+	maxTokens := gjson.GetBytes(body, "max_tokens").Int()
+
+	// Look up the model's max completion tokens from the registry
+	maxCompletionTokens := 0
+	if modelInfo := registry.GetGlobalRegistry().GetModelInfo(modelName); modelInfo != nil {
+		maxCompletionTokens = modelInfo.MaxCompletionTokens
+	}
+
+	// Fall back to budget + buffer if registry lookup fails or returns 0
+	const fallbackBuffer = 4000
+	requiredMaxTokens := budgetTokens + fallbackBuffer
+	if maxCompletionTokens > 0 {
+		requiredMaxTokens = int64(maxCompletionTokens)
+	}
+
+	if maxTokens < requiredMaxTokens {
+		body, _ = sjson.SetBytes(body, "max_tokens", requiredMaxTokens)
+	}
+	return body
+}
+
 func (e *ClaudeExecutor) resolveUpstreamModel(alias string, auth *cliproxyauth.Auth) string {
-	if alias == "" {
+	trimmed := strings.TrimSpace(alias)
+	if trimmed == "" {
 		return ""
 	}
+
 	entry := e.resolveClaudeConfig(auth)
 	if entry == nil {
 		return ""
 	}
+
+	normalizedModel, metadata := util.NormalizeThinkingModel(trimmed)
+
+	// Candidate names to match against configured aliases/names.
+	candidates := []string{strings.TrimSpace(normalizedModel)}
+	if !strings.EqualFold(normalizedModel, trimmed) {
+		candidates = append(candidates, trimmed)
+	}
+	if original := util.ResolveOriginalModel(normalizedModel, metadata); original != "" && !strings.EqualFold(original, normalizedModel) {
+		candidates = append(candidates, original)
+	}
+
 	for i := range entry.Models {
 		model := entry.Models[i]
 		name := strings.TrimSpace(model.Name)
 		modelAlias := strings.TrimSpace(model.Alias)
-		if modelAlias != "" {
-			if strings.EqualFold(modelAlias, alias) {
+
+		for _, candidate := range candidates {
+			if candidate == "" {
+				continue
+			}
+			if modelAlias != "" && strings.EqualFold(modelAlias, candidate) {
 				if name != "" {
 					return name
 				}
-				return alias
+				return candidate
+			}
+			if name != "" && strings.EqualFold(name, candidate) {
+				return name
 			}
-			continue
-		}
-		if name != "" && strings.EqualFold(name, alias) {
-			return name
 		}
 	}
 	return ""
@@ -532,7 +661,7 @@ func decodeResponseBody(body io.ReadCloser, contentEncoding string) (io.ReadClos
 	return body, nil
 }
 
-func applyClaudeHeaders(r *http.Request, auth *cliproxyauth.Auth, apiKey string, stream bool) {
+func applyClaudeHeaders(r *http.Request, auth *cliproxyauth.Auth, apiKey string, stream bool, extraBetas []string) {
 	r.Header.Set("Authorization", "Bearer "+apiKey)
 	r.Header.Set("Content-Type", "application/json")
 
@@ -541,15 +670,30 @@ func applyClaudeHeaders(r *http.Request, auth *cliproxyauth.Auth, apiKey string,
 		ginHeaders = ginCtx.Request.Header
 	}
 
+	baseBetas := "claude-code-20250219,oauth-2025-04-20,interleaved-thinking-2025-05-14,fine-grained-tool-streaming-2025-05-14"
 	if val := strings.TrimSpace(ginHeaders.Get("Anthropic-Beta")); val != "" {
+		baseBetas = val
 		if !strings.Contains(val, "oauth") {
-			val += ",oauth-2025-04-20"
+			baseBetas += ",oauth-2025-04-20"
 		}
-		r.Header.Set("Anthropic-Beta", val)
-	} else {
-		r.Header.Set("Anthropic-Beta", "claude-code-20250219,oauth-2025-04-20,interleaved-thinking-2025-05-14,fine-grained-tool-streaming-2025-05-14")
 	}
 
+	// Merge extra betas from request body
+	if len(extraBetas) > 0 {
+		existingSet := make(map[string]bool)
+		for _, b := range strings.Split(baseBetas, ",") {
+			existingSet[strings.TrimSpace(b)] = true
+		}
+		for _, beta := range extraBetas {
+			beta = strings.TrimSpace(beta)
+			if beta != "" && !existingSet[beta] {
+				baseBetas += "," + beta
+				existingSet[beta] = true
+			}
+		}
+	}
+	r.Header.Set("Anthropic-Beta", baseBetas)
+
 	misc.EnsureHeader(r.Header, ginHeaders, "Anthropic-Version", "2023-06-01")
 	misc.EnsureHeader(r.Header, ginHeaders, "Anthropic-Dangerous-Direct-Browser-Access", "true")
 	misc.EnsureHeader(r.Header, ginHeaders, "X-App", "cli")
@@ -592,3 +736,22 @@ func claudeCreds(a *cliproxyauth.Auth) (apiKey, baseURL string) {
 	}
 	return
 }
+
+func checkSystemInstructions(payload []byte) []byte {
+	system := gjson.GetBytes(payload, "system")
+	claudeCodeInstructions := `[{"type":"text","text":"You are Claude Code, Anthropic's official CLI for Claude."}]`
+	if system.IsArray() {
+		if gjson.GetBytes(payload, "system.0.text").String() != "You are Claude Code, Anthropic's official CLI for Claude." {
+			system.ForEach(func(_, part gjson.Result) bool {
+				if part.Get("type").String() == "text" {
+					claudeCodeInstructions, _ = sjson.SetRaw(claudeCodeInstructions, "-1", part.Raw)
+				}
+				return true
+			})
+			payload, _ = sjson.SetRawBytes(payload, "system", []byte(claudeCodeInstructions))
+		}
+	} else {
+		payload, _ = sjson.SetRawBytes(payload, "system", []byte(claudeCodeInstructions))
+	}
+	return payload
+}

internal/runtime/executor/codex_executor.go

@@ -49,77 +49,18 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
 
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("codex")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
-
-	if util.InArray([]string{"gpt-5", "gpt-5-minimal", "gpt-5-low", "gpt-5-medium", "gpt-5-high"}, req.Model) {
-		body, _ = sjson.SetBytes(body, "model", "gpt-5")
-		switch req.Model {
-		case "gpt-5-minimal":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "minimal")
-		case "gpt-5-low":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "low")
-		case "gpt-5-medium":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		case "gpt-5-high":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "high")
-		}
-	} else if util.InArray([]string{"gpt-5-codex", "gpt-5-codex-low", "gpt-5-codex-medium", "gpt-5-codex-high"}, req.Model) {
-		body, _ = sjson.SetBytes(body, "model", "gpt-5-codex")
-		switch req.Model {
-		case "gpt-5-codex-low":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "low")
-		case "gpt-5-codex-medium":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		case "gpt-5-codex-high":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "high")
-		}
-	} else if util.InArray([]string{"gpt-5-codex-mini", "gpt-5-codex-mini-medium", "gpt-5-codex-mini-high"}, req.Model) {
-		body, _ = sjson.SetBytes(body, "model", "gpt-5-codex-mini")
-		switch req.Model {
-		case "gpt-5-codex-mini-medium":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		case "gpt-5-codex-mini-high":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "high")
-		default:
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		}
-	} else if util.InArray([]string{"gpt-5.1", "gpt-5.1-minimal", "gpt-5.1-low", "gpt-5.1-medium", "gpt-5.1-high"}, req.Model) {
-		body, _ = sjson.SetBytes(body, "model", "gpt-5.1")
-		switch req.Model {
-		case "gpt-5.1-minimal":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "minimal")
-		case "gpt-5.1-low":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "low")
-		case "gpt-5.1-medium":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		case "gpt-5.1-high":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "high")
-		}
-	} else if util.InArray([]string{"gpt-5.1-codex", "gpt-5.1-codex-low", "gpt-5.1-codex-medium", "gpt-5.1-codex-high"}, req.Model) {
-		body, _ = sjson.SetBytes(body, "model", "gpt-5.1-codex")
-		switch req.Model {
-		case "gpt-5.1-codex-low":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "low")
-		case "gpt-5.1-codex-medium":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		case "gpt-5.1-codex-high":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "high")
-		}
-	} else if util.InArray([]string{"gpt-5.1-codex-mini", "gpt-5.1-codex-mini-medium", "gpt-5.1-codex-mini-high"}, req.Model) {
-		body, _ = sjson.SetBytes(body, "model", "gpt-5.1-codex-mini")
-		switch req.Model {
-		case "gpt-5.1-codex-mini-medium":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		case "gpt-5.1-codex-mini-high":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "high")
-		default:
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		}
+	body = applyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning.effort")
+	body = normalizeThinkingConfig(body, upstreamModel)
+	if errValidate := validateThinkingConfig(body, upstreamModel); errValidate != nil {
+		return resp, errValidate
 	}
 	body = applyPayloadConfig(e.cfg, req.Model, body)
-
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 	body, _ = sjson.SetBytes(body, "stream", true)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
 
@@ -205,75 +146,20 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
 
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("codex")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 
-	if util.InArray([]string{"gpt-5", "gpt-5-minimal", "gpt-5-low", "gpt-5-medium", "gpt-5-high"}, req.Model) {
-		body, _ = sjson.SetBytes(body, "model", "gpt-5")
-		switch req.Model {
-		case "gpt-5-minimal":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "minimal")
-		case "gpt-5-low":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "low")
-		case "gpt-5-medium":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		case "gpt-5-high":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "high")
-		}
-	} else if util.InArray([]string{"gpt-5-codex", "gpt-5-codex-low", "gpt-5-codex-medium", "gpt-5-codex-high"}, req.Model) {
-		body, _ = sjson.SetBytes(body, "model", "gpt-5-codex")
-		switch req.Model {
-		case "gpt-5-codex-low":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "low")
-		case "gpt-5-codex-medium":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		case "gpt-5-codex-high":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "high")
-		}
-	} else if util.InArray([]string{"gpt-5-codex-mini", "gpt-5-codex-mini-medium", "gpt-5-codex-mini-high"}, req.Model) {
-		body, _ = sjson.SetBytes(body, "model", "gpt-5-codex-mini")
-		switch req.Model {
-		case "gpt-5-codex-mini-medium":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		case "gpt-5-codex-mini-high":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "high")
-		}
-	} else if util.InArray([]string{"gpt-5.1", "gpt-5.1-minimal", "gpt-5.1-low", "gpt-5.1-medium", "gpt-5.1-high"}, req.Model) {
-		body, _ = sjson.SetBytes(body, "model", "gpt-5.1")
-		switch req.Model {
-		case "gpt-5.1-minimal":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "minimal")
-		case "gpt-5.1-low":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "low")
-		case "gpt-5.1-medium":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		case "gpt-5.1-high":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "high")
-		}
-	} else if util.InArray([]string{"gpt-5.1-codex", "gpt-5.1-codex-low", "gpt-5.1-codex-medium", "gpt-5.1-codex-high"}, req.Model) {
-		body, _ = sjson.SetBytes(body, "model", "gpt-5.1-codex")
-		switch req.Model {
-		case "gpt-5.1-codex-low":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "low")
-		case "gpt-5.1-codex-medium":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		case "gpt-5.1-codex-high":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "high")
-		}
-	} else if util.InArray([]string{"gpt-5.1-codex-mini", "gpt-5.1-codex-mini-medium", "gpt-5.1-codex-mini-high"}, req.Model) {
-		body, _ = sjson.SetBytes(body, "model", "gpt-5.1-codex-mini")
-		switch req.Model {
-		case "gpt-5.1-codex-mini-medium":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		case "gpt-5.1-codex-mini-high":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "high")
-		default:
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		}
+	body = applyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning.effort")
+	body = normalizeThinkingConfig(body, upstreamModel)
+	if errValidate := validateThinkingConfig(body, upstreamModel); errValidate != nil {
+		return nil, errValidate
 	}
 	body = applyPayloadConfig(e.cfg, req.Model, body)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 
 	url := strings.TrimSuffix(baseURL, "/") + "/responses"
 	httpReq, err := e.cacheHelper(ctx, from, url, req, body)
@@ -330,8 +216,7 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 			}
 		}()
 		scanner := bufio.NewScanner(httpResp.Body)
-		buf := make([]byte, 20_971_520)
-		scanner.Buffer(buf, 20_971_520)
+		scanner.Buffer(nil, 52_428_800) // 50MB
 		var param any
 		for scanner.Scan() {
 			line := scanner.Bytes()
@@ -361,53 +246,16 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 }
 
 func (e *CodexExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("codex")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
 
 	modelForCounting := req.Model
 
-	if util.InArray([]string{"gpt-5", "gpt-5-minimal", "gpt-5-low", "gpt-5-medium", "gpt-5-high"}, req.Model) {
-		modelForCounting = "gpt-5"
-		body, _ = sjson.SetBytes(body, "model", "gpt-5")
-		switch req.Model {
-		case "gpt-5-minimal":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "minimal")
-		case "gpt-5-low":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "low")
-		case "gpt-5-medium":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		case "gpt-5-high":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "high")
-		default:
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "low")
-		}
-	} else if util.InArray([]string{"gpt-5-codex", "gpt-5-codex-low", "gpt-5-codex-medium", "gpt-5-codex-high"}, req.Model) {
-		modelForCounting = "gpt-5"
-		body, _ = sjson.SetBytes(body, "model", "gpt-5-codex")
-		switch req.Model {
-		case "gpt-5-codex-low":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "low")
-		case "gpt-5-codex-medium":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		case "gpt-5-codex-high":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "high")
-		default:
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "low")
-		}
-	} else if util.InArray([]string{"gpt-5-codex-mini", "gpt-5-codex-mini-medium", "gpt-5-codex-mini-high"}, req.Model) {
-		modelForCounting = "gpt-5"
-		body, _ = sjson.SetBytes(body, "model", "codex-mini-latest")
-		switch req.Model {
-		case "gpt-5-codex-mini-medium":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		case "gpt-5-codex-mini-high":
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "high")
-		default:
-			body, _ = sjson.SetBytes(body, "reasoning.effort", "medium")
-		}
-	}
-
+	body = applyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning.effort")
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
 	body, _ = sjson.SetBytes(body, "stream", false)
 

internal/runtime/executor/gemini_cli_executor.go

@@ -1,3 +1,6 @@
+// Package executor provides runtime execution capabilities for various AI service providers.
+// This file implements the Gemini CLI executor that talks to Cloud Code Assist endpoints
+// using OAuth credentials from auth metadata.
 package executor
 
 import (
@@ -29,11 +32,11 @@ import (
 const (
 	codeAssistEndpoint      = "https://cloudcode-pa.googleapis.com"
 	codeAssistVersion       = "v1internal"
-	geminiOauthClientID     = "681255809395-oo8ft2oprdrnp9e3aqf6av3hmdib135j.apps.googleusercontent.com"
-	geminiOauthClientSecret = "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"
+	geminiOAuthClientID     = "681255809395-oo8ft2oprdrnp9e3aqf6av3hmdib135j.apps.googleusercontent.com"
+	geminiOAuthClientSecret = "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"
 )
 
-var geminiOauthScopes = []string{
+var geminiOAuthScopes = []string{
 	"https://www.googleapis.com/auth/cloud-platform",
 	"https://www.googleapis.com/auth/userinfo.email",
 	"https://www.googleapis.com/auth/userinfo.profile",
@@ -44,14 +47,24 @@ type GeminiCLIExecutor struct {
 	cfg *config.Config
 }
 
+// NewGeminiCLIExecutor creates a new Gemini CLI executor instance.
+//
+// Parameters:
+//   - cfg: The application configuration
+//
+// Returns:
+//   - *GeminiCLIExecutor: A new Gemini CLI executor instance
 func NewGeminiCLIExecutor(cfg *config.Config) *GeminiCLIExecutor {
 	return &GeminiCLIExecutor{cfg: cfg}
 }
 
+// Identifier returns the executor identifier.
 func (e *GeminiCLIExecutor) Identifier() string { return "gemini-cli" }
 
+// PrepareRequest prepares the HTTP request for execution (no-op for Gemini CLI).
 func (e *GeminiCLIExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error { return nil }
 
+// Execute performs a non-streaming request to the Gemini CLI API.
 func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
 	tokenSource, baseTokenData, err := prepareGeminiCLITokenSource(ctx, e.cfg, auth)
 	if err != nil {
@@ -62,15 +75,10 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini-cli")
-	budgetOverride, includeOverride, hasOverride := util.GeminiThinkingFromMetadata(req.Metadata)
 	basePayload := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
-	if hasOverride && util.ModelSupportsThinking(req.Model) {
-		if budgetOverride != nil {
-			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
-			budgetOverride = &norm
-		}
-		basePayload = util.ApplyGeminiCLIThinkingConfig(basePayload, budgetOverride, includeOverride)
-	}
+	basePayload = applyThinkingMetadataCLI(basePayload, req.Metadata, req.Model)
+	basePayload = util.ApplyDefaultThinkingIfNeededCLI(req.Model, basePayload)
+	basePayload = util.NormalizeGeminiCLIThinkingBudget(req.Model, basePayload)
 	basePayload = util.StripThinkingConfigIfUnsupported(req.Model, basePayload)
 	basePayload = fixGeminiCLIImageAspectRatio(req.Model, basePayload)
 	basePayload = applyPayloadConfigWithRoot(e.cfg, req.Model, "gemini", "request", basePayload)
@@ -180,7 +188,7 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
 			continue
 		}
 
-		err = statusErr{code: httpResp.StatusCode, msg: string(data)}
+		err = newGeminiStatusErr(httpResp.StatusCode, data)
 		return resp, err
 	}
 
@@ -190,10 +198,11 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
 	if lastStatus == 0 {
 		lastStatus = 429
 	}
-	err = statusErr{code: lastStatus, msg: string(lastBody)}
+	err = newGeminiStatusErr(lastStatus, lastBody)
 	return resp, err
 }
 
+// ExecuteStream performs a streaming request to the Gemini CLI API.
 func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
 	tokenSource, baseTokenData, err := prepareGeminiCLITokenSource(ctx, e.cfg, auth)
 	if err != nil {
@@ -204,15 +213,10 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini-cli")
-	budgetOverride, includeOverride, hasOverride := util.GeminiThinkingFromMetadata(req.Metadata)
 	basePayload := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
-	if hasOverride && util.ModelSupportsThinking(req.Model) {
-		if budgetOverride != nil {
-			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
-			budgetOverride = &norm
-		}
-		basePayload = util.ApplyGeminiCLIThinkingConfig(basePayload, budgetOverride, includeOverride)
-	}
+	basePayload = applyThinkingMetadataCLI(basePayload, req.Metadata, req.Model)
+	basePayload = util.ApplyDefaultThinkingIfNeededCLI(req.Model, basePayload)
+	basePayload = util.NormalizeGeminiCLIThinkingBudget(req.Model, basePayload)
 	basePayload = util.StripThinkingConfigIfUnsupported(req.Model, basePayload)
 	basePayload = fixGeminiCLIImageAspectRatio(req.Model, basePayload)
 	basePayload = applyPayloadConfigWithRoot(e.cfg, req.Model, "gemini", "request", basePayload)
@@ -304,7 +308,7 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 				}
 				continue
 			}
-			err = statusErr{code: httpResp.StatusCode, msg: string(data)}
+			err = newGeminiStatusErr(httpResp.StatusCode, data)
 			return nil, err
 		}
 
@@ -319,8 +323,7 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 			}()
 			if opts.Alt == "" {
 				scanner := bufio.NewScanner(resp.Body)
-				buf := make([]byte, 20_971_520)
-				scanner.Buffer(buf, 20_971_520)
+				scanner.Buffer(nil, streamScannerBuffer)
 				var param any
 				for scanner.Scan() {
 					line := scanner.Bytes()
@@ -378,10 +381,11 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 	if lastStatus == 0 {
 		lastStatus = 429
 	}
-	err = statusErr{code: lastStatus, msg: string(lastBody)}
+	err = newGeminiStatusErr(lastStatus, lastBody)
 	return nil, err
 }
 
+// CountTokens counts tokens for the given request using the Gemini CLI API.
 func (e *GeminiCLIExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
 	tokenSource, baseTokenData, err := prepareGeminiCLITokenSource(ctx, e.cfg, auth)
 	if err != nil {
@@ -409,16 +413,9 @@ func (e *GeminiCLIExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.
 	var lastStatus int
 	var lastBody []byte
 
-	budgetOverride, includeOverride, hasOverride := util.GeminiThinkingFromMetadata(req.Metadata)
 	for _, attemptModel := range models {
 		payload := sdktranslator.TranslateRequest(from, to, attemptModel, bytes.Clone(req.Payload), false)
-		if hasOverride && util.ModelSupportsThinking(req.Model) {
-			if budgetOverride != nil {
-				norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
-				budgetOverride = &norm
-			}
-			payload = util.ApplyGeminiCLIThinkingConfig(payload, budgetOverride, includeOverride)
-		}
+		payload = applyThinkingMetadataCLI(payload, req.Metadata, req.Model)
 		payload = deleteJSONField(payload, "project")
 		payload = deleteJSONField(payload, "model")
 		payload = deleteJSONField(payload, "request.safetySettings")
@@ -486,12 +483,11 @@ func (e *GeminiCLIExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.
 	if lastStatus == 0 {
 		lastStatus = 429
 	}
-	return cliproxyexecutor.Response{}, statusErr{code: lastStatus, msg: string(lastBody)}
+	return cliproxyexecutor.Response{}, newGeminiStatusErr(lastStatus, lastBody)
 }
 
-func (e *GeminiCLIExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
-	log.Debugf("gemini cli executor: refresh called")
-	_ = ctx
+// Refresh refreshes the authentication credentials (no-op for Gemini CLI).
+func (e *GeminiCLIExecutor) Refresh(_ context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
 	return auth, nil
 }
 
@@ -533,9 +529,9 @@ func prepareGeminiCLITokenSource(ctx context.Context, cfg *config.Config, auth *
 	}
 
 	conf := &oauth2.Config{
-		ClientID:     geminiOauthClientID,
-		ClientSecret: geminiOauthClientSecret,
-		Scopes:       geminiOauthScopes,
+		ClientID:     geminiOAuthClientID,
+		ClientSecret: geminiOAuthClientSecret,
+		Scopes:       geminiOAuthScopes,
 		Endpoint:     google.Endpoint,
 	}
 
@@ -689,7 +685,7 @@ func cliPreviewFallbackOrder(model string) []string {
 	case "gemini-2.5-pro":
 		return []string{
 			// "gemini-2.5-pro-preview-05-06",
-			"gemini-2.5-pro-preview-06-05",
+			// "gemini-2.5-pro-preview-06-05",
 		}
 	case "gemini-2.5-flash":
 		return []string{
@@ -770,3 +766,42 @@ func fixGeminiCLIImageAspectRatio(modelName string, rawJSON []byte) []byte {
 	}
 	return rawJSON
 }
+
+func newGeminiStatusErr(statusCode int, body []byte) statusErr {
+	err := statusErr{code: statusCode, msg: string(body)}
+	if statusCode == http.StatusTooManyRequests {
+		if retryAfter, parseErr := parseRetryDelay(body); parseErr == nil && retryAfter != nil {
+			err.retryAfter = retryAfter
+		}
+	}
+	return err
+}
+
+// parseRetryDelay extracts the retry delay from a Google API 429 error response.
+// The error response contains a RetryInfo.retryDelay field in the format "0.847655010s".
+// Returns the parsed duration or an error if it cannot be determined.
+func parseRetryDelay(errorBody []byte) (*time.Duration, error) {
+	// Try to parse the retryDelay from the error response
+	// Format: error.details[].retryDelay where @type == "type.googleapis.com/google.rpc.RetryInfo"
+	details := gjson.GetBytes(errorBody, "error.details")
+	if !details.Exists() || !details.IsArray() {
+		return nil, fmt.Errorf("no error.details found")
+	}
+
+	for _, detail := range details.Array() {
+		typeVal := detail.Get("@type").String()
+		if typeVal == "type.googleapis.com/google.rpc.RetryInfo" {
+			retryDelay := detail.Get("retryDelay").String()
+			if retryDelay != "" {
+				// Parse duration string like "0.847655010s"
+				duration, err := time.ParseDuration(retryDelay)
+				if err != nil {
+					return nil, fmt.Errorf("failed to parse duration")
+				}
+				return &duration, nil
+			}
+		}
+	}
+
+	return nil, fmt.Errorf("no RetryInfo found")
+}

internal/runtime/executor/gemini_executor.go

@@ -11,7 +11,6 @@ import (
 	"io"
 	"net/http"
 	"strings"
-	"time"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
@@ -21,8 +20,6 @@ import (
 	log "github.com/sirupsen/logrus"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
-	"golang.org/x/oauth2"
-	"golang.org/x/oauth2/google"
 )
 
 const (
@@ -31,6 +28,9 @@ const (
 
 	// glAPIVersion is the API version used for Gemini requests.
 	glAPIVersion = "v1beta"
+
+	// streamScannerBuffer is the buffer size for SSE stream scanning.
+	streamScannerBuffer = 52_428_800
 )
 
 // GeminiExecutor is a stateless executor for the official Gemini API using API keys.
@@ -48,9 +48,11 @@ type GeminiExecutor struct {
 //
 // Returns:
 //   - *GeminiExecutor: A new Gemini executor instance
-func NewGeminiExecutor(cfg *config.Config) *GeminiExecutor { return &GeminiExecutor{cfg: cfg} }
+func NewGeminiExecutor(cfg *config.Config) *GeminiExecutor {
+	return &GeminiExecutor{cfg: cfg}
+}
 
-// Identifier returns the executor identifier for Gemini.
+// Identifier returns the executor identifier.
 func (e *GeminiExecutor) Identifier() string { return "gemini" }
 
 // PrepareRequest prepares the HTTP request for execution (no-op for Gemini).
@@ -75,20 +77,19 @@ func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
 
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	// Official Gemini API via API key or OAuth bearer
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
-	if budgetOverride, includeOverride, ok := util.GeminiThinkingFromMetadata(req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
-		if budgetOverride != nil {
-			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
-			budgetOverride = &norm
-		}
-		body = util.ApplyGeminiThinkingConfig(body, budgetOverride, includeOverride)
-	}
+	body = applyThinkingMetadata(body, req.Metadata, req.Model)
+	body = util.ApplyDefaultThinkingIfNeeded(req.Model, body)
+	body = util.NormalizeGeminiThinkingBudget(req.Model, body)
 	body = util.StripThinkingConfigIfUnsupported(req.Model, body)
 	body = fixGeminiImageAspectRatio(req.Model, body)
 	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 
 	action := "generateContent"
 	if req.Metadata != nil {
@@ -97,7 +98,7 @@ func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 		}
 	}
 	baseURL := resolveGeminiBaseURL(auth)
-	url := fmt.Sprintf("%s/%s/models/%s:%s", baseURL, glAPIVersion, req.Model, action)
+	url := fmt.Sprintf("%s/%s/models/%s:%s", baseURL, glAPIVersion, upstreamModel, action)
 	if opts.Alt != "" && action != "countTokens" {
 		url = url + fmt.Sprintf("?$alt=%s", opts.Alt)
 	}
@@ -165,28 +166,28 @@ func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	return resp, nil
 }
 
+// ExecuteStream performs a streaming request to the Gemini API.
 func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
 	apiKey, bearer := geminiCreds(auth)
 
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
 
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
-	if budgetOverride, includeOverride, ok := util.GeminiThinkingFromMetadata(req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
-		if budgetOverride != nil {
-			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
-			budgetOverride = &norm
-		}
-		body = util.ApplyGeminiThinkingConfig(body, budgetOverride, includeOverride)
-	}
+	body = applyThinkingMetadata(body, req.Metadata, req.Model)
+	body = util.ApplyDefaultThinkingIfNeeded(req.Model, body)
+	body = util.NormalizeGeminiThinkingBudget(req.Model, body)
 	body = util.StripThinkingConfigIfUnsupported(req.Model, body)
 	body = fixGeminiImageAspectRatio(req.Model, body)
 	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 
 	baseURL := resolveGeminiBaseURL(auth)
-	url := fmt.Sprintf("%s/%s/models/%s:%s", baseURL, glAPIVersion, req.Model, "streamGenerateContent")
+	url := fmt.Sprintf("%s/%s/models/%s:%s", baseURL, glAPIVersion, upstreamModel, "streamGenerateContent")
 	if opts.Alt == "" {
 		url = url + "?alt=sse"
 	} else {
@@ -251,16 +252,20 @@ func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 			}
 		}()
 		scanner := bufio.NewScanner(httpResp.Body)
-		buf := make([]byte, 20_971_520)
-		scanner.Buffer(buf, 20_971_520)
+		scanner.Buffer(nil, streamScannerBuffer)
 		var param any
 		for scanner.Scan() {
 			line := scanner.Bytes()
 			appendAPIResponseChunk(ctx, e.cfg, line)
-			if detail, ok := parseGeminiStreamUsage(line); ok {
+			filtered := FilterSSEUsageMetadata(line)
+			payload := jsonPayload(filtered)
+			if len(payload) == 0 {
+				continue
+			}
+			if detail, ok := parseGeminiStreamUsage(payload); ok {
 				reporter.publish(ctx, detail)
 			}
-			lines := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), body, bytes.Clone(line), &param)
+			lines := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), body, bytes.Clone(payload), &param)
 			for i := range lines {
 				out <- cliproxyexecutor.StreamChunk{Payload: []byte(lines[i])}
 			}
@@ -278,19 +283,14 @@ func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	return stream, nil
 }
 
+// CountTokens counts tokens for the given request using the Gemini API.
 func (e *GeminiExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
 	apiKey, bearer := geminiCreds(auth)
 
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
 	translatedReq := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
-	if budgetOverride, includeOverride, ok := util.GeminiThinkingFromMetadata(req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
-		if budgetOverride != nil {
-			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
-			budgetOverride = &norm
-		}
-		translatedReq = util.ApplyGeminiThinkingConfig(translatedReq, budgetOverride, includeOverride)
-	}
+	translatedReq = applyThinkingMetadata(translatedReq, req.Metadata, req.Model)
 	translatedReq = util.StripThinkingConfigIfUnsupported(req.Model, translatedReq)
 	translatedReq = fixGeminiImageAspectRatio(req.Model, translatedReq)
 	respCtx := context.WithValue(ctx, "alt", opts.Alt)
@@ -357,106 +357,8 @@ func (e *GeminiExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
 	return cliproxyexecutor.Response{Payload: []byte(translated)}, nil
 }
 
-func (e *GeminiExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
-	log.Debugf("gemini executor: refresh called")
-	// OAuth bearer token refresh for official Gemini API.
-	if auth == nil {
-		return nil, fmt.Errorf("gemini executor: auth is nil")
-	}
-	if auth.Metadata == nil {
-		return auth, nil
-	}
-	// Token data is typically nested under "token" map in Gemini files.
-	tokenMap, _ := auth.Metadata["token"].(map[string]any)
-	var refreshToken, accessToken, clientID, clientSecret, tokenURI, expiryStr string
-	if tokenMap != nil {
-		if v, ok := tokenMap["refresh_token"].(string); ok {
-			refreshToken = v
-		}
-		if v, ok := tokenMap["access_token"].(string); ok {
-			accessToken = v
-		}
-		if v, ok := tokenMap["client_id"].(string); ok {
-			clientID = v
-		}
-		if v, ok := tokenMap["client_secret"].(string); ok {
-			clientSecret = v
-		}
-		if v, ok := tokenMap["token_uri"].(string); ok {
-			tokenURI = v
-		}
-		if v, ok := tokenMap["expiry"].(string); ok {
-			expiryStr = v
-		}
-	} else {
-		// Fallback to top-level keys if present
-		if v, ok := auth.Metadata["refresh_token"].(string); ok {
-			refreshToken = v
-		}
-		if v, ok := auth.Metadata["access_token"].(string); ok {
-			accessToken = v
-		}
-		if v, ok := auth.Metadata["client_id"].(string); ok {
-			clientID = v
-		}
-		if v, ok := auth.Metadata["client_secret"].(string); ok {
-			clientSecret = v
-		}
-		if v, ok := auth.Metadata["token_uri"].(string); ok {
-			tokenURI = v
-		}
-		if v, ok := auth.Metadata["expiry"].(string); ok {
-			expiryStr = v
-		}
-	}
-	if refreshToken == "" {
-		// Nothing to do for API key or cookie based entries
-		return auth, nil
-	}
-
-	// Prepare oauth2 config; default to Google endpoints
-	endpoint := google.Endpoint
-	if tokenURI != "" {
-		endpoint.TokenURL = tokenURI
-	}
-	conf := &oauth2.Config{ClientID: clientID, ClientSecret: clientSecret, Endpoint: endpoint}
-
-	// Ensure proxy-aware HTTP client for token refresh
-	httpClient := util.SetProxy(&e.cfg.SDKConfig, &http.Client{})
-	ctx = context.WithValue(ctx, oauth2.HTTPClient, httpClient)
-
-	// Build base token
-	tok := &oauth2.Token{AccessToken: accessToken, RefreshToken: refreshToken}
-	if t, err := time.Parse(time.RFC3339, expiryStr); err == nil {
-		tok.Expiry = t
-	}
-	newTok, err := conf.TokenSource(ctx, tok).Token()
-	if err != nil {
-		return nil, err
-	}
-
-	// Persist back to metadata; prefer nested token map if present
-	if tokenMap == nil {
-		tokenMap = make(map[string]any)
-	}
-	tokenMap["access_token"] = newTok.AccessToken
-	tokenMap["refresh_token"] = newTok.RefreshToken
-	tokenMap["expiry"] = newTok.Expiry.Format(time.RFC3339)
-	if clientID != "" {
-		tokenMap["client_id"] = clientID
-	}
-	if clientSecret != "" {
-		tokenMap["client_secret"] = clientSecret
-	}
-	if tokenURI != "" {
-		tokenMap["token_uri"] = tokenURI
-	}
-	auth.Metadata["token"] = tokenMap
-
-	// Also mirror top-level access_token for compatibility if previously present
-	if _, ok := auth.Metadata["access_token"]; ok {
-		auth.Metadata["access_token"] = newTok.AccessToken
-	}
+// Refresh refreshes the authentication credentials (no-op for Gemini API key).
+func (e *GeminiExecutor) Refresh(_ context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
 	return auth, nil
 }
 

internal/runtime/executor/gemini_vertex_executor.go

@@ -1,6 +1,6 @@
-// Package executor contains provider executors. This file implements the Vertex AI
-// Gemini executor that talks to Google Vertex AI endpoints using service account
-// credentials imported by the CLI.
+// Package executor provides runtime execution capabilities for various AI service providers.
+// This file implements the Vertex AI Gemini executor that talks to Google Vertex AI
+// endpoints using service account credentials or API keys.
 package executor
 
 import (
@@ -36,42 +36,108 @@ type GeminiVertexExecutor struct {
 	cfg *config.Config
 }
 
-// NewGeminiVertexExecutor constructs the Vertex executor.
+// NewGeminiVertexExecutor creates a new Vertex AI Gemini executor instance.
+//
+// Parameters:
+//   - cfg: The application configuration
+//
+// Returns:
+//   - *GeminiVertexExecutor: A new Vertex AI Gemini executor instance
 func NewGeminiVertexExecutor(cfg *config.Config) *GeminiVertexExecutor {
 	return &GeminiVertexExecutor{cfg: cfg}
 }
 
-// Identifier returns provider key for manager routing.
+// Identifier returns the executor identifier.
 func (e *GeminiVertexExecutor) Identifier() string { return "vertex" }
 
-// PrepareRequest is a no-op for Vertex.
+// PrepareRequest prepares the HTTP request for execution (no-op for Vertex).
 func (e *GeminiVertexExecutor) PrepareRequest(_ *http.Request, _ *cliproxyauth.Auth) error {
 	return nil
 }
 
-// Execute handles non-streaming requests.
+// Execute performs a non-streaming request to the Vertex AI API.
 func (e *GeminiVertexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
-	projectID, location, saJSON, errCreds := vertexCreds(auth)
-	if errCreds != nil {
-		return resp, errCreds
+	// Try API key authentication first
+	apiKey, baseURL := vertexAPICreds(auth)
+
+	// If no API key found, fall back to service account authentication
+	if apiKey == "" {
+		projectID, location, saJSON, errCreds := vertexCreds(auth)
+		if errCreds != nil {
+			return resp, errCreds
+		}
+		return e.executeWithServiceAccount(ctx, auth, req, opts, projectID, location, saJSON)
 	}
 
+	// Use API key authentication
+	return e.executeWithAPIKey(ctx, auth, req, opts, apiKey, baseURL)
+}
+
+// ExecuteStream performs a streaming request to the Vertex AI API.
+func (e *GeminiVertexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+	// Try API key authentication first
+	apiKey, baseURL := vertexAPICreds(auth)
+
+	// If no API key found, fall back to service account authentication
+	if apiKey == "" {
+		projectID, location, saJSON, errCreds := vertexCreds(auth)
+		if errCreds != nil {
+			return nil, errCreds
+		}
+		return e.executeStreamWithServiceAccount(ctx, auth, req, opts, projectID, location, saJSON)
+	}
+
+	// Use API key authentication
+	return e.executeStreamWithAPIKey(ctx, auth, req, opts, apiKey, baseURL)
+}
+
+// CountTokens counts tokens for the given request using the Vertex AI API.
+func (e *GeminiVertexExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
+	// Try API key authentication first
+	apiKey, baseURL := vertexAPICreds(auth)
+
+	// If no API key found, fall back to service account authentication
+	if apiKey == "" {
+		projectID, location, saJSON, errCreds := vertexCreds(auth)
+		if errCreds != nil {
+			return cliproxyexecutor.Response{}, errCreds
+		}
+		return e.countTokensWithServiceAccount(ctx, auth, req, opts, projectID, location, saJSON)
+	}
+
+	// Use API key authentication
+	return e.countTokensWithAPIKey(ctx, auth, req, opts, apiKey, baseURL)
+}
+
+// Refresh refreshes the authentication credentials (no-op for Vertex).
+func (e *GeminiVertexExecutor) Refresh(_ context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
+	return auth, nil
+}
+
+// executeWithServiceAccount handles authentication using service account credentials.
+// This method contains the original service account authentication logic.
+func (e *GeminiVertexExecutor) executeWithServiceAccount(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, projectID, location string, saJSON []byte) (resp cliproxyexecutor.Response, err error) {
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
 
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
-	if budgetOverride, includeOverride, ok := util.GeminiThinkingFromMetadata(req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
+	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(req.Model, req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
 		if budgetOverride != nil {
 			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
 			budgetOverride = &norm
 		}
 		body = util.ApplyGeminiThinkingConfig(body, budgetOverride, includeOverride)
 	}
+	body = util.ApplyDefaultThinkingIfNeeded(req.Model, body)
+	body = util.NormalizeGeminiThinkingBudget(req.Model, body)
 	body = util.StripThinkingConfigIfUnsupported(req.Model, body)
 	body = fixGeminiImageAspectRatio(req.Model, body)
 	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 
 	action := "generateContent"
 	if req.Metadata != nil {
@@ -80,7 +146,7 @@ func (e *GeminiVertexExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 		}
 	}
 	baseURL := vertexBaseURL(location)
-	url := fmt.Sprintf("%s/%s/projects/%s/locations/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, projectID, location, req.Model, action)
+	url := fmt.Sprintf("%s/%s/projects/%s/locations/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, projectID, location, upstreamModel, action)
 	if opts.Alt != "" && action != "countTokens" {
 		url = url + fmt.Sprintf("?$alt=%s", opts.Alt)
 	}
@@ -149,32 +215,133 @@ func (e *GeminiVertexExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 	return resp, nil
 }
 
-// ExecuteStream handles SSE streaming for Vertex.
-func (e *GeminiVertexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
-	projectID, location, saJSON, errCreds := vertexCreds(auth)
-	if errCreds != nil {
-		return nil, errCreds
-	}
-
+// executeWithAPIKey handles authentication using API key credentials.
+func (e *GeminiVertexExecutor) executeWithAPIKey(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, apiKey, baseURL string) (resp cliproxyexecutor.Response, err error) {
 	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
 	defer reporter.trackFailure(ctx, &err)
 
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
-	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
-	if budgetOverride, includeOverride, ok := util.GeminiThinkingFromMetadata(req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
+	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
+	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(req.Model, req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
 		if budgetOverride != nil {
 			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
 			budgetOverride = &norm
 		}
 		body = util.ApplyGeminiThinkingConfig(body, budgetOverride, includeOverride)
 	}
+	body = util.ApplyDefaultThinkingIfNeeded(req.Model, body)
+	body = util.NormalizeGeminiThinkingBudget(req.Model, body)
 	body = util.StripThinkingConfigIfUnsupported(req.Model, body)
 	body = fixGeminiImageAspectRatio(req.Model, body)
 	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
+
+	action := "generateContent"
+	if req.Metadata != nil {
+		if a, _ := req.Metadata["action"].(string); a == "countTokens" {
+			action = "countTokens"
+		}
+	}
+
+	// For API key auth, use simpler URL format without project/location
+	if baseURL == "" {
+		baseURL = "https://generativelanguage.googleapis.com"
+	}
+	url := fmt.Sprintf("%s/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, upstreamModel, action)
+	if opts.Alt != "" && action != "countTokens" {
+		url = url + fmt.Sprintf("?$alt=%s", opts.Alt)
+	}
+	body, _ = sjson.DeleteBytes(body, "session_id")
+
+	httpReq, errNewReq := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
+	if errNewReq != nil {
+		return resp, errNewReq
+	}
+	httpReq.Header.Set("Content-Type", "application/json")
+	if apiKey != "" {
+		httpReq.Header.Set("x-goog-api-key", apiKey)
+	}
+	applyGeminiHeaders(httpReq, auth)
+
+	var authID, authLabel, authType, authValue string
+	if auth != nil {
+		authID = auth.ID
+		authLabel = auth.Label
+		authType, authValue = auth.AccountInfo()
+	}
+	recordAPIRequest(ctx, e.cfg, upstreamRequestLog{
+		URL:       url,
+		Method:    http.MethodPost,
+		Headers:   httpReq.Header.Clone(),
+		Body:      body,
+		Provider:  e.Identifier(),
+		AuthID:    authID,
+		AuthLabel: authLabel,
+		AuthType:  authType,
+		AuthValue: authValue,
+	})
+
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	httpResp, errDo := httpClient.Do(httpReq)
+	if errDo != nil {
+		recordAPIResponseError(ctx, e.cfg, errDo)
+		return resp, errDo
+	}
+	defer func() {
+		if errClose := httpResp.Body.Close(); errClose != nil {
+			log.Errorf("vertex executor: close response body error: %v", errClose)
+		}
+	}()
+	recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
+	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
+		b, _ := io.ReadAll(httpResp.Body)
+		appendAPIResponseChunk(ctx, e.cfg, b)
+		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		err = statusErr{code: httpResp.StatusCode, msg: string(b)}
+		return resp, err
+	}
+	data, errRead := io.ReadAll(httpResp.Body)
+	if errRead != nil {
+		recordAPIResponseError(ctx, e.cfg, errRead)
+		return resp, errRead
+	}
+	appendAPIResponseChunk(ctx, e.cfg, data)
+	reporter.publish(ctx, parseGeminiUsage(data))
+	var param any
+	out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), body, data, &param)
+	resp = cliproxyexecutor.Response{Payload: []byte(out)}
+	return resp, nil
+}
+
+// executeStreamWithServiceAccount handles streaming authentication using service account credentials.
+func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, projectID, location string, saJSON []byte) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
+	defer reporter.trackFailure(ctx, &err)
+
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
+	from := opts.SourceFormat
+	to := sdktranslator.FromString("gemini")
+	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
+	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(req.Model, req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
+		if budgetOverride != nil {
+			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
+			budgetOverride = &norm
+		}
+		body = util.ApplyGeminiThinkingConfig(body, budgetOverride, includeOverride)
+	}
+	body = util.ApplyDefaultThinkingIfNeeded(req.Model, body)
+	body = util.NormalizeGeminiThinkingBudget(req.Model, body)
+	body = util.StripThinkingConfigIfUnsupported(req.Model, body)
+	body = fixGeminiImageAspectRatio(req.Model, body)
+	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
 
 	baseURL := vertexBaseURL(location)
-	url := fmt.Sprintf("%s/%s/projects/%s/locations/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, projectID, location, req.Model, "streamGenerateContent")
+	url := fmt.Sprintf("%s/%s/projects/%s/locations/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, projectID, location, upstreamModel, "streamGenerateContent")
 	if opts.Alt == "" {
 		url = url + "?alt=sse"
 	} else {
@@ -240,8 +407,7 @@ func (e *GeminiVertexExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 			}
 		}()
 		scanner := bufio.NewScanner(httpResp.Body)
-		buf := make([]byte, 20_971_520)
-		scanner.Buffer(buf, 20_971_520)
+		scanner.Buffer(nil, streamScannerBuffer)
 		var param any
 		for scanner.Scan() {
 			line := scanner.Bytes()
@@ -267,16 +433,131 @@ func (e *GeminiVertexExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 	return stream, nil
 }
 
-// CountTokens calls Vertex countTokens endpoint.
-func (e *GeminiVertexExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
-	projectID, location, saJSON, errCreds := vertexCreds(auth)
-	if errCreds != nil {
-		return cliproxyexecutor.Response{}, errCreds
+// executeStreamWithAPIKey handles streaming authentication using API key credentials.
+func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, apiKey, baseURL string) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+	reporter := newUsageReporter(ctx, e.Identifier(), req.Model, auth)
+	defer reporter.trackFailure(ctx, &err)
+
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
+	from := opts.SourceFormat
+	to := sdktranslator.FromString("gemini")
+	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
+	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(req.Model, req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
+		if budgetOverride != nil {
+			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
+			budgetOverride = &norm
+		}
+		body = util.ApplyGeminiThinkingConfig(body, budgetOverride, includeOverride)
 	}
+	body = util.ApplyDefaultThinkingIfNeeded(req.Model, body)
+	body = util.NormalizeGeminiThinkingBudget(req.Model, body)
+	body = util.StripThinkingConfigIfUnsupported(req.Model, body)
+	body = fixGeminiImageAspectRatio(req.Model, body)
+	body = applyPayloadConfig(e.cfg, req.Model, body)
+	body, _ = sjson.SetBytes(body, "model", upstreamModel)
+
+	// For API key auth, use simpler URL format without project/location
+	if baseURL == "" {
+		baseURL = "https://generativelanguage.googleapis.com"
+	}
+	url := fmt.Sprintf("%s/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, upstreamModel, "streamGenerateContent")
+	if opts.Alt == "" {
+		url = url + "?alt=sse"
+	} else {
+		url = url + fmt.Sprintf("?$alt=%s", opts.Alt)
+	}
+	body, _ = sjson.DeleteBytes(body, "session_id")
+
+	httpReq, errNewReq := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
+	if errNewReq != nil {
+		return nil, errNewReq
+	}
+	httpReq.Header.Set("Content-Type", "application/json")
+	if apiKey != "" {
+		httpReq.Header.Set("x-goog-api-key", apiKey)
+	}
+	applyGeminiHeaders(httpReq, auth)
+
+	var authID, authLabel, authType, authValue string
+	if auth != nil {
+		authID = auth.ID
+		authLabel = auth.Label
+		authType, authValue = auth.AccountInfo()
+	}
+	recordAPIRequest(ctx, e.cfg, upstreamRequestLog{
+		URL:       url,
+		Method:    http.MethodPost,
+		Headers:   httpReq.Header.Clone(),
+		Body:      body,
+		Provider:  e.Identifier(),
+		AuthID:    authID,
+		AuthLabel: authLabel,
+		AuthType:  authType,
+		AuthValue: authValue,
+	})
+
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	httpResp, errDo := httpClient.Do(httpReq)
+	if errDo != nil {
+		recordAPIResponseError(ctx, e.cfg, errDo)
+		return nil, errDo
+	}
+	recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
+	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
+		b, _ := io.ReadAll(httpResp.Body)
+		appendAPIResponseChunk(ctx, e.cfg, b)
+		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		if errClose := httpResp.Body.Close(); errClose != nil {
+			log.Errorf("vertex executor: close response body error: %v", errClose)
+		}
+		return nil, statusErr{code: httpResp.StatusCode, msg: string(b)}
+	}
+
+	out := make(chan cliproxyexecutor.StreamChunk)
+	stream = out
+	go func() {
+		defer close(out)
+		defer func() {
+			if errClose := httpResp.Body.Close(); errClose != nil {
+				log.Errorf("vertex executor: close response body error: %v", errClose)
+			}
+		}()
+		scanner := bufio.NewScanner(httpResp.Body)
+		scanner.Buffer(nil, streamScannerBuffer)
+		var param any
+		for scanner.Scan() {
+			line := scanner.Bytes()
+			appendAPIResponseChunk(ctx, e.cfg, line)
+			if detail, ok := parseGeminiStreamUsage(line); ok {
+				reporter.publish(ctx, detail)
+			}
+			lines := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), body, bytes.Clone(line), &param)
+			for i := range lines {
+				out <- cliproxyexecutor.StreamChunk{Payload: []byte(lines[i])}
+			}
+		}
+		lines := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), body, []byte("[DONE]"), &param)
+		for i := range lines {
+			out <- cliproxyexecutor.StreamChunk{Payload: []byte(lines[i])}
+		}
+		if errScan := scanner.Err(); errScan != nil {
+			recordAPIResponseError(ctx, e.cfg, errScan)
+			reporter.publishFailure(ctx)
+			out <- cliproxyexecutor.StreamChunk{Err: errScan}
+		}
+	}()
+	return stream, nil
+}
+
+// countTokensWithServiceAccount counts tokens using service account credentials.
+func (e *GeminiVertexExecutor) countTokensWithServiceAccount(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, projectID, location string, saJSON []byte) (cliproxyexecutor.Response, error) {
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("gemini")
 	translatedReq := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
-	if budgetOverride, includeOverride, ok := util.GeminiThinkingFromMetadata(req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
+	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(req.Model, req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
 		if budgetOverride != nil {
 			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
 			budgetOverride = &norm
@@ -285,13 +566,14 @@ func (e *GeminiVertexExecutor) CountTokens(ctx context.Context, auth *cliproxyau
 	}
 	translatedReq = util.StripThinkingConfigIfUnsupported(req.Model, translatedReq)
 	translatedReq = fixGeminiImageAspectRatio(req.Model, translatedReq)
+	translatedReq, _ = sjson.SetBytes(translatedReq, "model", upstreamModel)
 	respCtx := context.WithValue(ctx, "alt", opts.Alt)
 	translatedReq, _ = sjson.DeleteBytes(translatedReq, "tools")
 	translatedReq, _ = sjson.DeleteBytes(translatedReq, "generationConfig")
 	translatedReq, _ = sjson.DeleteBytes(translatedReq, "safetySettings")
 
 	baseURL := vertexBaseURL(location)
-	url := fmt.Sprintf("%s/%s/projects/%s/locations/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, projectID, location, req.Model, "countTokens")
+	url := fmt.Sprintf("%s/%s/projects/%s/locations/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, projectID, location, upstreamModel, "countTokens")
 
 	httpReq, errNewReq := http.NewRequestWithContext(respCtx, http.MethodPost, url, bytes.NewReader(translatedReq))
 	if errNewReq != nil {
@@ -357,9 +639,93 @@ func (e *GeminiVertexExecutor) CountTokens(ctx context.Context, auth *cliproxyau
 	return cliproxyexecutor.Response{Payload: []byte(out)}, nil
 }
 
-// Refresh is a no-op for service account based credentials.
-func (e *GeminiVertexExecutor) Refresh(_ context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
-	return auth, nil
+// countTokensWithAPIKey handles token counting using API key credentials.
+func (e *GeminiVertexExecutor) countTokensWithAPIKey(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, apiKey, baseURL string) (cliproxyexecutor.Response, error) {
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+
+	from := opts.SourceFormat
+	to := sdktranslator.FromString("gemini")
+	translatedReq := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
+	if budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(req.Model, req.Metadata); ok && util.ModelSupportsThinking(req.Model) {
+		if budgetOverride != nil {
+			norm := util.NormalizeThinkingBudget(req.Model, *budgetOverride)
+			budgetOverride = &norm
+		}
+		translatedReq = util.ApplyGeminiThinkingConfig(translatedReq, budgetOverride, includeOverride)
+	}
+	translatedReq = util.StripThinkingConfigIfUnsupported(req.Model, translatedReq)
+	translatedReq = fixGeminiImageAspectRatio(req.Model, translatedReq)
+	translatedReq, _ = sjson.SetBytes(translatedReq, "model", upstreamModel)
+	respCtx := context.WithValue(ctx, "alt", opts.Alt)
+	translatedReq, _ = sjson.DeleteBytes(translatedReq, "tools")
+	translatedReq, _ = sjson.DeleteBytes(translatedReq, "generationConfig")
+	translatedReq, _ = sjson.DeleteBytes(translatedReq, "safetySettings")
+
+	// For API key auth, use simpler URL format without project/location
+	if baseURL == "" {
+		baseURL = "https://generativelanguage.googleapis.com"
+	}
+	url := fmt.Sprintf("%s/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, req.Model, "countTokens")
+
+	httpReq, errNewReq := http.NewRequestWithContext(respCtx, http.MethodPost, url, bytes.NewReader(translatedReq))
+	if errNewReq != nil {
+		return cliproxyexecutor.Response{}, errNewReq
+	}
+	httpReq.Header.Set("Content-Type", "application/json")
+	if apiKey != "" {
+		httpReq.Header.Set("x-goog-api-key", apiKey)
+	}
+	applyGeminiHeaders(httpReq, auth)
+
+	var authID, authLabel, authType, authValue string
+	if auth != nil {
+		authID = auth.ID
+		authLabel = auth.Label
+		authType, authValue = auth.AccountInfo()
+	}
+	recordAPIRequest(ctx, e.cfg, upstreamRequestLog{
+		URL:       url,
+		Method:    http.MethodPost,
+		Headers:   httpReq.Header.Clone(),
+		Body:      translatedReq,
+		Provider:  e.Identifier(),
+		AuthID:    authID,
+		AuthLabel: authLabel,
+		AuthType:  authType,
+		AuthValue: authValue,
+	})
+
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	httpResp, errDo := httpClient.Do(httpReq)
+	if errDo != nil {
+		recordAPIResponseError(ctx, e.cfg, errDo)
+		return cliproxyexecutor.Response{}, errDo
+	}
+	defer func() {
+		if errClose := httpResp.Body.Close(); errClose != nil {
+			log.Errorf("vertex executor: close response body error: %v", errClose)
+		}
+	}()
+	recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
+	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
+		b, _ := io.ReadAll(httpResp.Body)
+		appendAPIResponseChunk(ctx, e.cfg, b)
+		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		return cliproxyexecutor.Response{}, statusErr{code: httpResp.StatusCode, msg: string(b)}
+	}
+	data, errRead := io.ReadAll(httpResp.Body)
+	if errRead != nil {
+		recordAPIResponseError(ctx, e.cfg, errRead)
+		return cliproxyexecutor.Response{}, errRead
+	}
+	appendAPIResponseChunk(ctx, e.cfg, data)
+	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
+		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
+		return cliproxyexecutor.Response{}, statusErr{code: httpResp.StatusCode, msg: string(data)}
+	}
+	count := gjson.GetBytes(data, "totalTokens").Int()
+	out := sdktranslator.TranslateTokenCount(ctx, to, from, count, data)
+	return cliproxyexecutor.Response{Payload: []byte(out)}, nil
 }
 
 // vertexCreds extracts project, location and raw service account JSON from auth metadata.
@@ -402,6 +768,23 @@ func vertexCreds(a *cliproxyauth.Auth) (projectID, location string, serviceAccou
 	return projectID, location, saJSON, nil
 }
 
+// vertexAPICreds extracts API key and base URL from auth attributes following the claudeCreds pattern.
+func vertexAPICreds(a *cliproxyauth.Auth) (apiKey, baseURL string) {
+	if a == nil {
+		return "", ""
+	}
+	if a.Attributes != nil {
+		apiKey = a.Attributes["api_key"]
+		baseURL = a.Attributes["base_url"]
+	}
+	if apiKey == "" && a.Metadata != nil {
+		if v, ok := a.Metadata["access_token"].(string); ok {
+			apiKey = v
+		}
+	}
+	return
+}
+
 func vertexBaseURL(location string) string {
 	loc := strings.TrimSpace(location)
 	if loc == "" {

internal/runtime/executor/iflow_executor.go

@@ -57,6 +57,15 @@ func (e *IFlowExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
+	body = applyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning_effort")
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel != "" {
+		body, _ = sjson.SetBytes(body, "model", upstreamModel)
+	}
+	body = normalizeThinkingConfig(body, upstreamModel)
+	if errValidate := validateThinkingConfig(body, upstreamModel); errValidate != nil {
+		return resp, errValidate
+	}
 	body = applyPayloadConfig(e.cfg, req.Model, body)
 
 	endpoint := strings.TrimSuffix(baseURL, "/") + iflowDefaultEndpoint
@@ -112,6 +121,8 @@ func (e *IFlowExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	}
 	appendAPIResponseChunk(ctx, e.cfg, data)
 	reporter.publish(ctx, parseOpenAIUsage(data))
+	// Ensure usage is recorded even if upstream omits usage metadata.
+	reporter.ensurePublished(ctx)
 
 	var param any
 	out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), body, data, &param)
@@ -137,6 +148,15 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 	to := sdktranslator.FromString("openai")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 
+	body = applyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning_effort")
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel != "" {
+		body, _ = sjson.SetBytes(body, "model", upstreamModel)
+	}
+	body = normalizeThinkingConfig(body, upstreamModel)
+	if errValidate := validateThinkingConfig(body, upstreamModel); errValidate != nil {
+		return nil, errValidate
+	}
 	// Ensure tools array exists to avoid provider quirks similar to Qwen's behaviour.
 	toolsResult := gjson.GetBytes(body, "tools")
 	if toolsResult.Exists() && toolsResult.IsArray() && len(toolsResult.Array()) == 0 {
@@ -199,8 +219,7 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 		}()
 
 		scanner := bufio.NewScanner(httpResp.Body)
-		buf := make([]byte, 20_971_520)
-		scanner.Buffer(buf, 20_971_520)
+		scanner.Buffer(nil, 52_428_800) // 50MB 
 		var param any
 		for scanner.Scan() {
 			line := scanner.Bytes()
@@ -218,6 +237,8 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 			reporter.publishFailure(ctx)
 			out <- cliproxyexecutor.StreamChunk{Err: errScan}
 		}
+		// Guarantee a usage record exists even if the stream never emitted usage data.
+		reporter.ensurePublished(ctx)
 	}()
 
 	return stream, nil
@@ -243,13 +264,87 @@ func (e *IFlowExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth
 	return cliproxyexecutor.Response{Payload: []byte(translated)}, nil
 }
 
-// Refresh refreshes OAuth tokens and updates the stored API key.
+// Refresh refreshes OAuth tokens or cookie-based API keys and updates the stored API key.
 func (e *IFlowExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
 	log.Debugf("iflow executor: refresh called")
 	if auth == nil {
 		return nil, fmt.Errorf("iflow executor: auth is nil")
 	}
 
+	// Check if this is cookie-based authentication
+	var cookie string
+	var email string
+	if auth.Metadata != nil {
+		if v, ok := auth.Metadata["cookie"].(string); ok {
+			cookie = strings.TrimSpace(v)
+		}
+		if v, ok := auth.Metadata["email"].(string); ok {
+			email = strings.TrimSpace(v)
+		}
+	}
+
+	// If cookie is present, use cookie-based refresh
+	if cookie != "" && email != "" {
+		return e.refreshCookieBased(ctx, auth, cookie, email)
+	}
+
+	// Otherwise, use OAuth-based refresh
+	return e.refreshOAuthBased(ctx, auth)
+}
+
+// refreshCookieBased refreshes API key using browser cookie
+func (e *IFlowExecutor) refreshCookieBased(ctx context.Context, auth *cliproxyauth.Auth, cookie, email string) (*cliproxyauth.Auth, error) {
+	log.Debugf("iflow executor: checking refresh need for cookie-based API key for user: %s", email)
+
+	// Get current expiry time from metadata
+	var currentExpire string
+	if auth.Metadata != nil {
+		if v, ok := auth.Metadata["expired"].(string); ok {
+			currentExpire = strings.TrimSpace(v)
+		}
+	}
+
+	// Check if refresh is needed
+	needsRefresh, _, err := iflowauth.ShouldRefreshAPIKey(currentExpire)
+	if err != nil {
+		log.Warnf("iflow executor: failed to check refresh need: %v", err)
+		// If we can't check, continue with refresh anyway as a safety measure
+	} else if !needsRefresh {
+		log.Debugf("iflow executor: no refresh needed for user: %s", email)
+		return auth, nil
+	}
+
+	log.Infof("iflow executor: refreshing cookie-based API key for user: %s", email)
+
+	svc := iflowauth.NewIFlowAuth(e.cfg)
+	keyData, err := svc.RefreshAPIKey(ctx, cookie, email)
+	if err != nil {
+		log.Errorf("iflow executor: cookie-based API key refresh failed: %v", err)
+		return nil, err
+	}
+
+	if auth.Metadata == nil {
+		auth.Metadata = make(map[string]any)
+	}
+	auth.Metadata["api_key"] = keyData.APIKey
+	auth.Metadata["expired"] = keyData.ExpireTime
+	auth.Metadata["type"] = "iflow"
+	auth.Metadata["last_refresh"] = time.Now().Format(time.RFC3339)
+	auth.Metadata["cookie"] = cookie
+	auth.Metadata["email"] = email
+
+	log.Infof("iflow executor: cookie-based API key refreshed successfully, new expiry: %s", keyData.ExpireTime)
+
+	if auth.Attributes == nil {
+		auth.Attributes = make(map[string]string)
+	}
+	auth.Attributes["api_key"] = keyData.APIKey
+
+	return auth, nil
+}
+
+// refreshOAuthBased refreshes tokens using OAuth refresh token
+func (e *IFlowExecutor) refreshOAuthBased(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
 	refreshToken := ""
 	oldAccessToken := ""
 	if auth.Metadata != nil {

internal/runtime/executor/logging_helpers.go

@@ -323,7 +323,14 @@ func formatAuthInfo(info upstreamRequestLog) string {
 }
 
 func summarizeErrorBody(contentType string, body []byte) string {
-	if strings.Contains(strings.ToLower(contentType), "text/html") {
+	isHTML := strings.Contains(strings.ToLower(contentType), "text/html")
+	if !isHTML {
+		trimmed := bytes.TrimSpace(bytes.ToLower(body))
+		if bytes.HasPrefix(trimmed, []byte("<!doctype html")) || bytes.HasPrefix(trimmed, []byte("<html")) {
+			isHTML = true
+		}
+	}
+	if isHTML {
 		if title := extractHTMLTitle(body); title != "" {
 			return title
 		}

internal/runtime/executor/openai_compat_executor.go

@@ -8,6 +8,7 @@ import (
 	"io"
 	"net/http"
 	"strings"
+	"time"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
@@ -57,6 +58,15 @@ func (e *OpenAICompatExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 		translated = e.overrideModel(translated, modelOverride)
 	}
 	translated = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", translated)
+	translated = applyReasoningEffortMetadata(translated, req.Metadata, req.Model, "reasoning_effort")
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel != "" {
+		translated, _ = sjson.SetBytes(translated, "model", upstreamModel)
+	}
+	translated = normalizeThinkingConfig(translated, upstreamModel)
+	if errValidate := validateThinkingConfig(translated, upstreamModel); errValidate != nil {
+		return resp, errValidate
+	}
 
 	url := strings.TrimSuffix(baseURL, "/") + "/chat/completions"
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(translated))
@@ -142,6 +152,15 @@ func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 		translated = e.overrideModel(translated, modelOverride)
 	}
 	translated = applyPayloadConfigWithRoot(e.cfg, req.Model, to.String(), "", translated)
+	translated = applyReasoningEffortMetadata(translated, req.Metadata, req.Model, "reasoning_effort")
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel != "" {
+		translated, _ = sjson.SetBytes(translated, "model", upstreamModel)
+	}
+	translated = normalizeThinkingConfig(translated, upstreamModel)
+	if errValidate := validateThinkingConfig(translated, upstreamModel); errValidate != nil {
+		return nil, errValidate
+	}
 
 	url := strings.TrimSuffix(baseURL, "/") + "/chat/completions"
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(translated))
@@ -205,8 +224,7 @@ func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 			}
 		}()
 		scanner := bufio.NewScanner(httpResp.Body)
-		buf := make([]byte, 20_971_520)
-		scanner.Buffer(buf, 20_971_520)
+		scanner.Buffer(nil, 52_428_800) // 50MB
 		var param any
 		for scanner.Scan() {
 			line := scanner.Bytes()
@@ -341,8 +359,9 @@ func (e *OpenAICompatExecutor) overrideModel(payload []byte, model string) []byt
 }
 
 type statusErr struct {
-	code int
-	msg  string
+	code       int
+	msg        string
+	retryAfter *time.Duration
 }
 
 func (e statusErr) Error() string {
@@ -351,4 +370,5 @@ func (e statusErr) Error() string {
 	}
 	return fmt.Sprintf("status %d", e.code)
 }
-func (e statusErr) StatusCode() int { return e.code }
+func (e statusErr) StatusCode() int            { return e.code }
+func (e statusErr) RetryAfter() *time.Duration { return e.retryAfter }

internal/runtime/executor/payload_helpers.go

@@ -1,13 +1,71 @@
 package executor
 
 import (
+	"fmt"
+	"net/http"
 	"strings"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
 )
 
+// applyThinkingMetadata applies thinking config from model suffix metadata (e.g., (high), (8192))
+// for standard Gemini format payloads. It normalizes the budget when the model supports thinking.
+func applyThinkingMetadata(payload []byte, metadata map[string]any, model string) []byte {
+	budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(model, metadata)
+	if !ok || (budgetOverride == nil && includeOverride == nil) {
+		return payload
+	}
+	if !util.ModelSupportsThinking(model) {
+		return payload
+	}
+	if budgetOverride != nil {
+		norm := util.NormalizeThinkingBudget(model, *budgetOverride)
+		budgetOverride = &norm
+	}
+	return util.ApplyGeminiThinkingConfig(payload, budgetOverride, includeOverride)
+}
+
+// applyThinkingMetadataCLI applies thinking config from model suffix metadata (e.g., (high), (8192))
+// for Gemini CLI format payloads (nested under "request"). It normalizes the budget when the model supports thinking.
+func applyThinkingMetadataCLI(payload []byte, metadata map[string]any, model string) []byte {
+	budgetOverride, includeOverride, ok := util.ResolveThinkingConfigFromMetadata(model, metadata)
+	if !ok || (budgetOverride == nil && includeOverride == nil) {
+		return payload
+	}
+	if !util.ModelSupportsThinking(model) {
+		return payload
+	}
+	if budgetOverride != nil {
+		norm := util.NormalizeThinkingBudget(model, *budgetOverride)
+		budgetOverride = &norm
+	}
+	return util.ApplyGeminiCLIThinkingConfig(payload, budgetOverride, includeOverride)
+}
+
+// applyReasoningEffortMetadata applies reasoning effort overrides from metadata to the given JSON path.
+// Metadata values take precedence over any existing field when the model supports thinking, intentionally
+// overwriting caller-provided values to honor suffix/default metadata priority.
+func applyReasoningEffortMetadata(payload []byte, metadata map[string]any, model, field string) []byte {
+	if len(metadata) == 0 {
+		return payload
+	}
+	if !util.ModelSupportsThinking(model) {
+		return payload
+	}
+	if field == "" {
+		return payload
+	}
+	if effort, ok := util.ReasoningEffortFromMetadata(metadata); ok && effort != "" {
+		if updated, err := sjson.SetBytes(payload, field, effort); err == nil {
+			return updated
+		}
+	}
+	return payload
+}
+
 // applyPayloadConfig applies payload default and override rules from configuration
 // to the given JSON payload for the specified model.
 // Defaults only fill missing fields, while overrides always overwrite existing values.
@@ -157,3 +215,93 @@ func matchModelPattern(pattern, model string) bool {
 	}
 	return pi == len(pattern)
 }
+
+// normalizeThinkingConfig normalizes thinking-related fields in the payload
+// based on model capabilities. For models without thinking support, it strips
+// reasoning fields. For models with level-based thinking, it validates and
+// normalizes the reasoning effort level.
+func normalizeThinkingConfig(payload []byte, model string) []byte {
+	if len(payload) == 0 || model == "" {
+		return payload
+	}
+
+	if !util.ModelSupportsThinking(model) {
+		return stripThinkingFields(payload)
+	}
+
+	if util.ModelUsesThinkingLevels(model) {
+		return normalizeReasoningEffortLevel(payload, model)
+	}
+
+	return payload
+}
+
+// stripThinkingFields removes thinking-related fields from the payload for
+// models that do not support thinking.
+func stripThinkingFields(payload []byte) []byte {
+	fieldsToRemove := []string{
+		"reasoning",
+		"reasoning_effort",
+		"reasoning.effort",
+	}
+	out := payload
+	for _, field := range fieldsToRemove {
+		if gjson.GetBytes(out, field).Exists() {
+			out, _ = sjson.DeleteBytes(out, field)
+		}
+	}
+	return out
+}
+
+// normalizeReasoningEffortLevel validates and normalizes the reasoning_effort
+// or reasoning.effort field for level-based thinking models.
+func normalizeReasoningEffortLevel(payload []byte, model string) []byte {
+	out := payload
+
+	if effort := gjson.GetBytes(out, "reasoning_effort"); effort.Exists() {
+		if normalized, ok := util.NormalizeReasoningEffortLevel(model, effort.String()); ok {
+			out, _ = sjson.SetBytes(out, "reasoning_effort", normalized)
+		}
+	}
+
+	if effort := gjson.GetBytes(out, "reasoning.effort"); effort.Exists() {
+		if normalized, ok := util.NormalizeReasoningEffortLevel(model, effort.String()); ok {
+			out, _ = sjson.SetBytes(out, "reasoning.effort", normalized)
+		}
+	}
+
+	return out
+}
+
+// validateThinkingConfig checks for unsupported reasoning levels on level-based models.
+// Returns a statusErr with 400 when an unsupported level is supplied to avoid silently
+// downgrading requests.
+func validateThinkingConfig(payload []byte, model string) error {
+	if len(payload) == 0 || model == "" {
+		return nil
+	}
+	if !util.ModelSupportsThinking(model) || !util.ModelUsesThinkingLevels(model) {
+		return nil
+	}
+
+	levels := util.GetModelThinkingLevels(model)
+	checkField := func(path string) error {
+		if effort := gjson.GetBytes(payload, path); effort.Exists() {
+			if _, ok := util.NormalizeReasoningEffortLevel(model, effort.String()); !ok {
+				return statusErr{
+					code: http.StatusBadRequest,
+					msg:  fmt.Sprintf("unsupported reasoning effort level %q for model %s (supported: %s)", effort.String(), model, strings.Join(levels, ", ")),
+				}
+			}
+		}
+		return nil
+	}
+
+	if err := checkField("reasoning_effort"); err != nil {
+		return err
+	}
+	if err := checkField("reasoning.effort"); err != nil {
+		return err
+	}
+	return nil
+}

internal/runtime/executor/qwen_executor.go

@@ -12,6 +12,7 @@ import (
 
 	qwenauth "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/qwen"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 	cliproxyauth "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/auth"
 	cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/executor"
 	sdktranslator "github.com/router-for-me/CLIProxyAPI/v6/sdk/translator"
@@ -50,6 +51,15 @@ func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), false)
+	body = applyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning_effort")
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel != "" {
+		body, _ = sjson.SetBytes(body, "model", upstreamModel)
+	}
+	body = normalizeThinkingConfig(body, upstreamModel)
+	if errValidate := validateThinkingConfig(body, upstreamModel); errValidate != nil {
+		return resp, errValidate
+	}
 	body = applyPayloadConfig(e.cfg, req.Model, body)
 
 	url := strings.TrimSuffix(baseURL, "/") + "/chat/completions"
@@ -121,6 +131,15 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
 	to := sdktranslator.FromString("openai")
 	body := sdktranslator.TranslateRequest(from, to, req.Model, bytes.Clone(req.Payload), true)
 
+	body = applyReasoningEffortMetadata(body, req.Metadata, req.Model, "reasoning_effort")
+	upstreamModel := util.ResolveOriginalModel(req.Model, req.Metadata)
+	if upstreamModel != "" {
+		body, _ = sjson.SetBytes(body, "model", upstreamModel)
+	}
+	body = normalizeThinkingConfig(body, upstreamModel)
+	if errValidate := validateThinkingConfig(body, upstreamModel); errValidate != nil {
+		return nil, errValidate
+	}
 	toolsResult := gjson.GetBytes(body, "tools")
 	// I'm addressing the Qwen3 "poisoning" issue, which is caused by the model needing a tool to be defined. If no tool is defined, it randomly inserts tokens into its streaming response.
 	// This will have no real consequences. It's just to scare Qwen3.
@@ -181,8 +200,7 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
 			}
 		}()
 		scanner := bufio.NewScanner(httpResp.Body)
-		buf := make([]byte, 20_971_520)
-		scanner.Buffer(buf, 20_971_520)
+		scanner.Buffer(nil, 52_428_800) // 50MB
 		var param any
 		for scanner.Scan() {
 			line := scanner.Bytes()

internal/runtime/executor/usage_helpers.go

@@ -12,12 +12,14 @@ import (
 	cliproxyauth "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/auth"
 	"github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/usage"
 	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
 )
 
 type usageReporter struct {
 	provider    string
 	model       string
 	authID      string
+	authIndex   uint64
 	apiKey      string
 	source      string
 	requestedAt time.Time
@@ -35,6 +37,7 @@ func newUsageReporter(ctx context.Context, provider, model string, auth *cliprox
 	}
 	if auth != nil {
 		reporter.authID = auth.ID
+		reporter.authIndex = auth.EnsureIndex()
 	}
 	return reporter
 }
@@ -76,6 +79,7 @@ func (r *usageReporter) publishWithOutcome(ctx context.Context, detail usage.Det
 			Source:      r.source,
 			APIKey:      r.apiKey,
 			AuthID:      r.authID,
+			AuthIndex:   r.authIndex,
 			RequestedAt: r.requestedAt,
 			Failed:      failed,
 			Detail:      detail,
@@ -98,6 +102,7 @@ func (r *usageReporter) ensurePublished(ctx context.Context) {
 			Source:      r.source,
 			APIKey:      r.apiKey,
 			AuthID:      r.authID,
+			AuthIndex:   r.authIndex,
 			RequestedAt: r.requestedAt,
 			Failed:      false,
 			Detail:      usage.Detail{},
@@ -360,6 +365,204 @@ func parseGeminiCLIStreamUsage(line []byte) (usage.Detail, bool) {
 	return detail, true
 }
 
+func parseAntigravityUsage(data []byte) usage.Detail {
+	usageNode := gjson.ParseBytes(data)
+	node := usageNode.Get("response.usageMetadata")
+	if !node.Exists() {
+		node = usageNode.Get("usageMetadata")
+	}
+	if !node.Exists() {
+		node = usageNode.Get("usage_metadata")
+	}
+	if !node.Exists() {
+		return usage.Detail{}
+	}
+	detail := usage.Detail{
+		InputTokens:     node.Get("promptTokenCount").Int(),
+		OutputTokens:    node.Get("candidatesTokenCount").Int(),
+		ReasoningTokens: node.Get("thoughtsTokenCount").Int(),
+		TotalTokens:     node.Get("totalTokenCount").Int(),
+	}
+	if detail.TotalTokens == 0 {
+		detail.TotalTokens = detail.InputTokens + detail.OutputTokens + detail.ReasoningTokens
+	}
+	return detail
+}
+
+func parseAntigravityStreamUsage(line []byte) (usage.Detail, bool) {
+	payload := jsonPayload(line)
+	if len(payload) == 0 || !gjson.ValidBytes(payload) {
+		return usage.Detail{}, false
+	}
+	node := gjson.GetBytes(payload, "response.usageMetadata")
+	if !node.Exists() {
+		node = gjson.GetBytes(payload, "usageMetadata")
+	}
+	if !node.Exists() {
+		node = gjson.GetBytes(payload, "usage_metadata")
+	}
+	if !node.Exists() {
+		return usage.Detail{}, false
+	}
+	detail := usage.Detail{
+		InputTokens:     node.Get("promptTokenCount").Int(),
+		OutputTokens:    node.Get("candidatesTokenCount").Int(),
+		ReasoningTokens: node.Get("thoughtsTokenCount").Int(),
+		TotalTokens:     node.Get("totalTokenCount").Int(),
+	}
+	if detail.TotalTokens == 0 {
+		detail.TotalTokens = detail.InputTokens + detail.OutputTokens + detail.ReasoningTokens
+	}
+	return detail, true
+}
+
+var stopChunkWithoutUsage sync.Map
+
+func rememberStopWithoutUsage(traceID string) {
+	stopChunkWithoutUsage.Store(traceID, struct{}{})
+	time.AfterFunc(10*time.Minute, func() { stopChunkWithoutUsage.Delete(traceID) })
+}
+
+// FilterSSEUsageMetadata removes usageMetadata from SSE events that are not
+// terminal (finishReason != "stop"). Stop chunks are left untouched. This
+// function is shared between aistudio and antigravity executors.
+func FilterSSEUsageMetadata(payload []byte) []byte {
+	if len(payload) == 0 {
+		return payload
+	}
+
+	lines := bytes.Split(payload, []byte("\n"))
+	modified := false
+	foundData := false
+	for idx, line := range lines {
+		trimmed := bytes.TrimSpace(line)
+		if len(trimmed) == 0 || !bytes.HasPrefix(trimmed, []byte("data:")) {
+			continue
+		}
+		foundData = true
+		dataIdx := bytes.Index(line, []byte("data:"))
+		if dataIdx < 0 {
+			continue
+		}
+		rawJSON := bytes.TrimSpace(line[dataIdx+5:])
+		traceID := gjson.GetBytes(rawJSON, "traceId").String()
+		if isStopChunkWithoutUsage(rawJSON) && traceID != "" {
+			rememberStopWithoutUsage(traceID)
+			continue
+		}
+		if traceID != "" {
+			if _, ok := stopChunkWithoutUsage.Load(traceID); ok && hasUsageMetadata(rawJSON) {
+				stopChunkWithoutUsage.Delete(traceID)
+				continue
+			}
+		}
+
+		cleaned, changed := StripUsageMetadataFromJSON(rawJSON)
+		if !changed {
+			continue
+		}
+		var rebuilt []byte
+		rebuilt = append(rebuilt, line[:dataIdx]...)
+		rebuilt = append(rebuilt, []byte("data:")...)
+		if len(cleaned) > 0 {
+			rebuilt = append(rebuilt, ' ')
+			rebuilt = append(rebuilt, cleaned...)
+		}
+		lines[idx] = rebuilt
+		modified = true
+	}
+	if !modified {
+		if !foundData {
+			// Handle payloads that are raw JSON without SSE data: prefix.
+			trimmed := bytes.TrimSpace(payload)
+			cleaned, changed := StripUsageMetadataFromJSON(trimmed)
+			if !changed {
+				return payload
+			}
+			return cleaned
+		}
+		return payload
+	}
+	return bytes.Join(lines, []byte("\n"))
+}
+
+// StripUsageMetadataFromJSON drops usageMetadata unless finishReason is present (terminal).
+// It handles both formats:
+// - Aistudio: candidates.0.finishReason
+// - Antigravity: response.candidates.0.finishReason
+func StripUsageMetadataFromJSON(rawJSON []byte) ([]byte, bool) {
+	jsonBytes := bytes.TrimSpace(rawJSON)
+	if len(jsonBytes) == 0 || !gjson.ValidBytes(jsonBytes) {
+		return rawJSON, false
+	}
+
+	// Check for finishReason in both aistudio and antigravity formats
+	finishReason := gjson.GetBytes(jsonBytes, "candidates.0.finishReason")
+	if !finishReason.Exists() {
+		finishReason = gjson.GetBytes(jsonBytes, "response.candidates.0.finishReason")
+	}
+	terminalReason := finishReason.Exists() && strings.TrimSpace(finishReason.String()) != ""
+
+	usageMetadata := gjson.GetBytes(jsonBytes, "usageMetadata")
+	if !usageMetadata.Exists() {
+		usageMetadata = gjson.GetBytes(jsonBytes, "response.usageMetadata")
+	}
+
+	// Terminal chunk: keep as-is.
+	if terminalReason {
+		return rawJSON, false
+	}
+
+	// Nothing to strip
+	if !usageMetadata.Exists() {
+		return rawJSON, false
+	}
+
+	// Remove usageMetadata from both possible locations
+	cleaned := jsonBytes
+	var changed bool
+
+	if gjson.GetBytes(cleaned, "usageMetadata").Exists() {
+		cleaned, _ = sjson.DeleteBytes(cleaned, "usageMetadata")
+		changed = true
+	}
+
+	if gjson.GetBytes(cleaned, "response.usageMetadata").Exists() {
+		cleaned, _ = sjson.DeleteBytes(cleaned, "response.usageMetadata")
+		changed = true
+	}
+
+	return cleaned, changed
+}
+
+func hasUsageMetadata(jsonBytes []byte) bool {
+	if len(jsonBytes) == 0 || !gjson.ValidBytes(jsonBytes) {
+		return false
+	}
+	if gjson.GetBytes(jsonBytes, "usageMetadata").Exists() {
+		return true
+	}
+	if gjson.GetBytes(jsonBytes, "response.usageMetadata").Exists() {
+		return true
+	}
+	return false
+}
+
+func isStopChunkWithoutUsage(jsonBytes []byte) bool {
+	if len(jsonBytes) == 0 || !gjson.ValidBytes(jsonBytes) {
+		return false
+	}
+	finishReason := gjson.GetBytes(jsonBytes, "candidates.0.finishReason")
+	if !finishReason.Exists() {
+		finishReason = gjson.GetBytes(jsonBytes, "response.candidates.0.finishReason")
+	}
+	trimmed := strings.TrimSpace(finishReason.String())
+	if !finishReason.Exists() || trimmed == "" {
+		return false
+	}
+	return !hasUsageMetadata(jsonBytes)
+}
+
 func jsonPayload(line []byte) []byte {
 	trimmed := bytes.TrimSpace(line)
 	if len(trimmed) == 0 {

internal/translator/antigravity/claude/antigravity_claude_request.go

@@ -0,0 +1,214 @@
+// Package claude provides request translation functionality for Claude Code API compatibility.
+// This package handles the conversion of Claude Code API requests into Gemini CLI-compatible
+// JSON format, transforming message contents, system instructions, and tool declarations
+// into the format expected by Gemini CLI API clients. It performs JSON data transformation
+// to ensure compatibility between Claude Code API format and Gemini CLI API's expected format.
+package claude
+
+import (
+	"bytes"
+	"encoding/json"
+	"strings"
+
+	client "github.com/router-for-me/CLIProxyAPI/v6/internal/interfaces"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/common"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+const geminiCLIClaudeThoughtSignature = "skip_thought_signature_validator"
+
+// ConvertClaudeRequestToAntigravity parses and transforms a Claude Code API request into Gemini CLI API format.
+// It extracts the model name, system instruction, message contents, and tool declarations
+// from the raw JSON request and returns them in the format expected by the Gemini CLI API.
+// The function performs the following transformations:
+// 1. Extracts the model information from the request
+// 2. Restructures the JSON to match Gemini CLI API format
+// 3. Converts system instructions to the expected format
+// 4. Maps message contents with proper role transformations
+// 5. Handles tool declarations and tool choices
+// 6. Maps generation configuration parameters
+//
+// Parameters:
+//   - modelName: The name of the model to use for the request
+//   - rawJSON: The raw JSON request data from the Claude Code API
+//   - stream: A boolean indicating if the request is for a streaming response (unused in current implementation)
+//
+// Returns:
+//   - []byte: The transformed request data in Gemini CLI API format
+func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _ bool) []byte {
+	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON = bytes.Replace(rawJSON, []byte(`"url":{"type":"string","format":"uri",`), []byte(`"url":{"type":"string",`), -1)
+
+	// system instruction
+	var systemInstruction *client.Content
+	systemResult := gjson.GetBytes(rawJSON, "system")
+	if systemResult.IsArray() {
+		systemResults := systemResult.Array()
+		systemInstruction = &client.Content{Role: "user", Parts: []client.Part{}}
+		for i := 0; i < len(systemResults); i++ {
+			systemPromptResult := systemResults[i]
+			systemTypePromptResult := systemPromptResult.Get("type")
+			if systemTypePromptResult.Type == gjson.String && systemTypePromptResult.String() == "text" {
+				systemPrompt := systemPromptResult.Get("text").String()
+				systemPart := client.Part{Text: systemPrompt}
+				systemInstruction.Parts = append(systemInstruction.Parts, systemPart)
+			}
+		}
+		if len(systemInstruction.Parts) == 0 {
+			systemInstruction = nil
+		}
+	}
+
+	// contents
+	contents := make([]client.Content, 0)
+	messagesResult := gjson.GetBytes(rawJSON, "messages")
+	if messagesResult.IsArray() {
+		messageResults := messagesResult.Array()
+		for i := 0; i < len(messageResults); i++ {
+			messageResult := messageResults[i]
+			roleResult := messageResult.Get("role")
+			if roleResult.Type != gjson.String {
+				continue
+			}
+			role := roleResult.String()
+			if role == "assistant" {
+				role = "model"
+			}
+			clientContent := client.Content{Role: role, Parts: []client.Part{}}
+			contentsResult := messageResult.Get("content")
+			if contentsResult.IsArray() {
+				contentResults := contentsResult.Array()
+				for j := 0; j < len(contentResults); j++ {
+					contentResult := contentResults[j]
+					contentTypeResult := contentResult.Get("type")
+					if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "thinking" {
+						prompt := contentResult.Get("thinking").String()
+						signatureResult := contentResult.Get("signature")
+						signature := geminiCLIClaudeThoughtSignature
+						if signatureResult.Exists() {
+							signature = signatureResult.String()
+						}
+						clientContent.Parts = append(clientContent.Parts, client.Part{Text: prompt, Thought: true, ThoughtSignature: signature})
+					} else if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "text" {
+						prompt := contentResult.Get("text").String()
+						clientContent.Parts = append(clientContent.Parts, client.Part{Text: prompt})
+					} else if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "tool_use" {
+						functionName := contentResult.Get("name").String()
+						functionArgs := contentResult.Get("input").String()
+						functionID := contentResult.Get("id").String()
+						var args map[string]any
+						if err := json.Unmarshal([]byte(functionArgs), &args); err == nil {
+							if strings.Contains(modelName, "claude") {
+								clientContent.Parts = append(clientContent.Parts, client.Part{
+									FunctionCall: &client.FunctionCall{ID: functionID, Name: functionName, Args: args},
+								})
+							} else {
+								clientContent.Parts = append(clientContent.Parts, client.Part{
+									FunctionCall:     &client.FunctionCall{ID: functionID, Name: functionName, Args: args},
+									ThoughtSignature: geminiCLIClaudeThoughtSignature,
+								})
+							}
+						}
+					} else if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "tool_result" {
+						toolCallID := contentResult.Get("tool_use_id").String()
+						if toolCallID != "" {
+							funcName := toolCallID
+							toolCallIDs := strings.Split(toolCallID, "-")
+							if len(toolCallIDs) > 1 {
+								funcName = strings.Join(toolCallIDs[0:len(toolCallIDs)-1], "-")
+							}
+							responseData := contentResult.Get("content").Raw
+							functionResponse := client.FunctionResponse{ID: toolCallID, Name: funcName, Response: map[string]interface{}{"result": responseData}}
+							clientContent.Parts = append(clientContent.Parts, client.Part{FunctionResponse: &functionResponse})
+						}
+					} else if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "image" {
+						sourceResult := contentResult.Get("source")
+						if sourceResult.Get("type").String() == "base64" {
+							inlineData := &client.InlineData{
+								MimeType: sourceResult.Get("media_type").String(),
+								Data:     sourceResult.Get("data").String(),
+							}
+							clientContent.Parts = append(clientContent.Parts, client.Part{InlineData: inlineData})
+						}
+					}
+				}
+				contents = append(contents, clientContent)
+			} else if contentsResult.Type == gjson.String {
+				prompt := contentsResult.String()
+				contents = append(contents, client.Content{Role: role, Parts: []client.Part{{Text: prompt}}})
+			}
+		}
+	}
+
+	// tools
+	var tools []client.ToolDeclaration
+	toolsResult := gjson.GetBytes(rawJSON, "tools")
+	if toolsResult.IsArray() {
+		tools = make([]client.ToolDeclaration, 1)
+		tools[0].FunctionDeclarations = make([]any, 0)
+		toolsResults := toolsResult.Array()
+		for i := 0; i < len(toolsResults); i++ {
+			toolResult := toolsResults[i]
+			inputSchemaResult := toolResult.Get("input_schema")
+			if inputSchemaResult.Exists() && inputSchemaResult.IsObject() {
+				inputSchema := inputSchemaResult.Raw
+				tool, _ := sjson.Delete(toolResult.Raw, "input_schema")
+				tool, _ = sjson.SetRaw(tool, "parametersJsonSchema", inputSchema)
+				tool, _ = sjson.Delete(tool, "strict")
+				tool, _ = sjson.Delete(tool, "input_examples")
+				var toolDeclaration any
+				if err := json.Unmarshal([]byte(tool), &toolDeclaration); err == nil {
+					tools[0].FunctionDeclarations = append(tools[0].FunctionDeclarations, toolDeclaration)
+				}
+			}
+		}
+	} else {
+		tools = make([]client.ToolDeclaration, 0)
+	}
+
+	// Build output Gemini CLI request JSON
+	out := `{"model":"","request":{"contents":[]}}`
+	out, _ = sjson.Set(out, "model", modelName)
+	if systemInstruction != nil {
+		b, _ := json.Marshal(systemInstruction)
+		out, _ = sjson.SetRaw(out, "request.systemInstruction", string(b))
+	}
+	if len(contents) > 0 {
+		b, _ := json.Marshal(contents)
+		out, _ = sjson.SetRaw(out, "request.contents", string(b))
+	}
+	if len(tools) > 0 && len(tools[0].FunctionDeclarations) > 0 {
+		b, _ := json.Marshal(tools)
+		out, _ = sjson.SetRaw(out, "request.tools", string(b))
+	}
+
+	// Map Anthropic thinking -> Gemini thinkingBudget/include_thoughts when type==enabled
+	if t := gjson.GetBytes(rawJSON, "thinking"); t.Exists() && t.IsObject() && util.ModelSupportsThinking(modelName) {
+		if t.Get("type").String() == "enabled" {
+			if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
+				budget := int(b.Int())
+				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
+				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
+			}
+		}
+	}
+	if v := gjson.GetBytes(rawJSON, "temperature"); v.Exists() && v.Type == gjson.Number {
+		out, _ = sjson.Set(out, "request.generationConfig.temperature", v.Num)
+	}
+	if v := gjson.GetBytes(rawJSON, "top_p"); v.Exists() && v.Type == gjson.Number {
+		out, _ = sjson.Set(out, "request.generationConfig.topP", v.Num)
+	}
+	if v := gjson.GetBytes(rawJSON, "top_k"); v.Exists() && v.Type == gjson.Number {
+		out, _ = sjson.Set(out, "request.generationConfig.topK", v.Num)
+	}
+	if v := gjson.GetBytes(rawJSON, "max_tokens"); v.Exists() && v.Type == gjson.Number {
+		out, _ = sjson.Set(out, "request.generationConfig.maxOutputTokens", v.Num)
+	}
+
+	outBytes := []byte(out)
+	outBytes = common.AttachDefaultSafetySettings(outBytes, "request.safetySettings")
+
+	return outBytes
+}

internal/translator/antigravity/claude/antigravity_claude_response.go

@@ -0,0 +1,473 @@
+// Package claude provides response translation functionality for Claude Code API compatibility.
+// This package handles the conversion of backend client responses into Claude Code-compatible
+// Server-Sent Events (SSE) format, implementing a sophisticated state machine that manages
+// different response types including text content, thinking processes, and function calls.
+// The translation ensures proper sequencing of SSE events and maintains state across
+// multiple response chunks to provide a seamless streaming experience.
+package claude
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"strings"
+	"sync/atomic"
+	"time"
+
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+// Params holds parameters for response conversion and maintains state across streaming chunks.
+// This structure tracks the current state of the response translation process to ensure
+// proper sequencing of SSE events and transitions between different content types.
+type Params struct {
+	HasFirstResponse     bool   // Indicates if the initial message_start event has been sent
+	ResponseType         int    // Current response type: 0=none, 1=content, 2=thinking, 3=function
+	ResponseIndex        int    // Index counter for content blocks in the streaming response
+	HasFinishReason      bool   // Tracks whether a finish reason has been observed
+	FinishReason         string // The finish reason string returned by the provider
+	HasUsageMetadata     bool   // Tracks whether usage metadata has been observed
+	PromptTokenCount     int64  // Cached prompt token count from usage metadata
+	CandidatesTokenCount int64  // Cached candidate token count from usage metadata
+	ThoughtsTokenCount   int64  // Cached thinking token count from usage metadata
+	TotalTokenCount      int64  // Cached total token count from usage metadata
+	HasSentFinalEvents   bool   // Indicates if final content/message events have been sent
+	HasToolUse           bool   // Indicates if tool use was observed in the stream
+	HasContent           bool   // Tracks whether any content (text, thinking, or tool use) has been output
+}
+
+// toolUseIDCounter provides a process-wide unique counter for tool use identifiers.
+var toolUseIDCounter uint64
+
+// ConvertAntigravityResponseToClaude performs sophisticated streaming response format conversion.
+// This function implements a complex state machine that translates backend client responses
+// into Claude Code-compatible Server-Sent Events (SSE) format. It manages different response types
+// and handles state transitions between content blocks, thinking processes, and function calls.
+//
+// Response type states: 0=none, 1=content, 2=thinking, 3=function
+// The function maintains state across multiple calls to ensure proper SSE event sequencing.
+//
+// Parameters:
+//   - ctx: The context for the request, used for cancellation and timeout handling
+//   - modelName: The name of the model being used for the response (unused in current implementation)
+//   - rawJSON: The raw JSON response from the Gemini CLI API
+//   - param: A pointer to a parameter object for maintaining state between calls
+//
+// Returns:
+//   - []string: A slice of strings, each containing a Claude Code-compatible JSON response
+func ConvertAntigravityResponseToClaude(_ context.Context, _ string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, param *any) []string {
+	if *param == nil {
+		*param = &Params{
+			HasFirstResponse: false,
+			ResponseType:     0,
+			ResponseIndex:    0,
+		}
+	}
+
+	params := (*param).(*Params)
+
+	if bytes.Equal(rawJSON, []byte("[DONE]")) {
+		output := ""
+		// Only send final events if we have actually output content
+		if params.HasContent {
+			appendFinalEvents(params, &output, true)
+			return []string{
+				output + "event: message_stop\ndata: {\"type\":\"message_stop\"}\n\n\n",
+			}
+		}
+		return []string{}
+	}
+
+	output := ""
+
+	// Initialize the streaming session with a message_start event
+	// This is only sent for the very first response chunk to establish the streaming session
+	if !params.HasFirstResponse {
+		output = "event: message_start\n"
+
+		// Create the initial message structure with default values according to Claude Code API specification
+		// This follows the Claude Code API specification for streaming message initialization
+		messageStartTemplate := `{"type": "message_start", "message": {"id": "msg_1nZdL29xx5MUA1yADyHTEsnR8uuvGzszyY", "type": "message", "role": "assistant", "content": [], "model": "claude-3-5-sonnet-20241022", "stop_reason": null, "stop_sequence": null, "usage": {"input_tokens": 0, "output_tokens": 0}}}`
+
+		// Override default values with actual response metadata if available from the Gemini CLI response
+		if modelVersionResult := gjson.GetBytes(rawJSON, "response.modelVersion"); modelVersionResult.Exists() {
+			messageStartTemplate, _ = sjson.Set(messageStartTemplate, "message.model", modelVersionResult.String())
+		}
+		if responseIDResult := gjson.GetBytes(rawJSON, "response.responseId"); responseIDResult.Exists() {
+			messageStartTemplate, _ = sjson.Set(messageStartTemplate, "message.id", responseIDResult.String())
+		}
+		output = output + fmt.Sprintf("data: %s\n\n\n", messageStartTemplate)
+
+		params.HasFirstResponse = true
+	}
+
+	// Process the response parts array from the backend client
+	// Each part can contain text content, thinking content, or function calls
+	partsResult := gjson.GetBytes(rawJSON, "response.candidates.0.content.parts")
+	if partsResult.IsArray() {
+		partResults := partsResult.Array()
+		for i := 0; i < len(partResults); i++ {
+			partResult := partResults[i]
+
+			// Extract the different types of content from each part
+			partTextResult := partResult.Get("text")
+			functionCallResult := partResult.Get("functionCall")
+
+			// Handle text content (both regular content and thinking)
+			if partTextResult.Exists() {
+				// Process thinking content (internal reasoning)
+				if partResult.Get("thought").Bool() {
+					if thoughtSignature := partResult.Get("thoughtSignature"); thoughtSignature.Exists() && thoughtSignature.String() != "" {
+						output = output + "event: content_block_delta\n"
+						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"signature_delta","signature":""}}`, params.ResponseIndex), "delta.signature", thoughtSignature.String())
+						output = output + fmt.Sprintf("data: %s\n\n\n", data)
+						params.HasContent = true
+					} else if params.ResponseType == 2 { // Continue existing thinking block if already in thinking state
+						output = output + "event: content_block_delta\n"
+						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"thinking_delta","thinking":""}}`, params.ResponseIndex), "delta.thinking", partTextResult.String())
+						output = output + fmt.Sprintf("data: %s\n\n\n", data)
+						params.HasContent = true
+					} else {
+						// Transition from another state to thinking
+						// First, close any existing content block
+						if params.ResponseType != 0 {
+							if params.ResponseType == 2 {
+								// output = output + "event: content_block_delta\n"
+								// output = output + fmt.Sprintf(`data: {"type":"content_block_delta","index":%d,"delta":{"type":"signature_delta","signature":null}}`, params.ResponseIndex)
+								// output = output + "\n\n\n"
+							}
+							output = output + "event: content_block_stop\n"
+							output = output + fmt.Sprintf(`data: {"type":"content_block_stop","index":%d}`, params.ResponseIndex)
+							output = output + "\n\n\n"
+							params.ResponseIndex++
+						}
+
+						// Start a new thinking content block
+						output = output + "event: content_block_start\n"
+						output = output + fmt.Sprintf(`data: {"type":"content_block_start","index":%d,"content_block":{"type":"thinking","thinking":""}}`, params.ResponseIndex)
+						output = output + "\n\n\n"
+						output = output + "event: content_block_delta\n"
+						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"thinking_delta","thinking":""}}`, params.ResponseIndex), "delta.thinking", partTextResult.String())
+						output = output + fmt.Sprintf("data: %s\n\n\n", data)
+						params.ResponseType = 2 // Set state to thinking
+						params.HasContent = true
+					}
+				} else {
+					finishReasonResult := gjson.GetBytes(rawJSON, "response.candidates.0.finishReason")
+					if partTextResult.String() != "" || !finishReasonResult.Exists() {
+						// Process regular text content (user-visible output)
+						// Continue existing text block if already in content state
+						if params.ResponseType == 1 {
+							output = output + "event: content_block_delta\n"
+							data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"text_delta","text":""}}`, params.ResponseIndex), "delta.text", partTextResult.String())
+							output = output + fmt.Sprintf("data: %s\n\n\n", data)
+							params.HasContent = true
+						} else {
+							// Transition from another state to text content
+							// First, close any existing content block
+							if params.ResponseType != 0 {
+								if params.ResponseType == 2 {
+									// output = output + "event: content_block_delta\n"
+									// output = output + fmt.Sprintf(`data: {"type":"content_block_delta","index":%d,"delta":{"type":"signature_delta","signature":null}}`, params.ResponseIndex)
+									// output = output + "\n\n\n"
+								}
+								output = output + "event: content_block_stop\n"
+								output = output + fmt.Sprintf(`data: {"type":"content_block_stop","index":%d}`, params.ResponseIndex)
+								output = output + "\n\n\n"
+								params.ResponseIndex++
+							}
+							if partTextResult.String() != "" {
+								// Start a new text content block
+								output = output + "event: content_block_start\n"
+								output = output + fmt.Sprintf(`data: {"type":"content_block_start","index":%d,"content_block":{"type":"text","text":""}}`, params.ResponseIndex)
+								output = output + "\n\n\n"
+								output = output + "event: content_block_delta\n"
+								data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"text_delta","text":""}}`, params.ResponseIndex), "delta.text", partTextResult.String())
+								output = output + fmt.Sprintf("data: %s\n\n\n", data)
+								params.ResponseType = 1 // Set state to content
+								params.HasContent = true
+							}
+						}
+					}
+				}
+			} else if functionCallResult.Exists() {
+				// Handle function/tool calls from the AI model
+				// This processes tool usage requests and formats them for Claude Code API compatibility
+				params.HasToolUse = true
+				fcName := functionCallResult.Get("name").String()
+
+				// Handle state transitions when switching to function calls
+				// Close any existing function call block first
+				if params.ResponseType == 3 {
+					output = output + "event: content_block_stop\n"
+					output = output + fmt.Sprintf(`data: {"type":"content_block_stop","index":%d}`, params.ResponseIndex)
+					output = output + "\n\n\n"
+					params.ResponseIndex++
+					params.ResponseType = 0
+				}
+
+				// Special handling for thinking state transition
+				if params.ResponseType == 2 {
+					// output = output + "event: content_block_delta\n"
+					// output = output + fmt.Sprintf(`data: {"type":"content_block_delta","index":%d,"delta":{"type":"signature_delta","signature":null}}`, params.ResponseIndex)
+					// output = output + "\n\n\n"
+				}
+
+				// Close any other existing content block
+				if params.ResponseType != 0 {
+					output = output + "event: content_block_stop\n"
+					output = output + fmt.Sprintf(`data: {"type":"content_block_stop","index":%d}`, params.ResponseIndex)
+					output = output + "\n\n\n"
+					params.ResponseIndex++
+				}
+
+				// Start a new tool use content block
+				// This creates the structure for a function call in Claude Code format
+				output = output + "event: content_block_start\n"
+
+				// Create the tool use block with unique ID and function details
+				data := fmt.Sprintf(`{"type":"content_block_start","index":%d,"content_block":{"type":"tool_use","id":"","name":"","input":{}}}`, params.ResponseIndex)
+				data, _ = sjson.Set(data, "content_block.id", fmt.Sprintf("%s-%d-%d", fcName, time.Now().UnixNano(), atomic.AddUint64(&toolUseIDCounter, 1)))
+				data, _ = sjson.Set(data, "content_block.name", fcName)
+				output = output + fmt.Sprintf("data: %s\n\n\n", data)
+
+				if fcArgsResult := functionCallResult.Get("args"); fcArgsResult.Exists() {
+					output = output + "event: content_block_delta\n"
+					data, _ = sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"input_json_delta","partial_json":""}}`, params.ResponseIndex), "delta.partial_json", fcArgsResult.Raw)
+					output = output + fmt.Sprintf("data: %s\n\n\n", data)
+				}
+				params.ResponseType = 3
+				params.HasContent = true
+			}
+		}
+	}
+
+	if finishReasonResult := gjson.GetBytes(rawJSON, "response.candidates.0.finishReason"); finishReasonResult.Exists() {
+		params.HasFinishReason = true
+		params.FinishReason = finishReasonResult.String()
+	}
+
+	if usageResult := gjson.GetBytes(rawJSON, "response.usageMetadata"); usageResult.Exists() {
+		params.HasUsageMetadata = true
+		params.PromptTokenCount = usageResult.Get("promptTokenCount").Int()
+		params.CandidatesTokenCount = usageResult.Get("candidatesTokenCount").Int()
+		params.ThoughtsTokenCount = usageResult.Get("thoughtsTokenCount").Int()
+		params.TotalTokenCount = usageResult.Get("totalTokenCount").Int()
+		if params.CandidatesTokenCount == 0 && params.TotalTokenCount > 0 {
+			params.CandidatesTokenCount = params.TotalTokenCount - params.PromptTokenCount - params.ThoughtsTokenCount
+			if params.CandidatesTokenCount < 0 {
+				params.CandidatesTokenCount = 0
+			}
+		}
+	}
+
+	if params.HasUsageMetadata && params.HasFinishReason {
+		appendFinalEvents(params, &output, false)
+	}
+
+	return []string{output}
+}
+
+func appendFinalEvents(params *Params, output *string, force bool) {
+	if params.HasSentFinalEvents {
+		return
+	}
+
+	if !params.HasUsageMetadata && !force {
+		return
+	}
+
+	// Only send final events if we have actually output content
+	if !params.HasContent {
+		return
+	}
+
+	if params.ResponseType != 0 {
+		*output = *output + "event: content_block_stop\n"
+		*output = *output + fmt.Sprintf(`data: {"type":"content_block_stop","index":%d}`, params.ResponseIndex)
+		*output = *output + "\n\n\n"
+		params.ResponseType = 0
+	}
+
+	stopReason := resolveStopReason(params)
+	usageOutputTokens := params.CandidatesTokenCount + params.ThoughtsTokenCount
+	if usageOutputTokens == 0 && params.TotalTokenCount > 0 {
+		usageOutputTokens = params.TotalTokenCount - params.PromptTokenCount
+		if usageOutputTokens < 0 {
+			usageOutputTokens = 0
+		}
+	}
+
+	*output = *output + "event: message_delta\n"
+	*output = *output + "data: "
+	delta := fmt.Sprintf(`{"type":"message_delta","delta":{"stop_reason":"%s","stop_sequence":null},"usage":{"input_tokens":%d,"output_tokens":%d}}`, stopReason, params.PromptTokenCount, usageOutputTokens)
+	*output = *output + delta + "\n\n\n"
+
+	params.HasSentFinalEvents = true
+}
+
+func resolveStopReason(params *Params) string {
+	if params.HasToolUse {
+		return "tool_use"
+	}
+
+	switch params.FinishReason {
+	case "MAX_TOKENS":
+		return "max_tokens"
+	case "STOP", "FINISH_REASON_UNSPECIFIED", "UNKNOWN":
+		return "end_turn"
+	}
+
+	return "end_turn"
+}
+
+// ConvertAntigravityResponseToClaudeNonStream converts a non-streaming Gemini CLI response to a non-streaming Claude response.
+//
+// Parameters:
+//   - ctx: The context for the request.
+//   - modelName: The name of the model.
+//   - rawJSON: The raw JSON response from the Gemini CLI API.
+//   - param: A pointer to a parameter object for the conversion.
+//
+// Returns:
+//   - string: A Claude-compatible JSON response.
+func ConvertAntigravityResponseToClaudeNonStream(_ context.Context, _ string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, _ *any) string {
+	_ = originalRequestRawJSON
+	_ = requestRawJSON
+
+	root := gjson.ParseBytes(rawJSON)
+	promptTokens := root.Get("response.usageMetadata.promptTokenCount").Int()
+	candidateTokens := root.Get("response.usageMetadata.candidatesTokenCount").Int()
+	thoughtTokens := root.Get("response.usageMetadata.thoughtsTokenCount").Int()
+	totalTokens := root.Get("response.usageMetadata.totalTokenCount").Int()
+	outputTokens := candidateTokens + thoughtTokens
+	if outputTokens == 0 && totalTokens > 0 {
+		outputTokens = totalTokens - promptTokens
+		if outputTokens < 0 {
+			outputTokens = 0
+		}
+	}
+
+	response := map[string]interface{}{
+		"id":            root.Get("response.responseId").String(),
+		"type":          "message",
+		"role":          "assistant",
+		"model":         root.Get("response.modelVersion").String(),
+		"content":       []interface{}{},
+		"stop_reason":   nil,
+		"stop_sequence": nil,
+		"usage": map[string]interface{}{
+			"input_tokens":  promptTokens,
+			"output_tokens": outputTokens,
+		},
+	}
+
+	parts := root.Get("response.candidates.0.content.parts")
+	var contentBlocks []interface{}
+	textBuilder := strings.Builder{}
+	thinkingBuilder := strings.Builder{}
+	toolIDCounter := 0
+	hasToolCall := false
+
+	flushText := func() {
+		if textBuilder.Len() == 0 {
+			return
+		}
+		contentBlocks = append(contentBlocks, map[string]interface{}{
+			"type": "text",
+			"text": textBuilder.String(),
+		})
+		textBuilder.Reset()
+	}
+
+	flushThinking := func() {
+		if thinkingBuilder.Len() == 0 {
+			return
+		}
+		contentBlocks = append(contentBlocks, map[string]interface{}{
+			"type":     "thinking",
+			"thinking": thinkingBuilder.String(),
+		})
+		thinkingBuilder.Reset()
+	}
+
+	if parts.IsArray() {
+		for _, part := range parts.Array() {
+			if text := part.Get("text"); text.Exists() && text.String() != "" {
+				if part.Get("thought").Bool() {
+					flushText()
+					thinkingBuilder.WriteString(text.String())
+					continue
+				}
+				flushThinking()
+				textBuilder.WriteString(text.String())
+				continue
+			}
+
+			if functionCall := part.Get("functionCall"); functionCall.Exists() {
+				flushThinking()
+				flushText()
+				hasToolCall = true
+
+				name := functionCall.Get("name").String()
+				toolIDCounter++
+				toolBlock := map[string]interface{}{
+					"type":  "tool_use",
+					"id":    fmt.Sprintf("tool_%d", toolIDCounter),
+					"name":  name,
+					"input": map[string]interface{}{},
+				}
+
+				if args := functionCall.Get("args"); args.Exists() {
+					var parsed interface{}
+					if err := json.Unmarshal([]byte(args.Raw), &parsed); err == nil {
+						toolBlock["input"] = parsed
+					}
+				}
+
+				contentBlocks = append(contentBlocks, toolBlock)
+				continue
+			}
+		}
+	}
+
+	flushThinking()
+	flushText()
+
+	response["content"] = contentBlocks
+
+	stopReason := "end_turn"
+	if hasToolCall {
+		stopReason = "tool_use"
+	} else {
+		if finish := root.Get("response.candidates.0.finishReason"); finish.Exists() {
+			switch finish.String() {
+			case "MAX_TOKENS":
+				stopReason = "max_tokens"
+			case "STOP", "FINISH_REASON_UNSPECIFIED", "UNKNOWN":
+				stopReason = "end_turn"
+			default:
+				stopReason = "end_turn"
+			}
+		}
+	}
+	response["stop_reason"] = stopReason
+
+	if usage := response["usage"].(map[string]interface{}); usage["input_tokens"] == int64(0) && usage["output_tokens"] == int64(0) {
+		if usageMeta := root.Get("response.usageMetadata"); !usageMeta.Exists() {
+			delete(response, "usage")
+		}
+	}
+
+	encoded, err := json.Marshal(response)
+	if err != nil {
+		return ""
+	}
+	return string(encoded)
+}
+
+func ClaudeTokenCount(ctx context.Context, count int64) string {
+	return fmt.Sprintf(`{"input_tokens":%d}`, count)
+}

internal/translator/antigravity/claude/init.go

@@ -0,0 +1,20 @@
+package claude
+
+import (
+	. "github.com/router-for-me/CLIProxyAPI/v6/internal/constant"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/interfaces"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/translator/translator"
+)
+
+func init() {
+	translator.Register(
+		Claude,
+		Antigravity,
+		ConvertClaudeRequestToAntigravity,
+		interfaces.TranslateResponse{
+			Stream:     ConvertAntigravityResponseToClaude,
+			NonStream:  ConvertAntigravityResponseToClaudeNonStream,
+			TokenCount: ClaudeTokenCount,
+		},
+	)
+}

internal/translator/antigravity/gemini/antigravity_gemini_request.go

@@ -0,0 +1,293 @@
+// Package gemini provides request translation functionality for Gemini CLI to Gemini API compatibility.
+// It handles parsing and transforming Gemini CLI API requests into Gemini API format,
+// extracting model information, system instructions, message contents, and tool declarations.
+// The package performs JSON data transformation to ensure compatibility
+// between Gemini CLI API format and Gemini API's expected format.
+package gemini
+
+import (
+	"bytes"
+	"encoding/json"
+	"fmt"
+
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/common"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
+	log "github.com/sirupsen/logrus"
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+// ConvertGeminiRequestToAntigravity parses and transforms a Gemini CLI API request into Gemini API format.
+// It extracts the model name, system instruction, message contents, and tool declarations
+// from the raw JSON request and returns them in the format expected by the Gemini API.
+// The function performs the following transformations:
+// 1. Extracts the model information from the request
+// 2. Restructures the JSON to match Gemini API format
+// 3. Converts system instructions to the expected format
+// 4. Fixes CLI tool response format and grouping
+//
+// Parameters:
+//   - modelName: The name of the model to use for the request (unused in current implementation)
+//   - rawJSON: The raw JSON request data from the Gemini CLI API
+//   - stream: A boolean indicating if the request is for a streaming response (unused in current implementation)
+//
+// Returns:
+//   - []byte: The transformed request data in Gemini API format
+func ConvertGeminiRequestToAntigravity(_ string, inputRawJSON []byte, _ bool) []byte {
+	rawJSON := bytes.Clone(inputRawJSON)
+	template := ""
+	template = `{"project":"","request":{},"model":""}`
+	template, _ = sjson.SetRaw(template, "request", string(rawJSON))
+	template, _ = sjson.Set(template, "model", gjson.Get(template, "request.model").String())
+	template, _ = sjson.Delete(template, "request.model")
+
+	template, errFixCLIToolResponse := fixCLIToolResponse(template)
+	if errFixCLIToolResponse != nil {
+		return []byte{}
+	}
+
+	systemInstructionResult := gjson.Get(template, "request.system_instruction")
+	if systemInstructionResult.Exists() {
+		template, _ = sjson.SetRaw(template, "request.systemInstruction", systemInstructionResult.Raw)
+		template, _ = sjson.Delete(template, "request.system_instruction")
+	}
+	rawJSON = []byte(template)
+
+	// Normalize roles in request.contents: default to valid values if missing/invalid
+	contents := gjson.GetBytes(rawJSON, "request.contents")
+	if contents.Exists() {
+		prevRole := ""
+		idx := 0
+		contents.ForEach(func(_ gjson.Result, value gjson.Result) bool {
+			role := value.Get("role").String()
+			valid := role == "user" || role == "model"
+			if role == "" || !valid {
+				var newRole string
+				if prevRole == "" {
+					newRole = "user"
+				} else if prevRole == "user" {
+					newRole = "model"
+				} else {
+					newRole = "user"
+				}
+				path := fmt.Sprintf("request.contents.%d.role", idx)
+				rawJSON, _ = sjson.SetBytes(rawJSON, path, newRole)
+				role = newRole
+			}
+			prevRole = role
+			idx++
+			return true
+		})
+	}
+
+	toolsResult := gjson.GetBytes(rawJSON, "request.tools")
+	if toolsResult.Exists() && toolsResult.IsArray() {
+		toolResults := toolsResult.Array()
+		for i := 0; i < len(toolResults); i++ {
+			functionDeclarationsResult := gjson.GetBytes(rawJSON, fmt.Sprintf("request.tools.%d.function_declarations", i))
+			if functionDeclarationsResult.Exists() && functionDeclarationsResult.IsArray() {
+				functionDeclarationsResults := functionDeclarationsResult.Array()
+				for j := 0; j < len(functionDeclarationsResults); j++ {
+					parametersResult := gjson.GetBytes(rawJSON, fmt.Sprintf("request.tools.%d.function_declarations.%d.parameters", i, j))
+					if parametersResult.Exists() {
+						strJson, _ := util.RenameKey(string(rawJSON), fmt.Sprintf("request.tools.%d.function_declarations.%d.parameters", i, j), fmt.Sprintf("request.tools.%d.function_declarations.%d.parametersJsonSchema", i, j))
+						rawJSON = []byte(strJson)
+					}
+				}
+			}
+		}
+	}
+
+	gjson.GetBytes(rawJSON, "request.contents").ForEach(func(key, content gjson.Result) bool {
+		if content.Get("role").String() == "model" {
+			content.Get("parts").ForEach(func(partKey, part gjson.Result) bool {
+				if part.Get("functionCall").Exists() {
+					rawJSON, _ = sjson.SetBytes(rawJSON, fmt.Sprintf("request.contents.%d.parts.%d.thoughtSignature", key.Int(), partKey.Int()), "skip_thought_signature_validator")
+				} else if part.Get("thoughtSignature").Exists() {
+					rawJSON, _ = sjson.SetBytes(rawJSON, fmt.Sprintf("request.contents.%d.parts.%d.thoughtSignature", key.Int(), partKey.Int()), "skip_thought_signature_validator")
+				}
+				return true
+			})
+		}
+		return true
+	})
+
+	return common.AttachDefaultSafetySettings(rawJSON, "request.safetySettings")
+}
+
+// FunctionCallGroup represents a group of function calls and their responses
+type FunctionCallGroup struct {
+	ModelContent    map[string]interface{}
+	FunctionCalls   []gjson.Result
+	ResponsesNeeded int
+}
+
+// fixCLIToolResponse performs sophisticated tool response format conversion and grouping.
+// This function transforms the CLI tool response format by intelligently grouping function calls
+// with their corresponding responses, ensuring proper conversation flow and API compatibility.
+// It converts from a linear format (1.json) to a grouped format (2.json) where function calls
+// and their responses are properly associated and structured.
+//
+// Parameters:
+//   - input: The input JSON string to be processed
+//
+// Returns:
+//   - string: The processed JSON string with grouped function calls and responses
+//   - error: An error if the processing fails
+func fixCLIToolResponse(input string) (string, error) {
+	// Parse the input JSON to extract the conversation structure
+	parsed := gjson.Parse(input)
+
+	// Extract the contents array which contains the conversation messages
+	contents := parsed.Get("request.contents")
+	if !contents.Exists() {
+		// log.Debugf(input)
+		return input, fmt.Errorf("contents not found in input")
+	}
+
+	// Initialize data structures for processing and grouping
+	var newContents []interface{}          // Final processed contents array
+	var pendingGroups []*FunctionCallGroup // Groups awaiting completion with responses
+	var collectedResponses []gjson.Result  // Standalone responses to be matched
+
+	// Process each content object in the conversation
+	// This iterates through messages and groups function calls with their responses
+	contents.ForEach(func(key, value gjson.Result) bool {
+		role := value.Get("role").String()
+		parts := value.Get("parts")
+
+		// Check if this content has function responses
+		var responsePartsInThisContent []gjson.Result
+		parts.ForEach(func(_, part gjson.Result) bool {
+			if part.Get("functionResponse").Exists() {
+				responsePartsInThisContent = append(responsePartsInThisContent, part)
+			}
+			return true
+		})
+
+		// If this content has function responses, collect them
+		if len(responsePartsInThisContent) > 0 {
+			collectedResponses = append(collectedResponses, responsePartsInThisContent...)
+
+			// Check if any pending groups can be satisfied
+			for i := len(pendingGroups) - 1; i >= 0; i-- {
+				group := pendingGroups[i]
+				if len(collectedResponses) >= group.ResponsesNeeded {
+					// Take the needed responses for this group
+					groupResponses := collectedResponses[:group.ResponsesNeeded]
+					collectedResponses = collectedResponses[group.ResponsesNeeded:]
+
+					// Create merged function response content
+					var responseParts []interface{}
+					for _, response := range groupResponses {
+						var responseMap map[string]interface{}
+						errUnmarshal := json.Unmarshal([]byte(response.Raw), &responseMap)
+						if errUnmarshal != nil {
+							log.Warnf("failed to unmarshal function response: %v\n", errUnmarshal)
+							continue
+						}
+						responseParts = append(responseParts, responseMap)
+					}
+
+					if len(responseParts) > 0 {
+						functionResponseContent := map[string]interface{}{
+							"parts": responseParts,
+							"role":  "function",
+						}
+						newContents = append(newContents, functionResponseContent)
+					}
+
+					// Remove this group as it's been satisfied
+					pendingGroups = append(pendingGroups[:i], pendingGroups[i+1:]...)
+					break
+				}
+			}
+
+			return true // Skip adding this content, responses are merged
+		}
+
+		// If this is a model with function calls, create a new group
+		if role == "model" {
+			var functionCallsInThisModel []gjson.Result
+			parts.ForEach(func(_, part gjson.Result) bool {
+				if part.Get("functionCall").Exists() {
+					functionCallsInThisModel = append(functionCallsInThisModel, part)
+				}
+				return true
+			})
+
+			if len(functionCallsInThisModel) > 0 {
+				// Add the model content
+				var contentMap map[string]interface{}
+				errUnmarshal := json.Unmarshal([]byte(value.Raw), &contentMap)
+				if errUnmarshal != nil {
+					log.Warnf("failed to unmarshal model content: %v\n", errUnmarshal)
+					return true
+				}
+				newContents = append(newContents, contentMap)
+
+				// Create a new group for tracking responses
+				group := &FunctionCallGroup{
+					ModelContent:    contentMap,
+					FunctionCalls:   functionCallsInThisModel,
+					ResponsesNeeded: len(functionCallsInThisModel),
+				}
+				pendingGroups = append(pendingGroups, group)
+			} else {
+				// Regular model content without function calls
+				var contentMap map[string]interface{}
+				errUnmarshal := json.Unmarshal([]byte(value.Raw), &contentMap)
+				if errUnmarshal != nil {
+					log.Warnf("failed to unmarshal content: %v\n", errUnmarshal)
+					return true
+				}
+				newContents = append(newContents, contentMap)
+			}
+		} else {
+			// Non-model content (user, etc.)
+			var contentMap map[string]interface{}
+			errUnmarshal := json.Unmarshal([]byte(value.Raw), &contentMap)
+			if errUnmarshal != nil {
+				log.Warnf("failed to unmarshal content: %v\n", errUnmarshal)
+				return true
+			}
+			newContents = append(newContents, contentMap)
+		}
+
+		return true
+	})
+
+	// Handle any remaining pending groups with remaining responses
+	for _, group := range pendingGroups {
+		if len(collectedResponses) >= group.ResponsesNeeded {
+			groupResponses := collectedResponses[:group.ResponsesNeeded]
+			collectedResponses = collectedResponses[group.ResponsesNeeded:]
+
+			var responseParts []interface{}
+			for _, response := range groupResponses {
+				var responseMap map[string]interface{}
+				errUnmarshal := json.Unmarshal([]byte(response.Raw), &responseMap)
+				if errUnmarshal != nil {
+					log.Warnf("failed to unmarshal function response: %v\n", errUnmarshal)
+					continue
+				}
+				responseParts = append(responseParts, responseMap)
+			}
+
+			if len(responseParts) > 0 {
+				functionResponseContent := map[string]interface{}{
+					"parts": responseParts,
+					"role":  "function",
+				}
+				newContents = append(newContents, functionResponseContent)
+			}
+		}
+	}
+
+	// Update the original JSON with the new contents
+	result := input
+	newContentsJSON, _ := json.Marshal(newContents)
+	result, _ = sjson.Set(result, "request.contents", json.RawMessage(newContentsJSON))
+
+	return result, nil
+}

internal/translator/antigravity/gemini/antigravity_gemini_response.go

@@ -0,0 +1,86 @@
+// Package gemini provides request translation functionality for Gemini to Gemini CLI API compatibility.
+// It handles parsing and transforming Gemini API requests into Gemini CLI API format,
+// extracting model information, system instructions, message contents, and tool declarations.
+// The package performs JSON data transformation to ensure compatibility
+// between Gemini API format and Gemini CLI API's expected format.
+package gemini
+
+import (
+	"bytes"
+	"context"
+	"fmt"
+
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+// ConvertAntigravityResponseToGemini parses and transforms a Gemini CLI API request into Gemini API format.
+// It extracts the model name, system instruction, message contents, and tool declarations
+// from the raw JSON request and returns them in the format expected by the Gemini API.
+// The function performs the following transformations:
+// 1. Extracts the response data from the request
+// 2. Handles alternative response formats
+// 3. Processes array responses by extracting individual response objects
+//
+// Parameters:
+//   - ctx: The context for the request, used for cancellation and timeout handling
+//   - modelName: The name of the model to use for the request (unused in current implementation)
+//   - rawJSON: The raw JSON request data from the Gemini CLI API
+//   - param: A pointer to a parameter object for the conversion (unused in current implementation)
+//
+// Returns:
+//   - []string: The transformed request data in Gemini API format
+func ConvertAntigravityResponseToGemini(ctx context.Context, _ string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, _ *any) []string {
+	if bytes.HasPrefix(rawJSON, []byte("data:")) {
+		rawJSON = bytes.TrimSpace(rawJSON[5:])
+	}
+
+	if alt, ok := ctx.Value("alt").(string); ok {
+		var chunk []byte
+		if alt == "" {
+			responseResult := gjson.GetBytes(rawJSON, "response")
+			if responseResult.Exists() {
+				chunk = []byte(responseResult.Raw)
+			}
+		} else {
+			chunkTemplate := "[]"
+			responseResult := gjson.ParseBytes(chunk)
+			if responseResult.IsArray() {
+				responseResultItems := responseResult.Array()
+				for i := 0; i < len(responseResultItems); i++ {
+					responseResultItem := responseResultItems[i]
+					if responseResultItem.Get("response").Exists() {
+						chunkTemplate, _ = sjson.SetRaw(chunkTemplate, "-1", responseResultItem.Get("response").Raw)
+					}
+				}
+			}
+			chunk = []byte(chunkTemplate)
+		}
+		return []string{string(chunk)}
+	}
+	return []string{}
+}
+
+// ConvertAntigravityResponseToGeminiNonStream converts a non-streaming Gemini CLI request to a non-streaming Gemini response.
+// This function processes the complete Gemini CLI request and transforms it into a single Gemini-compatible
+// JSON response. It extracts the response data from the request and returns it in the expected format.
+//
+// Parameters:
+//   - ctx: The context for the request, used for cancellation and timeout handling
+//   - modelName: The name of the model being used for the response (unused in current implementation)
+//   - rawJSON: The raw JSON request data from the Gemini CLI API
+//   - param: A pointer to a parameter object for the conversion (unused in current implementation)
+//
+// Returns:
+//   - string: A Gemini-compatible JSON response containing the response data
+func ConvertAntigravityResponseToGeminiNonStream(_ context.Context, _ string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, _ *any) string {
+	responseResult := gjson.GetBytes(rawJSON, "response")
+	if responseResult.Exists() {
+		return responseResult.Raw
+	}
+	return string(rawJSON)
+}
+
+func GeminiTokenCount(ctx context.Context, count int64) string {
+	return fmt.Sprintf(`{"totalTokens":%d,"promptTokensDetails":[{"modality":"TEXT","tokenCount":%d}]}`, count, count)
+}

internal/translator/antigravity/gemini/init.go

@@ -0,0 +1,20 @@
+package gemini
+
+import (
+	. "github.com/router-for-me/CLIProxyAPI/v6/internal/constant"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/interfaces"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/translator/translator"
+)
+
+func init() {
+	translator.Register(
+		Gemini,
+		Antigravity,
+		ConvertGeminiRequestToAntigravity,
+		interfaces.TranslateResponse{
+			Stream:     ConvertAntigravityResponseToGemini,
+			NonStream:  ConvertAntigravityResponseToGeminiNonStream,
+			TokenCount: GeminiTokenCount,
+		},
+	)
+}

internal/translator/antigravity/openai/chat-completions/antigravity_openai_request.go

@@ -0,0 +1,396 @@
+// Package openai provides request translation functionality for OpenAI to Gemini CLI API compatibility.
+// It converts OpenAI Chat Completions requests into Gemini CLI compatible JSON using gjson/sjson only.
+package chat_completions
+
+import (
+	"bytes"
+	"fmt"
+	"strings"
+
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/misc"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/common"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
+	log "github.com/sirupsen/logrus"
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+const geminiCLIFunctionThoughtSignature = "skip_thought_signature_validator"
+
+// ConvertOpenAIRequestToAntigravity converts an OpenAI Chat Completions request (raw JSON)
+// into a complete Gemini CLI request JSON. All JSON construction uses sjson and lookups use gjson.
+//
+// Parameters:
+//   - modelName: The name of the model to use for the request
+//   - rawJSON: The raw JSON request data from the OpenAI API
+//   - stream: A boolean indicating if the request is for a streaming response (unused in current implementation)
+//
+// Returns:
+//   - []byte: The transformed request data in Gemini CLI API format
+func ConvertOpenAIRequestToAntigravity(modelName string, inputRawJSON []byte, _ bool) []byte {
+	rawJSON := bytes.Clone(inputRawJSON)
+	// Base envelope (no default thinkingConfig)
+	out := []byte(`{"project":"","request":{"contents":[]},"model":"gemini-2.5-pro"}`)
+
+	// Model
+	out, _ = sjson.SetBytes(out, "model", modelName)
+
+	// Reasoning effort -> thinkingBudget/include_thoughts
+	// Note: OpenAI official fields take precedence over extra_body.google.thinking_config
+	re := gjson.GetBytes(rawJSON, "reasoning_effort")
+	hasOfficialThinking := re.Exists()
+	if hasOfficialThinking && util.ModelSupportsThinking(modelName) {
+		switch re.String() {
+		case "none":
+			out, _ = sjson.DeleteBytes(out, "request.generationConfig.thinkingConfig.include_thoughts")
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", 0)
+		case "auto":
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", -1)
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
+		case "low":
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", 1024)
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
+		case "medium":
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", 8192)
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
+		case "high":
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", 32768)
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
+		default:
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", -1)
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
+		}
+	}
+
+	// Cherry Studio extension extra_body.google.thinking_config (effective only when official fields are absent)
+	if !hasOfficialThinking && util.ModelSupportsThinking(modelName) {
+		if tc := gjson.GetBytes(rawJSON, "extra_body.google.thinking_config"); tc.Exists() && tc.IsObject() {
+			var setBudget bool
+			var budget int
+
+			if v := tc.Get("thinkingBudget"); v.Exists() {
+				budget = int(v.Int())
+				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
+				setBudget = true
+			} else if v := tc.Get("thinking_budget"); v.Exists() {
+				budget = int(v.Int())
+				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
+				setBudget = true
+			}
+
+			if v := tc.Get("includeThoughts"); v.Exists() {
+				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", v.Bool())
+			} else if v := tc.Get("include_thoughts"); v.Exists() {
+				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", v.Bool())
+			} else if setBudget && budget != 0 {
+				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
+			}
+		}
+	}
+
+	// Claude/Anthropic API format: thinking.type == "enabled" with budget_tokens
+	// This allows Claude Code and other Claude API clients to pass thinking configuration
+	if !gjson.GetBytes(out, "request.generationConfig.thinkingConfig").Exists() && util.ModelSupportsThinking(modelName) {
+		if t := gjson.GetBytes(rawJSON, "thinking"); t.Exists() && t.IsObject() {
+			if t.Get("type").String() == "enabled" {
+				if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
+					budget := int(b.Int())
+					out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
+					out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
+				}
+			}
+		}
+	}
+
+	// Temperature/top_p/top_k/max_tokens
+	if tr := gjson.GetBytes(rawJSON, "temperature"); tr.Exists() && tr.Type == gjson.Number {
+		out, _ = sjson.SetBytes(out, "request.generationConfig.temperature", tr.Num)
+	}
+	if tpr := gjson.GetBytes(rawJSON, "top_p"); tpr.Exists() && tpr.Type == gjson.Number {
+		out, _ = sjson.SetBytes(out, "request.generationConfig.topP", tpr.Num)
+	}
+	if tkr := gjson.GetBytes(rawJSON, "top_k"); tkr.Exists() && tkr.Type == gjson.Number {
+		out, _ = sjson.SetBytes(out, "request.generationConfig.topK", tkr.Num)
+	}
+	if maxTok := gjson.GetBytes(rawJSON, "max_tokens"); maxTok.Exists() && maxTok.Type == gjson.Number {
+		out, _ = sjson.SetBytes(out, "request.generationConfig.maxOutputTokens", maxTok.Num)
+	}
+
+	// Map OpenAI modalities -> Gemini CLI request.generationConfig.responseModalities
+	// e.g. "modalities": ["image", "text"] -> ["IMAGE", "TEXT"]
+	if mods := gjson.GetBytes(rawJSON, "modalities"); mods.Exists() && mods.IsArray() {
+		var responseMods []string
+		for _, m := range mods.Array() {
+			switch strings.ToLower(m.String()) {
+			case "text":
+				responseMods = append(responseMods, "TEXT")
+			case "image":
+				responseMods = append(responseMods, "IMAGE")
+			}
+		}
+		if len(responseMods) > 0 {
+			out, _ = sjson.SetBytes(out, "request.generationConfig.responseModalities", responseMods)
+		}
+	}
+
+	// OpenRouter-style image_config support
+	// If the input uses top-level image_config.aspect_ratio, map it into request.generationConfig.imageConfig.aspectRatio.
+	if imgCfg := gjson.GetBytes(rawJSON, "image_config"); imgCfg.Exists() && imgCfg.IsObject() {
+		if ar := imgCfg.Get("aspect_ratio"); ar.Exists() && ar.Type == gjson.String {
+			out, _ = sjson.SetBytes(out, "request.generationConfig.imageConfig.aspectRatio", ar.Str)
+		}
+		if size := imgCfg.Get("image_size"); size.Exists() && size.Type == gjson.String {
+			out, _ = sjson.SetBytes(out, "request.generationConfig.imageConfig.imageSize", size.Str)
+		}
+	}
+
+	// messages -> systemInstruction + contents
+	messages := gjson.GetBytes(rawJSON, "messages")
+	if messages.IsArray() {
+		arr := messages.Array()
+		// First pass: assistant tool_calls id->name map
+		tcID2Name := map[string]string{}
+		for i := 0; i < len(arr); i++ {
+			m := arr[i]
+			if m.Get("role").String() == "assistant" {
+				tcs := m.Get("tool_calls")
+				if tcs.IsArray() {
+					for _, tc := range tcs.Array() {
+						if tc.Get("type").String() == "function" {
+							id := tc.Get("id").String()
+							name := tc.Get("function.name").String()
+							if id != "" && name != "" {
+								tcID2Name[id] = name
+							}
+						}
+					}
+				}
+			}
+		}
+
+		// Second pass build systemInstruction/tool responses cache
+		toolResponses := map[string]string{} // tool_call_id -> response text
+		for i := 0; i < len(arr); i++ {
+			m := arr[i]
+			role := m.Get("role").String()
+			if role == "tool" {
+				toolCallID := m.Get("tool_call_id").String()
+				if toolCallID != "" {
+					c := m.Get("content")
+					toolResponses[toolCallID] = c.Raw
+				}
+			}
+		}
+
+		for i := 0; i < len(arr); i++ {
+			m := arr[i]
+			role := m.Get("role").String()
+			content := m.Get("content")
+
+			if role == "system" && len(arr) > 1 {
+				// system -> request.systemInstruction as a user message style
+				if content.Type == gjson.String {
+					out, _ = sjson.SetBytes(out, "request.systemInstruction.role", "user")
+					out, _ = sjson.SetBytes(out, "request.systemInstruction.parts.0.text", content.String())
+				} else if content.IsObject() && content.Get("type").String() == "text" {
+					out, _ = sjson.SetBytes(out, "request.systemInstruction.role", "user")
+					out, _ = sjson.SetBytes(out, "request.systemInstruction.parts.0.text", content.Get("text").String())
+				}
+			} else if role == "user" || (role == "system" && len(arr) == 1) {
+				// Build single user content node to avoid splitting into multiple contents
+				node := []byte(`{"role":"user","parts":[]}`)
+				if content.Type == gjson.String {
+					node, _ = sjson.SetBytes(node, "parts.0.text", content.String())
+				} else if content.IsArray() {
+					items := content.Array()
+					p := 0
+					for _, item := range items {
+						switch item.Get("type").String() {
+						case "text":
+							node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".text", item.Get("text").String())
+							p++
+						case "image_url":
+							imageURL := item.Get("image_url.url").String()
+							if len(imageURL) > 5 {
+								pieces := strings.SplitN(imageURL[5:], ";", 2)
+								if len(pieces) == 2 && len(pieces[1]) > 7 {
+									mime := pieces[0]
+									data := pieces[1][7:]
+									node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".inlineData.mime_type", mime)
+									node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".inlineData.data", data)
+									p++
+								}
+							}
+						case "file":
+							filename := item.Get("file.filename").String()
+							fileData := item.Get("file.file_data").String()
+							ext := ""
+							if sp := strings.Split(filename, "."); len(sp) > 1 {
+								ext = sp[len(sp)-1]
+							}
+							if mimeType, ok := misc.MimeTypes[ext]; ok {
+								node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".inlineData.mime_type", mimeType)
+								node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".inlineData.data", fileData)
+								p++
+							} else {
+								log.Warnf("Unknown file name extension '%s' in user message, skip", ext)
+							}
+						}
+					}
+				}
+				out, _ = sjson.SetRawBytes(out, "request.contents.-1", node)
+			} else if role == "assistant" {
+				if content.Type == gjson.String {
+					// Assistant text -> single model content
+					node := []byte(`{"role":"model","parts":[{"text":""}]}`)
+					node, _ = sjson.SetBytes(node, "parts.0.text", content.String())
+					out, _ = sjson.SetRawBytes(out, "request.contents.-1", node)
+				} else if !content.Exists() || content.Type == gjson.Null {
+					// Tool calls -> single model content with functionCall parts
+					tcs := m.Get("tool_calls")
+					if tcs.IsArray() {
+						node := []byte(`{"role":"model","parts":[]}`)
+						p := 0
+						fIDs := make([]string, 0)
+						for _, tc := range tcs.Array() {
+							if tc.Get("type").String() != "function" {
+								continue
+							}
+							fid := tc.Get("id").String()
+							fname := tc.Get("function.name").String()
+							fargs := tc.Get("function.arguments").String()
+							node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".functionCall.id", fid)
+							node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".functionCall.name", fname)
+							node, _ = sjson.SetRawBytes(node, "parts."+itoa(p)+".functionCall.args", []byte(fargs))
+							node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".thoughtSignature", geminiCLIFunctionThoughtSignature)
+							p++
+							if fid != "" {
+								fIDs = append(fIDs, fid)
+							}
+						}
+						out, _ = sjson.SetRawBytes(out, "request.contents.-1", node)
+
+						// Append a single tool content combining name + response per function
+						toolNode := []byte(`{"role":"user","parts":[]}`)
+						pp := 0
+						for _, fid := range fIDs {
+							if name, ok := tcID2Name[fid]; ok {
+								toolNode, _ = sjson.SetBytes(toolNode, "parts."+itoa(pp)+".functionResponse.id", fid)
+								toolNode, _ = sjson.SetBytes(toolNode, "parts."+itoa(pp)+".functionResponse.name", name)
+								resp := toolResponses[fid]
+								if resp == "" {
+									resp = "{}"
+								}
+								// Handle non-JSON output gracefully (matches dev branch approach)
+								if resp != "null" {
+									parsed := gjson.Parse(resp)
+									if parsed.Type == gjson.JSON {
+										toolNode, _ = sjson.SetRawBytes(toolNode, "parts."+itoa(pp)+".functionResponse.response.result", []byte(parsed.Raw))
+									} else {
+										toolNode, _ = sjson.SetBytes(toolNode, "parts."+itoa(pp)+".functionResponse.response.result", resp)
+									}
+								}
+								pp++
+							}
+						}
+						if pp > 0 {
+							out, _ = sjson.SetRawBytes(out, "request.contents.-1", toolNode)
+						}
+					}
+				}
+			}
+		}
+	}
+
+	// tools -> request.tools[0].functionDeclarations + request.tools[0].googleSearch passthrough
+	tools := gjson.GetBytes(rawJSON, "tools")
+	if tools.IsArray() && len(tools.Array()) > 0 {
+		toolNode := []byte(`{}`)
+		hasTool := false
+		hasFunction := false
+		for _, t := range tools.Array() {
+			if t.Get("type").String() == "function" {
+				fn := t.Get("function")
+				if fn.Exists() && fn.IsObject() {
+					fnRaw := fn.Raw
+					if fn.Get("parameters").Exists() {
+						renamed, errRename := util.RenameKey(fnRaw, "parameters", "parametersJsonSchema")
+						if errRename != nil {
+							log.Warnf("Failed to rename parameters for tool '%s': %v", fn.Get("name").String(), errRename)
+							var errSet error
+							fnRaw, errSet = sjson.Set(fnRaw, "parametersJsonSchema.type", "object")
+							if errSet != nil {
+								log.Warnf("Failed to set default schema type for tool '%s': %v", fn.Get("name").String(), errSet)
+								continue
+							}
+							fnRaw, errSet = sjson.Set(fnRaw, "parametersJsonSchema.properties", map[string]interface{}{})
+							if errSet != nil {
+								log.Warnf("Failed to set default schema properties for tool '%s': %v", fn.Get("name").String(), errSet)
+								continue
+							}
+						} else {
+							fnRaw = renamed
+						}
+					} else {
+						var errSet error
+						fnRaw, errSet = sjson.Set(fnRaw, "parametersJsonSchema.type", "object")
+						if errSet != nil {
+							log.Warnf("Failed to set default schema type for tool '%s': %v", fn.Get("name").String(), errSet)
+							continue
+						}
+						fnRaw, errSet = sjson.Set(fnRaw, "parametersJsonSchema.properties", map[string]interface{}{})
+						if errSet != nil {
+							log.Warnf("Failed to set default schema properties for tool '%s': %v", fn.Get("name").String(), errSet)
+							continue
+						}
+					}
+					fnRaw, _ = sjson.Delete(fnRaw, "strict")
+					if !hasFunction {
+						toolNode, _ = sjson.SetRawBytes(toolNode, "functionDeclarations", []byte("[]"))
+					}
+					tmp, errSet := sjson.SetRawBytes(toolNode, "functionDeclarations.-1", []byte(fnRaw))
+					if errSet != nil {
+						log.Warnf("Failed to append tool declaration for '%s': %v", fn.Get("name").String(), errSet)
+						continue
+					}
+					toolNode = tmp
+					hasFunction = true
+					hasTool = true
+				}
+			}
+			if gs := t.Get("google_search"); gs.Exists() {
+				var errSet error
+				toolNode, errSet = sjson.SetRawBytes(toolNode, "googleSearch", []byte(gs.Raw))
+				if errSet != nil {
+					log.Warnf("Failed to set googleSearch tool: %v", errSet)
+					continue
+				}
+				hasTool = true
+			}
+		}
+		if hasTool {
+			out, _ = sjson.SetRawBytes(out, "request.tools", []byte("[]"))
+			out, _ = sjson.SetRawBytes(out, "request.tools.0", toolNode)
+		}
+	}
+
+	return common.AttachDefaultSafetySettings(out, "request.safetySettings")
+}
+
+// itoa converts int to string without strconv import for few usages.
+func itoa(i int) string { return fmt.Sprintf("%d", i) }
+
+// quoteIfNeeded ensures a string is valid JSON value (quotes plain text), pass-through for JSON objects/arrays.
+func quoteIfNeeded(s string) string {
+	s = strings.TrimSpace(s)
+	if s == "" {
+		return "\"\""
+	}
+	if len(s) > 0 && (s[0] == '{' || s[0] == '[') {
+		return s
+	}
+	// escape quotes minimally
+	s = strings.ReplaceAll(s, "\\", "\\\\")
+	s = strings.ReplaceAll(s, "\"", "\\\"")
+	return "\"" + s + "\""
+}

internal/translator/antigravity/openai/chat-completions/antigravity_openai_response.go

@@ -0,0 +1,220 @@
+// Package openai provides response translation functionality for Gemini CLI to OpenAI API compatibility.
+// This package handles the conversion of Gemini CLI API responses into OpenAI Chat Completions-compatible
+// JSON format, transforming streaming events and non-streaming responses into the format
+// expected by OpenAI API clients. It supports both streaming and non-streaming modes,
+// handling text content, tool calls, reasoning content, and usage metadata appropriately.
+package chat_completions
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"strings"
+	"sync/atomic"
+	"time"
+
+	. "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/openai/chat-completions"
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+// convertCliResponseToOpenAIChatParams holds parameters for response conversion.
+type convertCliResponseToOpenAIChatParams struct {
+	UnixTimestamp int64
+	FunctionIndex int
+}
+
+// functionCallIDCounter provides a process-wide unique counter for function call identifiers.
+var functionCallIDCounter uint64
+
+// ConvertAntigravityResponseToOpenAI translates a single chunk of a streaming response from the
+// Gemini CLI API format to the OpenAI Chat Completions streaming format.
+// It processes various Gemini CLI event types and transforms them into OpenAI-compatible JSON responses.
+// The function handles text content, tool calls, reasoning content, and usage metadata, outputting
+// responses that match the OpenAI API format. It supports incremental updates for streaming responses.
+//
+// Parameters:
+//   - ctx: The context for the request, used for cancellation and timeout handling
+//   - modelName: The name of the model being used for the response (unused in current implementation)
+//   - rawJSON: The raw JSON response from the Gemini CLI API
+//   - param: A pointer to a parameter object for maintaining state between calls
+//
+// Returns:
+//   - []string: A slice of strings, each containing an OpenAI-compatible JSON response
+func ConvertAntigravityResponseToOpenAI(_ context.Context, _ string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, param *any) []string {
+	if *param == nil {
+		*param = &convertCliResponseToOpenAIChatParams{
+			UnixTimestamp: 0,
+			FunctionIndex: 0,
+		}
+	}
+
+	if bytes.Equal(rawJSON, []byte("[DONE]")) {
+		return []string{}
+	}
+
+	// Initialize the OpenAI SSE template.
+	template := `{"id":"","object":"chat.completion.chunk","created":12345,"model":"model","choices":[{"index":0,"delta":{"role":null,"content":null,"reasoning_content":null,"tool_calls":null},"finish_reason":null,"native_finish_reason":null}]}`
+
+	// Extract and set the model version.
+	if modelVersionResult := gjson.GetBytes(rawJSON, "response.modelVersion"); modelVersionResult.Exists() {
+		template, _ = sjson.Set(template, "model", modelVersionResult.String())
+	}
+
+	// Extract and set the creation timestamp.
+	if createTimeResult := gjson.GetBytes(rawJSON, "response.createTime"); createTimeResult.Exists() {
+		t, err := time.Parse(time.RFC3339Nano, createTimeResult.String())
+		if err == nil {
+			(*param).(*convertCliResponseToOpenAIChatParams).UnixTimestamp = t.Unix()
+		}
+		template, _ = sjson.Set(template, "created", (*param).(*convertCliResponseToOpenAIChatParams).UnixTimestamp)
+	} else {
+		template, _ = sjson.Set(template, "created", (*param).(*convertCliResponseToOpenAIChatParams).UnixTimestamp)
+	}
+
+	// Extract and set the response ID.
+	if responseIDResult := gjson.GetBytes(rawJSON, "response.responseId"); responseIDResult.Exists() {
+		template, _ = sjson.Set(template, "id", responseIDResult.String())
+	}
+
+	// Extract and set the finish reason.
+	if finishReasonResult := gjson.GetBytes(rawJSON, "response.candidates.0.finishReason"); finishReasonResult.Exists() {
+		template, _ = sjson.Set(template, "choices.0.finish_reason", strings.ToLower(finishReasonResult.String()))
+		template, _ = sjson.Set(template, "choices.0.native_finish_reason", strings.ToLower(finishReasonResult.String()))
+	}
+
+	// Extract and set usage metadata (token counts).
+	if usageResult := gjson.GetBytes(rawJSON, "response.usageMetadata"); usageResult.Exists() {
+		if candidatesTokenCountResult := usageResult.Get("candidatesTokenCount"); candidatesTokenCountResult.Exists() {
+			template, _ = sjson.Set(template, "usage.completion_tokens", candidatesTokenCountResult.Int())
+		}
+		if totalTokenCountResult := usageResult.Get("totalTokenCount"); totalTokenCountResult.Exists() {
+			template, _ = sjson.Set(template, "usage.total_tokens", totalTokenCountResult.Int())
+		}
+		promptTokenCount := usageResult.Get("promptTokenCount").Int()
+		thoughtsTokenCount := usageResult.Get("thoughtsTokenCount").Int()
+		template, _ = sjson.Set(template, "usage.prompt_tokens", promptTokenCount+thoughtsTokenCount)
+		if thoughtsTokenCount > 0 {
+			template, _ = sjson.Set(template, "usage.completion_tokens_details.reasoning_tokens", thoughtsTokenCount)
+		}
+	}
+
+	// Process the main content part of the response.
+	partsResult := gjson.GetBytes(rawJSON, "response.candidates.0.content.parts")
+	hasFunctionCall := false
+	if partsResult.IsArray() {
+		partResults := partsResult.Array()
+		for i := 0; i < len(partResults); i++ {
+			partResult := partResults[i]
+			partTextResult := partResult.Get("text")
+			functionCallResult := partResult.Get("functionCall")
+			thoughtSignatureResult := partResult.Get("thoughtSignature")
+			if !thoughtSignatureResult.Exists() {
+				thoughtSignatureResult = partResult.Get("thought_signature")
+			}
+			inlineDataResult := partResult.Get("inlineData")
+			if !inlineDataResult.Exists() {
+				inlineDataResult = partResult.Get("inline_data")
+			}
+
+			hasThoughtSignature := thoughtSignatureResult.Exists() && thoughtSignatureResult.String() != ""
+			hasContentPayload := partTextResult.Exists() || functionCallResult.Exists() || inlineDataResult.Exists()
+
+			// Ignore encrypted thoughtSignature but keep any actual content in the same part.
+			if hasThoughtSignature && !hasContentPayload {
+				continue
+			}
+
+			if partTextResult.Exists() {
+				textContent := partTextResult.String()
+
+				// Handle text content, distinguishing between regular content and reasoning/thoughts.
+				if partResult.Get("thought").Bool() {
+					template, _ = sjson.Set(template, "choices.0.delta.reasoning_content", textContent)
+				} else {
+					template, _ = sjson.Set(template, "choices.0.delta.content", textContent)
+				}
+				template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
+			} else if functionCallResult.Exists() {
+				// Handle function call content.
+				hasFunctionCall = true
+				toolCallsResult := gjson.Get(template, "choices.0.delta.tool_calls")
+				functionCallIndex := (*param).(*convertCliResponseToOpenAIChatParams).FunctionIndex
+				(*param).(*convertCliResponseToOpenAIChatParams).FunctionIndex++
+				if toolCallsResult.Exists() && toolCallsResult.IsArray() {
+					functionCallIndex = len(toolCallsResult.Array())
+				} else {
+					template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls", `[]`)
+				}
+
+				functionCallTemplate := `{"id": "","index": 0,"type": "function","function": {"name": "","arguments": ""}}`
+				fcName := functionCallResult.Get("name").String()
+				functionCallTemplate, _ = sjson.Set(functionCallTemplate, "id", fmt.Sprintf("%s-%d-%d", fcName, time.Now().UnixNano(), atomic.AddUint64(&functionCallIDCounter, 1)))
+				functionCallTemplate, _ = sjson.Set(functionCallTemplate, "index", functionCallIndex)
+				functionCallTemplate, _ = sjson.Set(functionCallTemplate, "function.name", fcName)
+				if fcArgsResult := functionCallResult.Get("args"); fcArgsResult.Exists() {
+					functionCallTemplate, _ = sjson.Set(functionCallTemplate, "function.arguments", fcArgsResult.Raw)
+				}
+				template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
+				template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls.-1", functionCallTemplate)
+			} else if inlineDataResult.Exists() {
+				data := inlineDataResult.Get("data").String()
+				if data == "" {
+					continue
+				}
+				mimeType := inlineDataResult.Get("mimeType").String()
+				if mimeType == "" {
+					mimeType = inlineDataResult.Get("mime_type").String()
+				}
+				if mimeType == "" {
+					mimeType = "image/png"
+				}
+				imageURL := fmt.Sprintf("data:%s;base64,%s", mimeType, data)
+				imagePayload, err := json.Marshal(map[string]any{
+					"type": "image_url",
+					"image_url": map[string]string{
+						"url": imageURL,
+					},
+				})
+				if err != nil {
+					continue
+				}
+				imagesResult := gjson.Get(template, "choices.0.delta.images")
+				if !imagesResult.Exists() || !imagesResult.IsArray() {
+					template, _ = sjson.SetRaw(template, "choices.0.delta.images", `[]`)
+				}
+				template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
+				template, _ = sjson.SetRaw(template, "choices.0.delta.images.-1", string(imagePayload))
+			}
+		}
+	}
+
+	if hasFunctionCall {
+		template, _ = sjson.Set(template, "choices.0.finish_reason", "tool_calls")
+		template, _ = sjson.Set(template, "choices.0.native_finish_reason", "tool_calls")
+	}
+
+	return []string{template}
+}
+
+// ConvertAntigravityResponseToOpenAINonStream converts a non-streaming Gemini CLI response to a non-streaming OpenAI response.
+// This function processes the complete Gemini CLI response and transforms it into a single OpenAI-compatible
+// JSON response. It handles message content, tool calls, reasoning content, and usage metadata, combining all
+// the information into a single response that matches the OpenAI API format.
+//
+// Parameters:
+//   - ctx: The context for the request, used for cancellation and timeout handling
+//   - modelName: The name of the model being used for the response
+//   - rawJSON: The raw JSON response from the Gemini CLI API
+//   - param: A pointer to a parameter object for the conversion
+//
+// Returns:
+//   - string: An OpenAI-compatible JSON response containing all message content and metadata
+func ConvertAntigravityResponseToOpenAINonStream(ctx context.Context, modelName string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, param *any) string {
+	responseResult := gjson.GetBytes(rawJSON, "response")
+	if responseResult.Exists() {
+		return ConvertGeminiResponseToOpenAINonStream(ctx, modelName, originalRequestRawJSON, requestRawJSON, []byte(responseResult.Raw), param)
+	}
+	return ""
+}

internal/translator/antigravity/openai/chat-completions/init.go

@@ -0,0 +1,19 @@
+package chat_completions
+
+import (
+	. "github.com/router-for-me/CLIProxyAPI/v6/internal/constant"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/interfaces"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/translator/translator"
+)
+
+func init() {
+	translator.Register(
+		OpenAI,
+		Antigravity,
+		ConvertOpenAIRequestToAntigravity,
+		interfaces.TranslateResponse{
+			Stream:    ConvertAntigravityResponseToOpenAI,
+			NonStream: ConvertAntigravityResponseToOpenAINonStream,
+		},
+	)
+}

internal/translator/antigravity/openai/responses/antigravity_openai-responses_request.go

@@ -0,0 +1,14 @@
+package responses
+
+import (
+	"bytes"
+
+	. "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/antigravity/gemini"
+	. "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/openai/responses"
+)
+
+func ConvertOpenAIResponsesRequestToAntigravity(modelName string, inputRawJSON []byte, stream bool) []byte {
+	rawJSON := bytes.Clone(inputRawJSON)
+	rawJSON = ConvertOpenAIResponsesRequestToGemini(modelName, rawJSON, stream)
+	return ConvertGeminiRequestToAntigravity(modelName, rawJSON, stream)
+}

internal/translator/antigravity/openai/responses/antigravity_openai-responses_response.go

@@ -0,0 +1,35 @@
+package responses
+
+import (
+	"context"
+
+	. "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/openai/responses"
+	"github.com/tidwall/gjson"
+)
+
+func ConvertAntigravityResponseToOpenAIResponses(ctx context.Context, modelName string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, param *any) []string {
+	responseResult := gjson.GetBytes(rawJSON, "response")
+	if responseResult.Exists() {
+		rawJSON = []byte(responseResult.Raw)
+	}
+	return ConvertGeminiResponseToOpenAIResponses(ctx, modelName, originalRequestRawJSON, requestRawJSON, rawJSON, param)
+}
+
+func ConvertAntigravityResponseToOpenAIResponsesNonStream(ctx context.Context, modelName string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, param *any) string {
+	responseResult := gjson.GetBytes(rawJSON, "response")
+	if responseResult.Exists() {
+		rawJSON = []byte(responseResult.Raw)
+	}
+
+	requestResult := gjson.GetBytes(originalRequestRawJSON, "request")
+	if responseResult.Exists() {
+		originalRequestRawJSON = []byte(requestResult.Raw)
+	}
+
+	requestResult = gjson.GetBytes(requestRawJSON, "request")
+	if responseResult.Exists() {
+		requestRawJSON = []byte(requestResult.Raw)
+	}
+
+	return ConvertGeminiResponseToOpenAIResponsesNonStream(ctx, modelName, originalRequestRawJSON, requestRawJSON, rawJSON, param)
+}

internal/translator/antigravity/openai/responses/init.go

@@ -0,0 +1,19 @@
+package responses
+
+import (
+	. "github.com/router-for-me/CLIProxyAPI/v6/internal/constant"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/interfaces"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/translator/translator"
+)
+
+func init() {
+	translator.Register(
+		OpenaiResponse,
+		Antigravity,
+		ConvertOpenAIResponsesRequestToAntigravity,
+		interfaces.TranslateResponse{
+			Stream:    ConvertAntigravityResponseToOpenAIResponses,
+			NonStream: ConvertAntigravityResponseToOpenAIResponsesNonStream,
+		},
+	)
+}

internal/translator/claude/gemini/claude_gemini_response.go

@@ -331,8 +331,8 @@ func ConvertClaudeResponseToGeminiNonStream(_ context.Context, modelName string,
 	streamingEvents := make([][]byte, 0)
 
 	scanner := bufio.NewScanner(bytes.NewReader(rawJSON))
-	buffer := make([]byte, 20_971_520)
-	scanner.Buffer(buffer, 20_971_520)
+	buffer := make([]byte, 52_428_800) // 50MB
+	scanner.Buffer(buffer, 52_428_800)
 	for scanner.Scan() {
 		line := scanner.Bytes()
 		// log.Debug(string(line))

internal/translator/claude/openai/responses/claude_openai-responses_response.go

@@ -445,8 +445,8 @@ func ConvertClaudeResponseToOpenAIResponsesNonStream(_ context.Context, _ string
 		// Use a simple scanner to iterate through raw bytes
 		// Note: extremely large responses may require increasing the buffer
 		scanner := bufio.NewScanner(bytes.NewReader(rawJSON))
-		buf := make([]byte, 20_971_520)
-		scanner.Buffer(buf, 20_971_520)
+		buf := make([]byte, 52_428_800) // 50MB
+		scanner.Buffer(buf, 52_428_800)
 		for scanner.Scan() {
 			line := scanner.Bytes()
 			if !bytes.HasPrefix(line, dataTag) {

internal/translator/codex/claude/codex_claude_request.go

@@ -204,7 +204,7 @@ func ConvertClaudeRequestToCodex(modelName string, inputRawJSON []byte, _ bool)
 				}
 				tool, _ = sjson.Set(tool, "name", name)
 			}
-			tool, _ = sjson.SetRaw(tool, "parameters", toolResult.Get("input_schema").Raw)
+			tool, _ = sjson.SetRaw(tool, "parameters", normalizeToolParameters(toolResult.Get("input_schema").Raw))
 			tool, _ = sjson.Delete(tool, "input_schema")
 			tool, _ = sjson.Delete(tool, "parameters.$schema")
 			tool, _ = sjson.Set(tool, "strict", false)
@@ -289,7 +289,7 @@ func buildShortNameMap(names []string) map[string]string {
 		}
 		base := cand
 		for i := 1; ; i++ {
-			suffix := "~" + strconv.Itoa(i)
+			suffix := "_" + strconv.Itoa(i)
 			allowed := limit - len(suffix)
 			if allowed < 0 {
 				allowed = 0
@@ -334,3 +334,22 @@ func buildReverseMapFromClaudeOriginalToShort(original []byte) map[string]string
 	}
 	return m
 }
+
+// normalizeToolParameters ensures object schemas contain at least an empty properties map.
+func normalizeToolParameters(raw string) string {
+	raw = strings.TrimSpace(raw)
+	if raw == "" || raw == "null" || !gjson.Valid(raw) {
+		return `{"type":"object","properties":{}}`
+	}
+	schema := raw
+	result := gjson.Parse(raw)
+	schemaType := result.Get("type").String()
+	if schemaType == "" {
+		schema, _ = sjson.Set(schema, "type", "object")
+		schemaType = "object"
+	}
+	if schemaType == "object" && !result.Get("properties").Exists() {
+		schema, _ = sjson.SetRaw(schema, "properties", `{}`)
+	}
+	return schema
+}

internal/translator/codex/gemini/codex_gemini_request.go

@@ -310,7 +310,7 @@ func buildShortNameMap(names []string) map[string]string {
 		}
 		base := cand
 		for i := 1; ; i++ {
-			suffix := "~" + strconv.Itoa(i)
+			suffix := "_" + strconv.Itoa(i)
 			allowed := limit - len(suffix)
 			if allowed < 0 {
 				allowed = 0

internal/translator/codex/gemini/codex_gemini_response.go

@@ -327,7 +327,7 @@ func buildReverseMapFromGeminiOriginal(original []byte) map[string]string {
 func mustMarshalJSON(v interface{}) string {
 	data, err := json.Marshal(v)
 	if err != nil {
-		panic(err)
+		return ""
 	}
 	return string(data)
 }

internal/translator/codex/openai/chat-completions/codex_openai_request.go

@@ -361,7 +361,7 @@ func buildShortNameMap(names []string) map[string]string {
 		}
 		base := cand
 		for i := 1; ; i++ {
-			suffix := "~" + strconv.Itoa(i)
+			suffix := "_" + strconv.Itoa(i)
 			allowed := limit - len(suffix)
 			if allowed < 0 {
 				allowed = 0

internal/translator/codex/openai/responses/codex_openai-responses_request.go

@@ -22,6 +22,7 @@ func ConvertOpenAIResponsesRequestToCodex(modelName string, inputRawJSON []byte,
 	rawJSON, _ = sjson.DeleteBytes(rawJSON, "max_completion_tokens")
 	rawJSON, _ = sjson.DeleteBytes(rawJSON, "temperature")
 	rawJSON, _ = sjson.DeleteBytes(rawJSON, "top_p")
+	rawJSON, _ = sjson.DeleteBytes(rawJSON, "service_tier")
 
 	originalInstructions := ""
 	originalInstructionsText := ""

internal/translator/gemini-cli/claude/gemini-cli_claude_request.go

@@ -17,6 +17,8 @@ import (
 	"github.com/tidwall/sjson"
 )
 
+const geminiCLIClaudeThoughtSignature = "skip_thought_signature_validator"
+
 // ConvertClaudeRequestToCLI parses and transforms a Claude Code API request into Gemini CLI API format.
 // It extracts the model name, system instruction, message contents, and tool declarations
 // from the raw JSON request and returns them in the format expected by the Gemini CLI API.
@@ -89,7 +91,10 @@ func ConvertClaudeRequestToCLI(modelName string, inputRawJSON []byte, _ bool) []
 						functionArgs := contentResult.Get("input").String()
 						var args map[string]any
 						if err := json.Unmarshal([]byte(functionArgs), &args); err == nil {
-							clientContent.Parts = append(clientContent.Parts, client.Part{FunctionCall: &client.FunctionCall{Name: functionName, Args: args}})
+							clientContent.Parts = append(clientContent.Parts, client.Part{
+								FunctionCall:     &client.FunctionCall{Name: functionName, Args: args},
+								ThoughtSignature: geminiCLIClaudeThoughtSignature,
+							})
 						}
 					} else if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "tool_result" {
 						toolCallID := contentResult.Get("tool_use_id").String()
@@ -128,6 +133,7 @@ func ConvertClaudeRequestToCLI(modelName string, inputRawJSON []byte, _ bool) []
 				tool, _ := sjson.Delete(toolResult.Raw, "input_schema")
 				tool, _ = sjson.SetRaw(tool, "parametersJsonSchema", inputSchema)
 				tool, _ = sjson.Delete(tool, "strict")
+				tool, _ = sjson.Delete(tool, "input_examples")
 				var toolDeclaration any
 				if err := json.Unmarshal([]byte(tool), &toolDeclaration); err == nil {
 					tools[0].FunctionDeclarations = append(tools[0].FunctionDeclarations, toolDeclaration)
@@ -159,7 +165,6 @@ func ConvertClaudeRequestToCLI(modelName string, inputRawJSON []byte, _ bool) []
 		if t.Get("type").String() == "enabled" {
 			if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
 				budget := int(b.Int())
-				budget = util.NormalizeThinkingBudget(modelName, budget)
 				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
 				out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 			}

internal/translator/gemini-cli/claude/gemini-cli_claude_response.go

@@ -12,6 +12,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"strings"
+	"sync/atomic"
 	"time"
 
 	"github.com/tidwall/gjson"
@@ -25,8 +26,12 @@ type Params struct {
 	HasFirstResponse bool // Indicates if the initial message_start event has been sent
 	ResponseType     int  // Current response type: 0=none, 1=content, 2=thinking, 3=function
 	ResponseIndex    int  // Index counter for content blocks in the streaming response
+	HasContent       bool // Tracks whether any content (text, thinking, or tool use) has been output
 }
 
+// toolUseIDCounter provides a process-wide unique counter for tool use identifiers.
+var toolUseIDCounter uint64
+
 // ConvertGeminiCLIResponseToClaude performs sophisticated streaming response format conversion.
 // This function implements a complex state machine that translates backend client responses
 // into Claude Code-compatible Server-Sent Events (SSE) format. It manages different response types
@@ -53,9 +58,13 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 	}
 
 	if bytes.Equal(rawJSON, []byte("[DONE]")) {
-		return []string{
-			"event: message_stop\ndata: {\"type\":\"message_stop\"}\n\n\n",
+		// Only send message_stop if we have actually output content
+		if (*param).(*Params).HasContent {
+			return []string{
+				"event: message_stop\ndata: {\"type\":\"message_stop\"}\n\n\n",
+			}
 		}
+		return []string{}
 	}
 
 	// Track whether tools are being used in this response chunk
@@ -104,6 +113,7 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 						output = output + "event: content_block_delta\n"
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"thinking_delta","thinking":""}}`, (*param).(*Params).ResponseIndex), "delta.thinking", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
+						(*param).(*Params).HasContent = true
 					} else {
 						// Transition from another state to thinking
 						// First, close any existing content block
@@ -127,6 +137,7 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"thinking_delta","thinking":""}}`, (*param).(*Params).ResponseIndex), "delta.thinking", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
 						(*param).(*Params).ResponseType = 2 // Set state to thinking
+						(*param).(*Params).HasContent = true
 					}
 				} else {
 					// Process regular text content (user-visible output)
@@ -135,6 +146,7 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 						output = output + "event: content_block_delta\n"
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"text_delta","text":""}}`, (*param).(*Params).ResponseIndex), "delta.text", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
+						(*param).(*Params).HasContent = true
 					} else {
 						// Transition from another state to text content
 						// First, close any existing content block
@@ -158,6 +170,7 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"text_delta","text":""}}`, (*param).(*Params).ResponseIndex), "delta.text", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
 						(*param).(*Params).ResponseType = 1 // Set state to content
+						(*param).(*Params).HasContent = true
 					}
 				}
 			} else if functionCallResult.Exists() {
@@ -197,7 +210,7 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 
 				// Create the tool use block with unique ID and function details
 				data := fmt.Sprintf(`{"type":"content_block_start","index":%d,"content_block":{"type":"tool_use","id":"","name":"","input":{}}}`, (*param).(*Params).ResponseIndex)
-				data, _ = sjson.Set(data, "content_block.id", fmt.Sprintf("%s-%d", fcName, time.Now().UnixNano()))
+				data, _ = sjson.Set(data, "content_block.id", fmt.Sprintf("%s-%d-%d", fcName, time.Now().UnixNano(), atomic.AddUint64(&toolUseIDCounter, 1)))
 				data, _ = sjson.Set(data, "content_block.name", fcName)
 				output = output + fmt.Sprintf("data: %s\n\n\n", data)
 
@@ -207,6 +220,7 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 					output = output + fmt.Sprintf("data: %s\n\n\n", data)
 				}
 				(*param).(*Params).ResponseType = 3
+				(*param).(*Params).HasContent = true
 			}
 		}
 	}
@@ -215,28 +229,31 @@ func ConvertGeminiCLIResponseToClaude(_ context.Context, _ string, originalReque
 	// Process usage metadata and finish reason when present in the response
 	if usageResult.Exists() && bytes.Contains(rawJSON, []byte(`"finishReason"`)) {
 		if candidatesTokenCountResult := usageResult.Get("candidatesTokenCount"); candidatesTokenCountResult.Exists() {
-			// Close the final content block
-			output = output + "event: content_block_stop\n"
-			output = output + fmt.Sprintf(`data: {"type":"content_block_stop","index":%d}`, (*param).(*Params).ResponseIndex)
-			output = output + "\n\n\n"
+			// Only send final events if we have actually output content
+			if (*param).(*Params).HasContent {
+				// Close the final content block
+				output = output + "event: content_block_stop\n"
+				output = output + fmt.Sprintf(`data: {"type":"content_block_stop","index":%d}`, (*param).(*Params).ResponseIndex)
+				output = output + "\n\n\n"
 
-			// Send the final message delta with usage information and stop reason
-			output = output + "event: message_delta\n"
-			output = output + `data: `
+				// Send the final message delta with usage information and stop reason
+				output = output + "event: message_delta\n"
+				output = output + `data: `
 
-			// Create the message delta template with appropriate stop reason
-			template := `{"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
-			// Set tool_use stop reason if tools were used in this response
-			if usedTool {
-				template = `{"type":"message_delta","delta":{"stop_reason":"tool_use","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
+				// Create the message delta template with appropriate stop reason
+				template := `{"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
+				// Set tool_use stop reason if tools were used in this response
+				if usedTool {
+					template = `{"type":"message_delta","delta":{"stop_reason":"tool_use","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
+				}
+
+				// Include thinking tokens in output token count if present
+				thoughtsTokenCount := usageResult.Get("thoughtsTokenCount").Int()
+				template, _ = sjson.Set(template, "usage.output_tokens", candidatesTokenCountResult.Int()+thoughtsTokenCount)
+				template, _ = sjson.Set(template, "usage.input_tokens", usageResult.Get("promptTokenCount").Int())
+
+				output = output + template + "\n\n\n"
 			}
-
-			// Include thinking tokens in output token count if present
-			thoughtsTokenCount := usageResult.Get("thoughtsTokenCount").Int()
-			template, _ = sjson.Set(template, "usage.output_tokens", candidatesTokenCountResult.Int()+thoughtsTokenCount)
-			template, _ = sjson.Set(template, "usage.input_tokens", usageResult.Get("promptTokenCount").Int())
-
-			output = output + template + "\n\n\n"
 		}
 	}
 

internal/translator/gemini-cli/gemini/gemini-cli_gemini_request.go

@@ -98,6 +98,20 @@ func ConvertGeminiRequestToGeminiCLI(_ string, inputRawJSON []byte, _ bool) []by
 		}
 	}
 
+	gjson.GetBytes(rawJSON, "request.contents").ForEach(func(key, content gjson.Result) bool {
+		if content.Get("role").String() == "model" {
+			content.Get("parts").ForEach(func(partKey, part gjson.Result) bool {
+				if part.Get("functionCall").Exists() {
+					rawJSON, _ = sjson.SetBytes(rawJSON, fmt.Sprintf("request.contents.%d.parts.%d.thoughtSignature", key.Int(), partKey.Int()), "skip_thought_signature_validator")
+				} else if part.Get("thoughtSignature").Exists() {
+					rawJSON, _ = sjson.SetBytes(rawJSON, fmt.Sprintf("request.contents.%d.parts.%d.thoughtSignature", key.Int(), partKey.Int()), "skip_thought_signature_validator")
+				}
+				return true
+			})
+		}
+		return true
+	})
+
 	return common.AttachDefaultSafetySettings(rawJSON, "request.safetySettings")
 }
 

internal/translator/gemini-cli/gemini/gemini-cli_gemini_response.go

@@ -6,6 +6,7 @@
 package gemini
 
 import (
+	"bytes"
 	"context"
 	"fmt"
 
@@ -13,7 +14,7 @@ import (
 	"github.com/tidwall/sjson"
 )
 
-// ConvertGeminiCliRequestToGemini parses and transforms a Gemini CLI API request into Gemini API format.
+// ConvertGeminiCliResponseToGemini parses and transforms a Gemini CLI API request into Gemini API format.
 // It extracts the model name, system instruction, message contents, and tool declarations
 // from the raw JSON request and returns them in the format expected by the Gemini API.
 // The function performs the following transformations:
@@ -29,7 +30,11 @@ import (
 //
 // Returns:
 //   - []string: The transformed request data in Gemini API format
-func ConvertGeminiCliRequestToGemini(ctx context.Context, _ string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, _ *any) []string {
+func ConvertGeminiCliResponseToGemini(ctx context.Context, _ string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, _ *any) []string {
+	if bytes.HasPrefix(rawJSON, []byte("data:")) {
+		rawJSON = bytes.TrimSpace(rawJSON[5:])
+	}
+
 	if alt, ok := ctx.Value("alt").(string); ok {
 		var chunk []byte
 		if alt == "" {
@@ -56,7 +61,7 @@ func ConvertGeminiCliRequestToGemini(ctx context.Context, _ string, originalRequ
 	return []string{}
 }
 
-// ConvertGeminiCliRequestToGeminiNonStream converts a non-streaming Gemini CLI request to a non-streaming Gemini response.
+// ConvertGeminiCliResponseToGeminiNonStream converts a non-streaming Gemini CLI request to a non-streaming Gemini response.
 // This function processes the complete Gemini CLI request and transforms it into a single Gemini-compatible
 // JSON response. It extracts the response data from the request and returns it in the expected format.
 //
@@ -68,7 +73,7 @@ func ConvertGeminiCliRequestToGemini(ctx context.Context, _ string, originalRequ
 //
 // Returns:
 //   - string: A Gemini-compatible JSON response containing the response data
-func ConvertGeminiCliRequestToGeminiNonStream(_ context.Context, _ string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, _ *any) string {
+func ConvertGeminiCliResponseToGeminiNonStream(_ context.Context, _ string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, _ *any) string {
 	responseResult := gjson.GetBytes(rawJSON, "response")
 	if responseResult.Exists() {
 		return responseResult.Raw

internal/translator/gemini-cli/gemini/init.go

@@ -12,8 +12,8 @@ func init() {
 		GeminiCLI,
 		ConvertGeminiRequestToGeminiCLI,
 		interfaces.TranslateResponse{
-			Stream:     ConvertGeminiCliRequestToGemini,
-			NonStream:  ConvertGeminiCliRequestToGeminiNonStream,
+			Stream:     ConvertGeminiCliResponseToGemini,
+			NonStream:  ConvertGeminiCliResponseToGeminiNonStream,
 			TokenCount: GeminiTokenCount,
 		},
 	)

internal/translator/gemini-cli/openai/chat-completions/gemini-cli_openai_request.go

@@ -15,6 +15,8 @@ import (
 	"github.com/tidwall/sjson"
 )
 
+const geminiCLIFunctionThoughtSignature = "skip_thought_signature_validator"
+
 // ConvertOpenAIRequestToGeminiCLI converts an OpenAI Chat Completions request (raw JSON)
 // into a complete Gemini CLI request JSON. All JSON construction uses sjson and lookups use gjson.
 //
@@ -46,13 +48,13 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", -1)
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 		case "low":
-			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 1024))
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", 1024)
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 		case "medium":
-			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 8192))
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", 8192)
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 		case "high":
-			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", util.NormalizeThinkingBudget(modelName, 32768))
+			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", 32768)
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 		default:
 			out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", -1)
@@ -64,15 +66,15 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 	if !hasOfficialThinking && util.ModelSupportsThinking(modelName) {
 		if tc := gjson.GetBytes(rawJSON, "extra_body.google.thinking_config"); tc.Exists() && tc.IsObject() {
 			var setBudget bool
-			var normalized int
+			var budget int
 
 			if v := tc.Get("thinkingBudget"); v.Exists() {
-				normalized = util.NormalizeThinkingBudget(modelName, int(v.Int()))
-				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", normalized)
+				budget = int(v.Int())
+				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
 				setBudget = true
 			} else if v := tc.Get("thinking_budget"); v.Exists() {
-				normalized = util.NormalizeThinkingBudget(modelName, int(v.Int()))
-				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", normalized)
+				budget = int(v.Int())
+				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.thinkingBudget", budget)
 				setBudget = true
 			}
 
@@ -80,7 +82,7 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", v.Bool())
 			} else if v := tc.Get("include_thoughts"); v.Exists() {
 				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", v.Bool())
-			} else if setBudget && normalized != 0 {
+			} else if setBudget && budget != 0 {
 				out, _ = sjson.SetBytes(out, "request.generationConfig.thinkingConfig.include_thoughts", true)
 			}
 		}
@@ -120,6 +122,9 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 		if ar := imgCfg.Get("aspect_ratio"); ar.Exists() && ar.Type == gjson.String {
 			out, _ = sjson.SetBytes(out, "request.generationConfig.imageConfig.aspectRatio", ar.Str)
 		}
+		if size := imgCfg.Get("image_size"); size.Exists() && size.Type == gjson.String {
+			out, _ = sjson.SetBytes(out, "request.generationConfig.imageConfig.imageSize", size.Str)
+		}
 	}
 
 	// messages -> systemInstruction + contents
@@ -239,6 +244,7 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 							fargs := tc.Get("function.arguments").String()
 							node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".functionCall.name", fname)
 							node, _ = sjson.SetRawBytes(node, "parts."+itoa(p)+".functionCall.args", []byte(fargs))
+							node, _ = sjson.SetBytes(node, "parts."+itoa(p)+".thoughtSignature", geminiCLIFunctionThoughtSignature)
 							p++
 							if fid != "" {
 								fIDs = append(fIDs, fid)
@@ -269,11 +275,12 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 		}
 	}
 
-	// tools -> request.tools[0].functionDeclarations
+	// tools -> request.tools[0].functionDeclarations + request.tools[0].googleSearch passthrough
 	tools := gjson.GetBytes(rawJSON, "tools")
 	if tools.IsArray() && len(tools.Array()) > 0 {
-		out, _ = sjson.SetRawBytes(out, "request.tools", []byte(`[{"functionDeclarations":[]}]`))
-		fdPath := "request.tools.0.functionDeclarations"
+		toolNode := []byte(`{}`)
+		hasTool := false
+		hasFunction := false
 		for _, t := range tools.Array() {
 			if t.Get("type").String() == "function" {
 				fn := t.Get("function")
@@ -283,6 +290,17 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 						renamed, errRename := util.RenameKey(fnRaw, "parameters", "parametersJsonSchema")
 						if errRename != nil {
 							log.Warnf("Failed to rename parameters for tool '%s': %v", fn.Get("name").String(), errRename)
+							var errSet error
+							fnRaw, errSet = sjson.Set(fnRaw, "parametersJsonSchema.type", "object")
+							if errSet != nil {
+								log.Warnf("Failed to set default schema type for tool '%s': %v", fn.Get("name").String(), errSet)
+								continue
+							}
+							fnRaw, errSet = sjson.Set(fnRaw, "parametersJsonSchema.properties", map[string]interface{}{})
+							if errSet != nil {
+								log.Warnf("Failed to set default schema properties for tool '%s': %v", fn.Get("name").String(), errSet)
+								continue
+							}
 						} else {
 							fnRaw = renamed
 						}
@@ -300,14 +318,32 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 						}
 					}
 					fnRaw, _ = sjson.Delete(fnRaw, "strict")
-					tmp, errSet := sjson.SetRawBytes(out, fdPath+".-1", []byte(fnRaw))
+					if !hasFunction {
+						toolNode, _ = sjson.SetRawBytes(toolNode, "functionDeclarations", []byte("[]"))
+					}
+					tmp, errSet := sjson.SetRawBytes(toolNode, "functionDeclarations.-1", []byte(fnRaw))
 					if errSet != nil {
 						log.Warnf("Failed to append tool declaration for '%s': %v", fn.Get("name").String(), errSet)
 						continue
 					}
-					out = tmp
+					toolNode = tmp
+					hasFunction = true
+					hasTool = true
 				}
 			}
+			if gs := t.Get("google_search"); gs.Exists() {
+				var errSet error
+				toolNode, errSet = sjson.SetRawBytes(toolNode, "googleSearch", []byte(gs.Raw))
+				if errSet != nil {
+					log.Warnf("Failed to set googleSearch tool: %v", errSet)
+					continue
+				}
+				hasTool = true
+			}
+		}
+		if hasTool {
+			out, _ = sjson.SetRawBytes(out, "request.tools", []byte("[]"))
+			out, _ = sjson.SetRawBytes(out, "request.tools.0", toolNode)
 		}
 	}
 

internal/translator/gemini-cli/openai/chat-completions/gemini-cli_openai_response.go

@@ -10,6 +10,8 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
+	"strings"
+	"sync/atomic"
 	"time"
 
 	. "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/openai/chat-completions"
@@ -23,6 +25,9 @@ type convertCliResponseToOpenAIChatParams struct {
 	FunctionIndex int
 }
 
+// functionCallIDCounter provides a process-wide unique counter for function call identifiers.
+var functionCallIDCounter uint64
+
 // ConvertCliResponseToOpenAI translates a single chunk of a streaming response from the
 // Gemini CLI API format to the OpenAI Chat Completions streaming format.
 // It processes various Gemini CLI event types and transforms them into OpenAI-compatible JSON responses.
@@ -75,8 +80,8 @@ func ConvertCliResponseToOpenAI(_ context.Context, _ string, originalRequestRawJ
 
 	// Extract and set the finish reason.
 	if finishReasonResult := gjson.GetBytes(rawJSON, "response.candidates.0.finishReason"); finishReasonResult.Exists() {
-		template, _ = sjson.Set(template, "choices.0.finish_reason", finishReasonResult.String())
-		template, _ = sjson.Set(template, "choices.0.native_finish_reason", finishReasonResult.String())
+		template, _ = sjson.Set(template, "choices.0.finish_reason", strings.ToLower(finishReasonResult.String()))
+		template, _ = sjson.Set(template, "choices.0.native_finish_reason", strings.ToLower(finishReasonResult.String()))
 	}
 
 	// Extract and set usage metadata (token counts).
@@ -104,17 +109,31 @@ func ConvertCliResponseToOpenAI(_ context.Context, _ string, originalRequestRawJ
 			partResult := partResults[i]
 			partTextResult := partResult.Get("text")
 			functionCallResult := partResult.Get("functionCall")
+			thoughtSignatureResult := partResult.Get("thoughtSignature")
+			if !thoughtSignatureResult.Exists() {
+				thoughtSignatureResult = partResult.Get("thought_signature")
+			}
 			inlineDataResult := partResult.Get("inlineData")
 			if !inlineDataResult.Exists() {
 				inlineDataResult = partResult.Get("inline_data")
 			}
 
+			hasThoughtSignature := thoughtSignatureResult.Exists() && thoughtSignatureResult.String() != ""
+			hasContentPayload := partTextResult.Exists() || functionCallResult.Exists() || inlineDataResult.Exists()
+
+			// Ignore encrypted thoughtSignature but keep any actual content in the same part.
+			if hasThoughtSignature && !hasContentPayload {
+				continue
+			}
+
 			if partTextResult.Exists() {
+				textContent := partTextResult.String()
+
 				// Handle text content, distinguishing between regular content and reasoning/thoughts.
 				if partResult.Get("thought").Bool() {
-					template, _ = sjson.Set(template, "choices.0.delta.reasoning_content", partTextResult.String())
+					template, _ = sjson.Set(template, "choices.0.delta.reasoning_content", textContent)
 				} else {
-					template, _ = sjson.Set(template, "choices.0.delta.content", partTextResult.String())
+					template, _ = sjson.Set(template, "choices.0.delta.content", textContent)
 				}
 				template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
 			} else if functionCallResult.Exists() {
@@ -131,7 +150,7 @@ func ConvertCliResponseToOpenAI(_ context.Context, _ string, originalRequestRawJ
 
 				functionCallTemplate := `{"id": "","index": 0,"type": "function","function": {"name": "","arguments": ""}}`
 				fcName := functionCallResult.Get("name").String()
-				functionCallTemplate, _ = sjson.Set(functionCallTemplate, "id", fmt.Sprintf("%s-%d", fcName, time.Now().UnixNano()))
+				functionCallTemplate, _ = sjson.Set(functionCallTemplate, "id", fmt.Sprintf("%s-%d-%d", fcName, time.Now().UnixNano(), atomic.AddUint64(&functionCallIDCounter, 1)))
 				functionCallTemplate, _ = sjson.Set(functionCallTemplate, "index", functionCallIndex)
 				functionCallTemplate, _ = sjson.Set(functionCallTemplate, "function.name", fcName)
 				if fcArgsResult := functionCallResult.Get("args"); fcArgsResult.Exists() {

internal/translator/gemini/claude/gemini_claude_request.go

@@ -17,6 +17,8 @@ import (
 	"github.com/tidwall/sjson"
 )
 
+const geminiClaudeThoughtSignature = "skip_thought_signature_validator"
+
 // ConvertClaudeRequestToGemini parses a Claude API request and returns a complete
 // Gemini CLI request body (as JSON bytes) ready to be sent via SendRawMessageStream.
 // All JSON transformations are performed using gjson/sjson.
@@ -82,7 +84,10 @@ func ConvertClaudeRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 						functionArgs := contentResult.Get("input").String()
 						var args map[string]any
 						if err := json.Unmarshal([]byte(functionArgs), &args); err == nil {
-							clientContent.Parts = append(clientContent.Parts, client.Part{FunctionCall: &client.FunctionCall{Name: functionName, Args: args}})
+							clientContent.Parts = append(clientContent.Parts, client.Part{
+								FunctionCall:     &client.FunctionCall{Name: functionName, Args: args},
+								ThoughtSignature: geminiClaudeThoughtSignature,
+							})
 						}
 					} else if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "tool_result" {
 						toolCallID := contentResult.Get("tool_use_id").String()
@@ -121,6 +126,7 @@ func ConvertClaudeRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 				tool, _ := sjson.Delete(toolResult.Raw, "input_schema")
 				tool, _ = sjson.SetRaw(tool, "parametersJsonSchema", inputSchema)
 				tool, _ = sjson.Delete(tool, "strict")
+				tool, _ = sjson.Delete(tool, "input_examples")
 				var toolDeclaration any
 				if err := json.Unmarshal([]byte(tool), &toolDeclaration); err == nil {
 					tools[0].FunctionDeclarations = append(tools[0].FunctionDeclarations, toolDeclaration)
@@ -152,7 +158,6 @@ func ConvertClaudeRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 		if t.Get("type").String() == "enabled" {
 			if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
 				budget := int(b.Int())
-				budget = util.NormalizeThinkingBudget(modelName, budget)
 				out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", budget)
 				out, _ = sjson.Set(out, "generationConfig.thinkingConfig.include_thoughts", true)
 			}

internal/translator/gemini/claude/gemini_claude_response.go

@@ -12,6 +12,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"strings"
+	"sync/atomic"
 	"time"
 
 	"github.com/tidwall/gjson"
@@ -24,8 +25,12 @@ type Params struct {
 	HasFirstResponse bool
 	ResponseType     int
 	ResponseIndex    int
+	HasContent       bool // Tracks whether any content (text, thinking, or tool use) has been output
 }
 
+// toolUseIDCounter provides a process-wide unique counter for tool use identifiers.
+var toolUseIDCounter uint64
+
 // ConvertGeminiResponseToClaude performs sophisticated streaming response format conversion.
 // This function implements a complex state machine that translates backend client responses
 // into Claude-compatible Server-Sent Events (SSE) format. It manages different response types
@@ -53,9 +58,13 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 	}
 
 	if bytes.Equal(rawJSON, []byte("[DONE]")) {
-		return []string{
-			"event: message_stop\ndata: {\"type\":\"message_stop\"}\n\n\n",
+		// Only send message_stop if we have actually output content
+		if (*param).(*Params).HasContent {
+			return []string{
+				"event: message_stop\ndata: {\"type\":\"message_stop\"}\n\n\n",
+			}
 		}
+		return []string{}
 	}
 
 	// Track whether tools are being used in this response chunk
@@ -104,6 +113,7 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 						output = output + "event: content_block_delta\n"
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"thinking_delta","thinking":""}}`, (*param).(*Params).ResponseIndex), "delta.thinking", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
+						(*param).(*Params).HasContent = true
 					} else {
 						// Transition from another state to thinking
 						// First, close any existing content block
@@ -127,6 +137,7 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"thinking_delta","thinking":""}}`, (*param).(*Params).ResponseIndex), "delta.thinking", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
 						(*param).(*Params).ResponseType = 2 // Set state to thinking
+						(*param).(*Params).HasContent = true
 					}
 				} else {
 					// Process regular text content (user-visible output)
@@ -135,6 +146,7 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 						output = output + "event: content_block_delta\n"
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"text_delta","text":""}}`, (*param).(*Params).ResponseIndex), "delta.text", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
+						(*param).(*Params).HasContent = true
 					} else {
 						// Transition from another state to text content
 						// First, close any existing content block
@@ -158,6 +170,7 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 						data, _ := sjson.Set(fmt.Sprintf(`{"type":"content_block_delta","index":%d,"delta":{"type":"text_delta","text":""}}`, (*param).(*Params).ResponseIndex), "delta.text", partTextResult.String())
 						output = output + fmt.Sprintf("data: %s\n\n\n", data)
 						(*param).(*Params).ResponseType = 1 // Set state to content
+						(*param).(*Params).HasContent = true
 					}
 				}
 			} else if functionCallResult.Exists() {
@@ -197,7 +210,7 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 
 				// Create the tool use block with unique ID and function details
 				data := fmt.Sprintf(`{"type":"content_block_start","index":%d,"content_block":{"type":"tool_use","id":"","name":"","input":{}}}`, (*param).(*Params).ResponseIndex)
-				data, _ = sjson.Set(data, "content_block.id", fmt.Sprintf("%s-%d", fcName, time.Now().UnixNano()))
+				data, _ = sjson.Set(data, "content_block.id", fmt.Sprintf("%s-%d-%d", fcName, time.Now().UnixNano(), atomic.AddUint64(&toolUseIDCounter, 1)))
 				data, _ = sjson.Set(data, "content_block.name", fcName)
 				output = output + fmt.Sprintf("data: %s\n\n\n", data)
 
@@ -207,6 +220,7 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 					output = output + fmt.Sprintf("data: %s\n\n\n", data)
 				}
 				(*param).(*Params).ResponseType = 3
+				(*param).(*Params).HasContent = true
 			}
 		}
 	}
@@ -214,23 +228,26 @@ func ConvertGeminiResponseToClaude(_ context.Context, _ string, originalRequestR
 	usageResult := gjson.GetBytes(rawJSON, "usageMetadata")
 	if usageResult.Exists() && bytes.Contains(rawJSON, []byte(`"finishReason"`)) {
 		if candidatesTokenCountResult := usageResult.Get("candidatesTokenCount"); candidatesTokenCountResult.Exists() {
-			output = output + "event: content_block_stop\n"
-			output = output + fmt.Sprintf(`data: {"type":"content_block_stop","index":%d}`, (*param).(*Params).ResponseIndex)
-			output = output + "\n\n\n"
+			// Only send final events if we have actually output content
+			if (*param).(*Params).HasContent {
+				output = output + "event: content_block_stop\n"
+				output = output + fmt.Sprintf(`data: {"type":"content_block_stop","index":%d}`, (*param).(*Params).ResponseIndex)
+				output = output + "\n\n\n"
 
-			output = output + "event: message_delta\n"
-			output = output + `data: `
+				output = output + "event: message_delta\n"
+				output = output + `data: `
 
-			template := `{"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
-			if usedTool {
-				template = `{"type":"message_delta","delta":{"stop_reason":"tool_use","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
+				template := `{"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
+				if usedTool {
+					template = `{"type":"message_delta","delta":{"stop_reason":"tool_use","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
+				}
+
+				thoughtsTokenCount := usageResult.Get("thoughtsTokenCount").Int()
+				template, _ = sjson.Set(template, "usage.output_tokens", candidatesTokenCountResult.Int()+thoughtsTokenCount)
+				template, _ = sjson.Set(template, "usage.input_tokens", usageResult.Get("promptTokenCount").Int())
+
+				output = output + template + "\n\n\n"
 			}
-
-			thoughtsTokenCount := usageResult.Get("thoughtsTokenCount").Int()
-			template, _ = sjson.Set(template, "usage.output_tokens", candidatesTokenCountResult.Int()+thoughtsTokenCount)
-			template, _ = sjson.Set(template, "usage.input_tokens", usageResult.Get("promptTokenCount").Int())
-
-			output = output + template + "\n\n\n"
 		}
 	}
 

internal/translator/gemini/gemini-cli/gemini_gemini-cli_request.go

@@ -46,5 +46,19 @@ func ConvertGeminiCLIRequestToGemini(_ string, inputRawJSON []byte, _ bool) []by
 		}
 	}
 
+	gjson.GetBytes(rawJSON, "contents").ForEach(func(key, content gjson.Result) bool {
+		if content.Get("role").String() == "model" {
+			content.Get("parts").ForEach(func(partKey, part gjson.Result) bool {
+				if part.Get("functionCall").Exists() {
+					rawJSON, _ = sjson.SetBytes(rawJSON, fmt.Sprintf("contents.%d.parts.%d.thoughtSignature", key.Int(), partKey.Int()), "skip_thought_signature_validator")
+				} else if part.Get("thoughtSignature").Exists() {
+					rawJSON, _ = sjson.SetBytes(rawJSON, fmt.Sprintf("contents.%d.parts.%d.thoughtSignature", key.Int(), partKey.Int()), "skip_thought_signature_validator")
+				}
+				return true
+			})
+		}
+		return true
+	})
+
 	return common.AttachDefaultSafetySettings(rawJSON, "safetySettings")
 }