Merge pull request #1683 from dusty-du/codex/device-login-flow

Add additive Codex device-code login flow

.github/workflows/release.yaml

@@ -19,7 +19,7 @@ jobs:
       - run: git fetch --force --tags
       - uses: actions/setup-go@v4
         with:
-          go-version: '>=1.24.0'
+          go-version: '>=1.26.0'
           cache: true
       - name: Generate Build Metadata
         run: |

Dockerfile

@@ -1,4 +1,4 @@
-FROM golang:1.24-alpine AS builder
+FROM golang:1.26-alpine AS builder
 
 WORKDIR /app
 

README.md

@@ -27,10 +27,6 @@ Get 10% OFF GLM CODING PLAN：https://z.ai/subscribe?ic=8JVLJQFSKB
 <td>Thanks to PackyCode for sponsoring this project! PackyCode is a reliable and efficient API relay service provider, offering relay services for Claude Code, Codex, Gemini, and more. PackyCode provides special discounts for our software users: register using <a href="https://www.packyapi.com/register?aff=cliproxyapi">this link</a> and enter the "cliproxyapi" promo code during recharge to get 10% off.</td>
 </tr>
 <tr>
-<td width="180"><a href="https://cubence.com/signup?code=CLIPROXYAPI&source=cpa"><img src="./assets/cubence.png" alt="Cubence" width="150"></a></td>
-<td>Thanks to Cubence for sponsoring this project! Cubence is a reliable and efficient API relay service provider, offering relay services for Claude Code, Codex, Gemini, and more. Cubence provides special discounts for our software users: register using <a href="https://cubence.com/signup?code=CLIPROXYAPI&source=cpa">this link</a> and enter the "CLIPROXYAPI" promo code during recharge to get 10% off.</td>
-</tr>
-<tr>
 <td width="180"><a href="https://www.aicodemirror.com/register?invitecode=TJNAIF"><img src="./assets/aicodemirror.png" alt="AICodeMirror" width="150"></a></td>
 <td>Thanks to AICodeMirror for sponsoring this project! AICodeMirror provides official high-stability relay services for Claude Code / Codex / Gemini CLI, with enterprise-grade concurrency, fast invoicing, and 24/7 dedicated technical support. Claude Code / Codex / Gemini official channels at 38% / 2% / 9% of original price, with extra discounts on top-ups! AICodeMirror offers special benefits for CLIProxyAPI users: register via <a href="https://www.aicodemirror.com/register?invitecode=TJNAIF">this link</a> to enjoy 20% off your first top-up, and enterprise customers can get up to 25% off!</td>
 </tr>
@@ -150,6 +146,10 @@ A Windows tray application implemented using PowerShell scripts, without relying
 
 霖君 is a cross-platform desktop application for managing AI programming assistants, supporting macOS, Windows, and Linux systems. Unified management of Claude Code, Gemini CLI, OpenAI Codex, Qwen Code, and other AI coding tools, with local proxy for multi-account quota tracking and one-click configuration.
 
+### [CLIProxyAPI Dashboard](https://github.com/itsmylife44/cliproxyapi-dashboard)
+
+A modern web-based management dashboard for CLIProxyAPI built with Next.js, React, and PostgreSQL. Features real-time log streaming, structured configuration editing, API key management, OAuth provider integration for Claude/Gemini/Codex, usage analytics, container management, and config sync with OpenCode via companion plugin - no manual YAML editing needed.
+
 > [!NOTE]  
 > If you developed a project based on CLIProxyAPI, please open a PR to add it to this list.
 
@@ -161,6 +161,12 @@ Those projects are ports of CLIProxyAPI or inspired by it:
 
 A Next.js implementation inspired by CLIProxyAPI, easy to install and use, built from scratch with format translation (OpenAI/Claude/Gemini/Ollama), combo system with auto-fallback, multi-account management with exponential backoff, a Next.js web dashboard, and support for CLI tools (Cursor, Claude Code, Cline, RooCode) - no API keys needed.
 
+### [OmniRoute](https://github.com/diegosouzapw/OmniRoute)
+
+Never stop coding. Smart routing to FREE & low-cost AI models with automatic fallback.
+
+OmniRoute is an AI gateway for multi-provider LLMs: an OpenAI-compatible endpoint with smart routing, load balancing, retries, and fallbacks. Add policies, rate limits, caching, and observability for reliable, cost-aware inference.
+
 > [!NOTE]  
 > If you have developed a port of CLIProxyAPI or a project inspired by it, please open a PR to add it to this list.
 

README_CN.md

@@ -27,10 +27,6 @@ GLM CODING PLAN 是专为AI编码打造的订阅套餐，每月最低仅需20元
 <td>感谢 PackyCode 对本项目的赞助！PackyCode 是一家可靠高效的 API 中转服务商，提供 Claude Code、Codex、Gemini 等多种服务的中转。PackyCode 为本软件用户提供了特别优惠：使用<a href="https://www.packyapi.com/register?aff=cliproxyapi">此链接</a>注册，并在充值时输入 "cliproxyapi" 优惠码即可享受九折优惠。</td>
 </tr>
 <tr>
-<td width="180"><a href="https://cubence.com/signup?code=CLIPROXYAPI&source=cpa"><img src="./assets/cubence.png" alt="Cubence" width="150"></a></td>
-<td>感谢 Cubence 对本项目的赞助！Cubence 是一家可靠高效的 API 中转服务商，提供 Claude Code、Codex、Gemini 等多种服务的中转。Cubence 为本软件用户提供了特别优惠：使用<a href="https://cubence.com/signup?code=CLIPROXYAPI&source=cpa">此链接</a>注册，并在充值时输入 "CLIPROXYAPI" 优惠码即可享受九折优惠。</td>
-</tr>
-<tr>
 <td width="180"><a href="https://www.aicodemirror.com/register?invitecode=TJNAIF"><img src="./assets/aicodemirror.png" alt="AICodeMirror" width="150"></a></td>
 <td>感谢 AICodeMirror 赞助了本项目！AICodeMirror 提供 Claude Code / Codex / Gemini CLI 官方高稳定中转服务，支持企业级高并发、极速开票、7×24 专属技术支持。 Claude Code / Codex / Gemini 官方渠道低至 3.8 / 0.2 / 0.9 折，充值更有折上折！AICodeMirror 为 CLIProxyAPI 的用户提供了特别福利，通过<a href="https://www.aicodemirror.com/register?invitecode=TJNAIF">此链接</a>注册的用户，可享受首充8折，企业客户最高可享 7.5 折！</td>
 </tr>
@@ -149,6 +145,10 @@ Windows 托盘应用，基于 PowerShell 脚本实现，不依赖任何第三方
 
 霖君是一款用于管理AI编程助手的跨平台桌面应用，支持macOS、Windows、Linux系统。统一管理Claude Code、Gemini CLI、OpenAI Codex、Qwen Code等AI编程工具，本地代理实现多账户配额跟踪和一键配置。
 
+### [CLIProxyAPI Dashboard](https://github.com/itsmylife44/cliproxyapi-dashboard)
+
+一个面向 CLIProxyAPI 的现代化 Web 管理仪表盘，基于 Next.js、React 和 PostgreSQL 构建。支持实时日志流、结构化配置编辑、API Key 管理、Claude/Gemini/Codex 的 OAuth 提供方集成、使用量分析、容器管理，并可通过配套插件与 OpenCode 同步配置，无需手动编辑 YAML。
+
 > [!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。
 
@@ -160,6 +160,12 @@ Windows 托盘应用，基于 PowerShell 脚本实现，不依赖任何第三方
 
 基于 Next.js 的实现，灵感来自 CLIProxyAPI，易于安装使用；自研格式转换（OpenAI/Claude/Gemini/Ollama）、组合系统与自动回退、多账户管理（指数退避）、Next.js Web 控制台，并支持 Cursor、Claude Code、Cline、RooCode 等 CLI 工具，无需 API 密钥。
 
+### [OmniRoute](https://github.com/diegosouzapw/OmniRoute)
+
+代码不止，创新不停。智能路由至免费及低成本 AI 模型，并支持自动故障转移。
+
+OmniRoute 是一个面向多供应商大语言模型的 AI 网关：它提供兼容 OpenAI 的端点，具备智能路由、负载均衡、重试及回退机制。通过添加策略、速率限制、缓存和可观测性，确保推理过程既可靠又具备成本意识。
+
 > [!NOTE]  
 > 如果你开发了 CLIProxyAPI 的移植或衍生项目，请提交 PR 将其添加到此列表中。
 

cmd/server/main.go

@@ -8,6 +8,7 @@ import (
 	"errors"
 	"flag"
 	"fmt"
+	"io"
 	"io/fs"
 	"net/url"
 	"os"
@@ -25,6 +26,7 @@ import (
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/misc"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/store"
 	_ "github.com/router-for-me/CLIProxyAPI/v6/internal/translator"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/tui"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/usage"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 	sdkAuth "github.com/router-for-me/CLIProxyAPI/v6/sdk/auth"
@@ -56,6 +58,7 @@ func main() {
 	// Command-line flags to control the application's behavior.
 	var login bool
 	var codexLogin bool
+	var codexDeviceLogin bool
 	var claudeLogin bool
 	var qwenLogin bool
 	var iflowLogin bool
@@ -63,14 +66,18 @@ func main() {
 	var noBrowser bool
 	var oauthCallbackPort int
 	var antigravityLogin bool
+	var kimiLogin bool
 	var projectID string
 	var vertexImport string
 	var configPath string
 	var password string
+	var tuiMode bool
+	var standalone bool
 
 	// Define command-line flags for different operation modes.
 	flag.BoolVar(&login, "login", false, "Login Google Account")
 	flag.BoolVar(&codexLogin, "codex-login", false, "Login to Codex using OAuth")
+	flag.BoolVar(&codexDeviceLogin, "codex-device-login", false, "Login to Codex using device code flow")
 	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")
@@ -78,10 +85,13 @@ func main() {
 	flag.BoolVar(&noBrowser, "no-browser", false, "Don't open browser automatically for OAuth")
 	flag.IntVar(&oauthCallbackPort, "oauth-callback-port", 0, "Override OAuth callback port (defaults to provider-specific port)")
 	flag.BoolVar(&antigravityLogin, "antigravity-login", false, "Login to Antigravity using OAuth")
+	flag.BoolVar(&kimiLogin, "kimi-login", false, "Login to Kimi 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")
 	flag.StringVar(&password, "password", "", "")
+	flag.BoolVar(&tuiMode, "tui", false, "Start with terminal management UI")
+	flag.BoolVar(&standalone, "standalone", false, "In TUI mode, start an embedded local server")
 
 	flag.CommandLine.Usage = func() {
 		out := flag.CommandLine.Output()
@@ -443,7 +453,7 @@ func main() {
 	}
 
 	// Register built-in access providers before constructing services.
-	configaccess.Register()
+	configaccess.Register(&cfg.SDKConfig)
 
 	// Handle different command modes based on the provided flags.
 
@@ -459,6 +469,9 @@ func main() {
 	} else if codexLogin {
 		// Handle Codex login
 		cmd.DoCodexLogin(cfg, options)
+	} else if codexDeviceLogin {
+		// Handle Codex device-code login
+		cmd.DoCodexDeviceLogin(cfg, options)
 	} else if claudeLogin {
 		// Handle Claude login
 		cmd.DoClaudeLogin(cfg, options)
@@ -468,6 +481,8 @@ func main() {
 		cmd.DoIFlowLogin(cfg, options)
 	} else if iflowCookie {
 		cmd.DoIFlowCookieAuth(cfg, options)
+	} else if kimiLogin {
+		cmd.DoKimiLogin(cfg, options)
 	} else {
 		// In cloud deploy mode without config file, just wait for shutdown signals
 		if isCloudDeploy && !configFileExists {
@@ -475,8 +490,83 @@ func main() {
 			cmd.WaitForCloudDeploy()
 			return
 		}
-		// Start the main proxy service
-		managementasset.StartAutoUpdater(context.Background(), configFilePath)
-		cmd.StartService(cfg, configFilePath, password)
+		if tuiMode {
+			if standalone {
+				// Standalone mode: start an embedded local server and connect TUI client to it.
+				managementasset.StartAutoUpdater(context.Background(), configFilePath)
+				hook := tui.NewLogHook(2000)
+				hook.SetFormatter(&logging.LogFormatter{})
+				log.AddHook(hook)
+
+				origStdout := os.Stdout
+				origStderr := os.Stderr
+				origLogOutput := log.StandardLogger().Out
+				log.SetOutput(io.Discard)
+
+				devNull, errOpenDevNull := os.Open(os.DevNull)
+				if errOpenDevNull == nil {
+					os.Stdout = devNull
+					os.Stderr = devNull
+				}
+
+				restoreIO := func() {
+					os.Stdout = origStdout
+					os.Stderr = origStderr
+					log.SetOutput(origLogOutput)
+					if devNull != nil {
+						_ = devNull.Close()
+					}
+				}
+
+				localMgmtPassword := fmt.Sprintf("tui-%d-%d", os.Getpid(), time.Now().UnixNano())
+				if password == "" {
+					password = localMgmtPassword
+				}
+
+				cancel, done := cmd.StartServiceBackground(cfg, configFilePath, password)
+
+				client := tui.NewClient(cfg.Port, password)
+				ready := false
+				backoff := 100 * time.Millisecond
+				for i := 0; i < 30; i++ {
+					if _, errGetConfig := client.GetConfig(); errGetConfig == nil {
+						ready = true
+						break
+					}
+					time.Sleep(backoff)
+					if backoff < time.Second {
+						backoff = time.Duration(float64(backoff) * 1.5)
+					}
+				}
+
+				if !ready {
+					restoreIO()
+					cancel()
+					<-done
+					fmt.Fprintf(os.Stderr, "TUI error: embedded server is not ready\n")
+					return
+				}
+
+				if errRun := tui.Run(cfg.Port, password, hook, origStdout); errRun != nil {
+					restoreIO()
+					fmt.Fprintf(os.Stderr, "TUI error: %v\n", errRun)
+				} else {
+					restoreIO()
+				}
+
+				cancel()
+				<-done
+			} else {
+				// Default TUI mode: pure management client.
+				// The proxy server must already be running.
+				if errRun := tui.Run(cfg.Port, password, nil, os.Stdout); errRun != nil {
+					fmt.Fprintf(os.Stderr, "TUI error: %v\n", errRun)
+				}
+			}
+		} else {
+			// Start the main proxy service
+			managementasset.StartAutoUpdater(context.Background(), configFilePath)
+			cmd.StartService(cfg, configFilePath, password)
+		}
 	}
 }

config.example.yaml

@@ -68,6 +68,10 @@ proxy-url: ""
 # When true, unprefixed model requests only use credentials without a prefix (except when prefix == model name).
 force-model-prefix: false
 
+# When true, forward filtered upstream response headers to downstream clients.
+# Default is false (disabled).
+passthrough-headers: false
+
 # 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
 
@@ -155,6 +159,15 @@ nonstream-keepalive-interval: 0
 #       sensitive-words:             # optional: words to obfuscate with zero-width characters
 #         - "API"
 #         - "proxy"
+#       cache-user-id: true          # optional: default is false; set true to reuse cached user_id per API key instead of generating a random one each request
+
+# Default headers for Claude API requests. Update when Claude Code releases new versions.
+# These are used as fallbacks when the client does not send its own headers.
+# claude-header-defaults:
+#   user-agent: "claude-cli/2.1.44 (external, sdk-cli)"
+#   package-version: "0.74.0"
+#   runtime-version: "v24.3.0"
+#   timeout: "600"
 
 # OpenAI compatibility providers
 # openai-compatibility:
@@ -221,25 +234,10 @@ nonstream-keepalive-interval: 0
 
 # Global OAuth model name aliases (per channel)
 # These aliases rename model IDs for both model listing and request routing.
-# Supported channels: gemini-cli, vertex, aistudio, antigravity, claude, codex, qwen, iflow.
+# Supported channels: gemini-cli, vertex, aistudio, antigravity, claude, codex, qwen, iflow, kimi.
 # NOTE: Aliases do not apply to gemini-api-key, codex-api-key, claude-api-key, openai-compatibility, vertex-api-key, or ampcode.
 # You can repeat the same name with different aliases to expose multiple client model names.
-oauth-model-alias:
-  antigravity:
-    - name: "rev19-uic3-1p"
-      alias: "gemini-2.5-computer-use-preview-10-2025"
-    - name: "gemini-3-pro-image"
-      alias: "gemini-3-pro-image-preview"
-    - name: "gemini-3-pro-high"
-      alias: "gemini-3-pro-preview"
-    - name: "gemini-3-flash"
-      alias: "gemini-3-flash-preview"
-    - name: "claude-sonnet-4-5"
-      alias: "gemini-claude-sonnet-4-5"
-    - name: "claude-sonnet-4-5-thinking"
-      alias: "gemini-claude-sonnet-4-5-thinking"
-    - name: "claude-opus-4-5-thinking"
-      alias: "gemini-claude-opus-4-5-thinking"
+# oauth-model-alias:
 #   gemini-cli:
 #     - name: "gemini-2.5-pro"          # original model name under this channel
 #       alias: "g2.5p"                  # client-visible alias
@@ -250,6 +248,9 @@ oauth-model-alias:
 #   aistudio:
 #     - name: "gemini-2.5-pro"
 #       alias: "g2.5p"
+#   antigravity:
+#     - name: "gemini-3-pro-high"
+#       alias: "gemini-3-pro-preview"
 #   claude:
 #     - name: "claude-sonnet-4-5-20250929"
 #       alias: "cs4.5"
@@ -262,6 +263,9 @@ oauth-model-alias:
 #   iflow:
 #     - name: "glm-4.7"
 #       alias: "glm-god"
+#   kimi:
+#     - name: "kimi-k2.5"
+#       alias: "k2.5"
 
 # OAuth provider excluded models
 # oauth-excluded-models:
@@ -284,6 +288,8 @@ oauth-model-alias:
 #     - "vision-model"
 #   iflow:
 #     - "tstars2.0"
+#   kimi:
+#     - "kimi-k2-thinking"
 
 # Optional payload configuration
 # payload:

docs/sdk-access.md

@@ -7,80 +7,71 @@ The `github.com/router-for-me/CLIProxyAPI/v6/sdk/access` package centralizes inb
 ```go
 import (
     sdkaccess "github.com/router-for-me/CLIProxyAPI/v6/sdk/access"
-    "github.com/router-for-me/CLIProxyAPI/v6/internal/config"
 )
 ```
 
 Add the module with `go get github.com/router-for-me/CLIProxyAPI/v6/sdk/access`.
 
+## Provider Registry
+
+Providers are registered globally and then attached to a `Manager` as a snapshot:
+
+- `RegisterProvider(type, provider)` installs a pre-initialized provider instance.
+- Registration order is preserved the first time each `type` is seen.
+- `RegisteredProviders()` returns the providers in that order.
+
 ## Manager Lifecycle
 
 ```go
 manager := sdkaccess.NewManager()
-providers, err := sdkaccess.BuildProviders(cfg)
-if err != nil {
-    return err
-}
-manager.SetProviders(providers)
+manager.SetProviders(sdkaccess.RegisteredProviders())
 ```
 
 * `NewManager` constructs an empty manager.
 * `SetProviders` replaces the provider slice using a defensive copy.
 * `Providers` retrieves a snapshot that can be iterated safely from other goroutines.
-* `BuildProviders` translates `config.Config` access declarations into runnable providers. When the config omits explicit providers but defines inline API keys, the helper auto-installs the built-in `config-api-key` provider.
+
+If the manager itself is `nil` or no providers are configured, the call returns `nil, nil`, allowing callers to treat access control as disabled.
 
 ## Authenticating Requests
 
 ```go
-result, err := manager.Authenticate(ctx, req)
+result, authErr := manager.Authenticate(ctx, req)
 switch {
-case err == nil:
+case authErr == nil:
     // Authentication succeeded; result describes the provider and principal.
-case errors.Is(err, sdkaccess.ErrNoCredentials):
+case sdkaccess.IsAuthErrorCode(authErr, sdkaccess.AuthErrorCodeNoCredentials):
     // No recognizable credentials were supplied.
-case errors.Is(err, sdkaccess.ErrInvalidCredential):
+case sdkaccess.IsAuthErrorCode(authErr, sdkaccess.AuthErrorCodeInvalidCredential):
     // Supplied credentials were present but rejected.
 default:
-    // Transport-level failure was returned by a provider.
+    // Internal/transport failure was returned by a provider.
 }
 ```
 
-`Manager.Authenticate` walks the configured providers in order. It returns on the first success, skips providers that surface `ErrNotHandled`, and tracks whether any provider reported `ErrNoCredentials` or `ErrInvalidCredential` for downstream error reporting.
-
-If the manager itself is `nil` or no providers are registered, the call returns `nil, nil`, allowing callers to treat access control as disabled without branching on errors.
+`Manager.Authenticate` walks the configured providers in order. It returns on the first success, skips providers that return `AuthErrorCodeNotHandled`, and aggregates `AuthErrorCodeNoCredentials` / `AuthErrorCodeInvalidCredential` for a final result.
 
 Each `Result` includes the provider identifier, the resolved principal, and optional metadata (for example, which header carried the credential).
 
-## Configuration Layout
+## Built-in `config-api-key` Provider
 
-The manager expects access providers under the `auth.providers` key inside `config.yaml`:
+The proxy includes one built-in access provider:
+
+- `config-api-key`: Validates API keys declared under top-level `api-keys`.
+  - Credential sources: `Authorization: Bearer`, `X-Goog-Api-Key`, `X-Api-Key`, `?key=`, `?auth_token=`
+  - Metadata: `Result.Metadata["source"]` is set to the matched source label.
+
+In the CLI server and `sdk/cliproxy`, this provider is registered automatically based on the loaded configuration.
 
 ```yaml
-auth:
-  providers:
-    - name: inline-api
-      type: config-api-key
-      api-keys:
-        - sk-test-123
-        - sk-prod-456
+api-keys:
+  - sk-test-123
+  - sk-prod-456
 ```
 
-Fields map directly to `config.AccessProvider`: `name` labels the provider, `type` selects the registered factory, `sdk` can name an external module, `api-keys` seeds inline credentials, and `config` passes provider-specific options.
+## Loading Providers from External Go Modules
 
-### Loading providers from external SDK modules
-
-To consume a provider shipped in another Go module, point the `sdk` field at the module path and import it for its registration side effect:
-
-```yaml
-auth:
-  providers:
-    - name: partner-auth
-      type: partner-token
-      sdk: github.com/acme/xplatform/sdk/access/providers/partner
-      config:
-        region: us-west-2
-        audience: cli-proxy
-```
+To consume a provider shipped in another Go module, import it for its registration side effect:
 
 ```go
 import (
@@ -89,19 +80,11 @@ import (
 )
 ```
 
-The blank identifier import ensures `init` runs so `sdkaccess.RegisterProvider` executes before `BuildProviders` is called.
-
-## Built-in Providers
-
-The SDK ships with one provider out of the box:
-
-- `config-api-key`: Validates API keys declared inline or under top-level `api-keys`. It accepts the key from `Authorization: Bearer`, `X-Goog-Api-Key`, `X-Api-Key`, or the `?key=` query string and reports `ErrInvalidCredential` when no match is found.
-
-Additional providers can be delivered by third-party packages. When a provider package is imported, it registers itself with `sdkaccess.RegisterProvider`.
+The blank identifier import ensures `init` runs so `sdkaccess.RegisterProvider` executes before you call `RegisteredProviders()` (or before `cliproxy.NewBuilder().Build()`).
 
 ### Metadata and auditing
 
-`Result.Metadata` carries provider-specific context. The built-in `config-api-key` provider, for example, stores the credential source (`authorization`, `x-goog-api-key`, `x-api-key`, or `query-key`). Populate this map in custom providers to enrich logs and downstream auditing.
+`Result.Metadata` carries provider-specific context. The built-in `config-api-key` provider, for example, stores the credential source (`authorization`, `x-goog-api-key`, `x-api-key`, `query-key`, `query-auth-token`). Populate this map in custom providers to enrich logs and downstream auditing.
 
 ## Writing Custom Providers
 
@@ -110,13 +93,13 @@ type customProvider struct{}
 
 func (p *customProvider) Identifier() string { return "my-provider" }
 
-func (p *customProvider) Authenticate(ctx context.Context, r *http.Request) (*sdkaccess.Result, error) {
+func (p *customProvider) Authenticate(ctx context.Context, r *http.Request) (*sdkaccess.Result, *sdkaccess.AuthError) {
     token := r.Header.Get("X-Custom")
     if token == "" {
-        return nil, sdkaccess.ErrNoCredentials
+        return nil, sdkaccess.NewNotHandledError()
     }
     if token != "expected" {
-        return nil, sdkaccess.ErrInvalidCredential
+        return nil, sdkaccess.NewInvalidCredentialError()
     }
     return &sdkaccess.Result{
         Provider:  p.Identifier(),
@@ -126,51 +109,46 @@ func (p *customProvider) Authenticate(ctx context.Context, r *http.Request) (*sd
 }
 
 func init() {
-    sdkaccess.RegisterProvider("custom", func(cfg *config.AccessProvider, root *config.Config) (sdkaccess.Provider, error) {
-        return &customProvider{}, nil
-    })
+    sdkaccess.RegisterProvider("custom", &customProvider{})
 }
 ```
 
-A provider must implement `Identifier()` and `Authenticate()`. To expose it to configuration, call `RegisterProvider` inside `init`. Provider factories receive the specific `AccessProvider` block plus the full root configuration for contextual needs.
+A provider must implement `Identifier()` and `Authenticate()`. To make it available to the access manager, call `RegisterProvider` inside `init` with an initialized provider instance.
 
 ## Error Semantics
 
-- `ErrNoCredentials`: no credentials were present or recognized by any provider.
-- `ErrInvalidCredential`: at least one provider processed the credentials but rejected them.
-- `ErrNotHandled`: instructs the manager to fall through to the next provider without affecting aggregate error reporting.
+- `NewNoCredentialsError()` (`AuthErrorCodeNoCredentials`): no credentials were present or recognized. (HTTP 401)
+- `NewInvalidCredentialError()` (`AuthErrorCodeInvalidCredential`): credentials were present but rejected. (HTTP 401)
+- `NewNotHandledError()` (`AuthErrorCodeNotHandled`): fall through to the next provider.
+- `NewInternalAuthError(message, cause)` (`AuthErrorCodeInternal`): transport/system failure. (HTTP 500)
 
-Return custom errors to surface transport failures; they propagate immediately to the caller instead of being masked.
+Errors propagate immediately to the caller unless they are classified as `not_handled` / `no_credentials` / `invalid_credential` and can be aggregated by the manager.
 
 ## Integration with cliproxy Service
 
-`sdk/cliproxy` wires `@sdk/access` automatically when you build a CLI service via `cliproxy.NewBuilder`. Supplying a preconfigured manager allows you to extend or override the default providers:
+`sdk/cliproxy` wires `@sdk/access` automatically when you build a CLI service via `cliproxy.NewBuilder`. Supplying a manager lets you reuse the same instance in your host process:
 
 ```go
 coreCfg, _ := config.LoadConfig("config.yaml")
-providers, _ := sdkaccess.BuildProviders(coreCfg)
-manager := sdkaccess.NewManager()
-manager.SetProviders(providers)
+accessManager := sdkaccess.NewManager()
 
 svc, _ := cliproxy.NewBuilder().
   WithConfig(coreCfg).
-  WithAccessManager(manager).
+  WithConfigPath("config.yaml").
+  WithRequestAccessManager(accessManager).
   Build()
 ```
 
-The service reuses the manager for every inbound request, ensuring consistent authentication across embedded deployments and the canonical CLI binary.
+Register any custom providers (typically via blank imports) before calling `Build()` so they are present in the global registry snapshot.
 
-### Hot reloading providers
+### Hot reloading
 
-When configuration changes, rebuild providers and swap them into the manager:
+When configuration changes, refresh any config-backed providers and then reset the manager's provider chain:
 
 ```go
-providers, err := sdkaccess.BuildProviders(newCfg)
-if err != nil {
-    log.Errorf("reload auth providers failed: %v", err)
-    return
-}
-accessManager.SetProviders(providers)
+// configaccess is github.com/router-for-me/CLIProxyAPI/v6/internal/access/config_access
+configaccess.Register(&newCfg.SDKConfig)
+accessManager.SetProviders(sdkaccess.RegisteredProviders())
 ```
 
-This mirrors the behaviour in `cliproxy.Service.refreshAccessProviders` and `api.Server.applyAccessConfig`, enabling runtime updates without restarting the process.
+This mirrors the behaviour in `internal/access.ApplyAccessProviders`, enabling runtime updates without restarting the process.

docs/sdk-access_CN.md

@@ -7,80 +7,71 @@
 ```go
 import (
     sdkaccess "github.com/router-for-me/CLIProxyAPI/v6/sdk/access"
-    "github.com/router-for-me/CLIProxyAPI/v6/internal/config"
 )
 ```
 
 通过 `go get github.com/router-for-me/CLIProxyAPI/v6/sdk/access` 添加依赖。
 
+## Provider Registry
+
+访问提供者是全局注册，然后以快照形式挂到 `Manager` 上：
+
+- `RegisterProvider(type, provider)` 注册一个已经初始化好的 provider 实例。
+- 每个 `type` 第一次出现时会记录其注册顺序。
+- `RegisteredProviders()` 会按该顺序返回 provider 列表。
+
 ## 管理器生命周期
 
 ```go
 manager := sdkaccess.NewManager()
-providers, err := sdkaccess.BuildProviders(cfg)
-if err != nil {
-    return err
-}
-manager.SetProviders(providers)
+manager.SetProviders(sdkaccess.RegisteredProviders())
 ```
 
 - `NewManager` 创建空管理器。
 - `SetProviders` 替换提供者切片并做防御性拷贝。
 - `Providers` 返回适合并发读取的快照。
-- `BuildProviders` 将 `config.Config` 中的访问配置转换成可运行的提供者。当配置没有显式声明但包含顶层 `api-keys` 时，会自动挂载内建的 `config-api-key` 提供者。
+
+如果管理器本身为 `nil` 或未配置任何 provider，调用会返回 `nil, nil`，可视为关闭访问控制。
 
 ## 认证请求
 
 ```go
-result, err := manager.Authenticate(ctx, req)
+result, authErr := manager.Authenticate(ctx, req)
 switch {
-case err == nil:
+case authErr == nil:
     // Authentication succeeded; result carries provider and principal.
-case errors.Is(err, sdkaccess.ErrNoCredentials):
+case sdkaccess.IsAuthErrorCode(authErr, sdkaccess.AuthErrorCodeNoCredentials):
     // No recognizable credentials were supplied.
-case errors.Is(err, sdkaccess.ErrInvalidCredential):
+case sdkaccess.IsAuthErrorCode(authErr, sdkaccess.AuthErrorCodeInvalidCredential):
     // Credentials were present but rejected.
 default:
     // Provider surfaced a transport-level failure.
 }
 ```
 
-`Manager.Authenticate` 按配置顺序遍历提供者。遇到成功立即返回，`ErrNotHandled` 会继续尝试下一个；若发现 `ErrNoCredentials` 或 `ErrInvalidCredential`，会在遍历结束后汇总给调用方。
-
-若管理器本身为 `nil` 或尚未注册提供者，调用会返回 `nil, nil`，让调用方无需针对错误做额外分支即可关闭访问控制。
+`Manager.Authenticate` 会按顺序遍历 provider：遇到成功立即返回，`AuthErrorCodeNotHandled` 会继续尝试下一个；`AuthErrorCodeNoCredentials` / `AuthErrorCodeInvalidCredential` 会在遍历结束后汇总给调用方。
 
 `Result` 提供认证提供者标识、解析出的主体以及可选元数据（例如凭证来源）。
 
-## 配置结构
+## 内建 `config-api-key` Provider
 
-在 `config.yaml` 的 `auth.providers` 下定义访问提供者：
+代理内置一个访问提供者：
+
+- `config-api-key`：校验 `config.yaml` 顶层的 `api-keys`。
+  - 凭证来源：`Authorization: Bearer`、`X-Goog-Api-Key`、`X-Api-Key`、`?key=`、`?auth_token=`
+  - 元数据：`Result.Metadata["source"]` 会写入匹配到的来源标识
+
+在 CLI 服务端与 `sdk/cliproxy` 中，该 provider 会根据加载到的配置自动注册。
 
 ```yaml
-auth:
-  providers:
-    - name: inline-api
-      type: config-api-key
-      api-keys:
-        - sk-test-123
-        - sk-prod-456
+api-keys:
+  - sk-test-123
+  - sk-prod-456
 ```
 
-条目映射到 `config.AccessProvider`：`name` 指定实例名，`type` 选择注册的工厂，`sdk` 可引用第三方模块，`api-keys` 提供内联凭证，`config` 用于传递特定选项。
+## 引入外部 Go 模块提供者
 
-### 引入外部 SDK 提供者
-
-若要消费其它 Go 模块输出的访问提供者，可在配置里填写 `sdk` 字段并在代码中引入该包，利用其 `init` 注册过程：
-
-```yaml
-auth:
-  providers:
-    - name: partner-auth
-      type: partner-token
-      sdk: github.com/acme/xplatform/sdk/access/providers/partner
-      config:
-        region: us-west-2
-        audience: cli-proxy
-```
+若要消费其它 Go 模块输出的访问提供者，直接用空白标识符导入以触发其 `init` 注册即可：
 
 ```go
 import (
@@ -89,19 +80,11 @@ import (
 )
 ```
 
-通过空白标识符导入即可确保 `init` 调用，先于 `BuildProviders` 完成 `sdkaccess.RegisterProvider`。
-
-## 内建提供者
-
-当前 SDK 默认内置：
-
-- `config-api-key`：校验配置中的 API Key。它从 `Authorization: Bearer`、`X-Goog-Api-Key`、`X-Api-Key` 以及查询参数 `?key=` 提取凭证，不匹配时抛出 `ErrInvalidCredential`。
-
-导入第三方包即可通过 `sdkaccess.RegisterProvider` 注册更多类型。
+空白导入可确保 `init` 先执行，从而在你调用 `RegisteredProviders()`（或 `cliproxy.NewBuilder().Build()`）之前完成 `sdkaccess.RegisterProvider`。
 
 ### 元数据与审计
 
-`Result.Metadata` 用于携带提供者特定的上下文信息。内建的 `config-api-key` 会记录凭证来源（`authorization`、`x-goog-api-key`、`x-api-key` 或 `query-key`）。自定义提供者同样可以填充该 Map，以便丰富日志与审计场景。
+`Result.Metadata` 用于携带提供者特定的上下文信息。内建的 `config-api-key` 会记录凭证来源（`authorization`、`x-goog-api-key`、`x-api-key`、`query-key`、`query-auth-token`）。自定义提供者同样可以填充该 Map，以便丰富日志与审计场景。
 
 ## 编写自定义提供者
 
@@ -110,13 +93,13 @@ type customProvider struct{}
 
 func (p *customProvider) Identifier() string { return "my-provider" }
 
-func (p *customProvider) Authenticate(ctx context.Context, r *http.Request) (*sdkaccess.Result, error) {
+func (p *customProvider) Authenticate(ctx context.Context, r *http.Request) (*sdkaccess.Result, *sdkaccess.AuthError) {
     token := r.Header.Get("X-Custom")
     if token == "" {
-        return nil, sdkaccess.ErrNoCredentials
+        return nil, sdkaccess.NewNotHandledError()
     }
     if token != "expected" {
-        return nil, sdkaccess.ErrInvalidCredential
+        return nil, sdkaccess.NewInvalidCredentialError()
     }
     return &sdkaccess.Result{
         Provider:  p.Identifier(),
@@ -126,51 +109,46 @@ func (p *customProvider) Authenticate(ctx context.Context, r *http.Request) (*sd
 }
 
 func init() {
-    sdkaccess.RegisterProvider("custom", func(cfg *config.AccessProvider, root *config.Config) (sdkaccess.Provider, error) {
-        return &customProvider{}, nil
-    })
+    sdkaccess.RegisterProvider("custom", &customProvider{})
 }
 ```
 
-自定义提供者需要实现 `Identifier()` 与 `Authenticate()`。在 `init` 中调用 `RegisterProvider` 暴露给配置层，工厂函数既能读取当前条目，也能访问完整根配置。
+自定义提供者需要实现 `Identifier()` 与 `Authenticate()`。在 `init` 中用已初始化实例调用 `RegisterProvider` 注册到全局 registry。
 
 ## 错误语义
 
-- `ErrNoCredentials`：任何提供者都未识别到凭证。
-- `ErrInvalidCredential`：至少一个提供者处理了凭证但判定无效。
-- `ErrNotHandled`：告诉管理器跳到下一个提供者，不影响最终错误统计。
+- `NewNoCredentialsError()`（`AuthErrorCodeNoCredentials`）：未提供或未识别到凭证。（HTTP 401）
+- `NewInvalidCredentialError()`（`AuthErrorCodeInvalidCredential`）：凭证存在但校验失败。（HTTP 401）
+- `NewNotHandledError()`（`AuthErrorCodeNotHandled`）：告诉管理器跳到下一个 provider。
+- `NewInternalAuthError(message, cause)`（`AuthErrorCodeInternal`）：网络/系统错误。（HTTP 500）
 
-自定义错误（例如网络异常）会马上冒泡返回。
+除可汇总的 `not_handled` / `no_credentials` / `invalid_credential` 外，其它错误会立即冒泡返回。
 
 ## 与 cliproxy 集成
 
-使用 `sdk/cliproxy` 构建服务时会自动接入 `@sdk/access`。如果需要扩展内置行为，可传入自定义管理器：
+使用 `sdk/cliproxy` 构建服务时会自动接入 `@sdk/access`。如果希望在宿主进程里复用同一个 `Manager` 实例，可传入自定义管理器：
 
 ```go
 coreCfg, _ := config.LoadConfig("config.yaml")
-providers, _ := sdkaccess.BuildProviders(coreCfg)
-manager := sdkaccess.NewManager()
-manager.SetProviders(providers)
+accessManager := sdkaccess.NewManager()
 
 svc, _ := cliproxy.NewBuilder().
   WithConfig(coreCfg).
-  WithAccessManager(manager).
+  WithConfigPath("config.yaml").
+  WithRequestAccessManager(accessManager).
   Build()
 ```
 
-服务会复用该管理器处理每一个入站请求，实现与 CLI 二进制一致的访问控制体验。
+请在调用 `Build()` 之前完成自定义 provider 的注册（通常通过空白导入触发 `init`），以确保它们被包含在全局 registry 的快照中。
 
 ### 动态热更新提供者
 
-当配置发生变化时，可以重新构建提供者并替换当前列表：
+当配置发生变化时，刷新依赖配置的 provider，然后重置 manager 的 provider 链：
 
 ```go
-providers, err := sdkaccess.BuildProviders(newCfg)
-if err != nil {
-    log.Errorf("reload auth providers failed: %v", err)
-    return
-}
-accessManager.SetProviders(providers)
+// configaccess is github.com/router-for-me/CLIProxyAPI/v6/internal/access/config_access
+configaccess.Register(&newCfg.SDKConfig)
+accessManager.SetProviders(sdkaccess.RegisteredProviders())
 ```
 
-这一流程与 `cliproxy.Service.refreshAccessProviders` 和 `api.Server.applyAccessConfig` 保持一致，避免为更新访问策略而重启进程。
+这一流程与 `internal/access.ApplyAccessProviders` 保持一致，避免为更新访问策略而重启进程。

examples/custom-provider/main.go

@@ -159,13 +159,13 @@ func (MyExecutor) CountTokens(context.Context, *coreauth.Auth, clipexec.Request,
 	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) {
+func (MyExecutor) ExecuteStream(ctx context.Context, a *coreauth.Auth, req clipexec.Request, opts clipexec.Options) (*clipexec.StreamResult, error) {
 	ch := make(chan clipexec.StreamChunk, 1)
 	go func() {
 		defer close(ch)
 		ch <- clipexec.StreamChunk{Payload: []byte("data: {\"ok\":true}\n\n")}
 	}()
-	return ch, nil
+	return &clipexec.StreamResult{Chunks: ch}, nil
 }
 
 func (MyExecutor) Refresh(ctx context.Context, a *coreauth.Auth) (*coreauth.Auth, error) {

examples/http-request/main.go

@@ -58,7 +58,7 @@ func (EchoExecutor) Execute(context.Context, *coreauth.Auth, clipexec.Request, c
 	return clipexec.Response{}, errors.New("echo executor: Execute not implemented")
 }
 
-func (EchoExecutor) ExecuteStream(context.Context, *coreauth.Auth, clipexec.Request, clipexec.Options) (<-chan clipexec.StreamChunk, error) {
+func (EchoExecutor) ExecuteStream(context.Context, *coreauth.Auth, clipexec.Request, clipexec.Options) (*clipexec.StreamResult, error) {
 	return nil, errors.New("echo executor: ExecuteStream not implemented")
 }
 

go.mod

@@ -1,9 +1,13 @@
 module github.com/router-for-me/CLIProxyAPI/v6
 
-go 1.24.0
+go 1.26.0
 
 require (
 	github.com/andybalholm/brotli v1.0.6
+	github.com/atotto/clipboard v0.1.4
+	github.com/charmbracelet/bubbles v1.0.0
+	github.com/charmbracelet/bubbletea v1.3.10
+	github.com/charmbracelet/lipgloss v1.1.0
 	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
@@ -22,6 +26,7 @@ require (
 	golang.org/x/crypto v0.45.0
 	golang.org/x/net v0.47.0
 	golang.org/x/oauth2 v0.30.0
+	golang.org/x/sync v0.18.0
 	gopkg.in/natefinch/lumberjack.v2 v2.2.1
 	gopkg.in/yaml.v3 v3.0.1
 )
@@ -30,8 +35,16 @@ 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/aymanbagabas/go-osc52/v2 v2.0.1 // indirect
 	github.com/bytedance/sonic v1.11.6 // indirect
 	github.com/bytedance/sonic/loader v0.1.1 // indirect
+	github.com/charmbracelet/colorprofile v0.4.1 // indirect
+	github.com/charmbracelet/x/ansi v0.11.6 // indirect
+	github.com/charmbracelet/x/cellbuf v0.0.15 // indirect
+	github.com/charmbracelet/x/term v0.2.2 // indirect
+	github.com/clipperhouse/displaywidth v0.9.0 // indirect
+	github.com/clipperhouse/stringish v0.1.1 // indirect
+	github.com/clipperhouse/uax29/v2 v2.5.0 // indirect
 	github.com/cloudflare/circl v1.6.1 // indirect
 	github.com/cloudwego/base64x v0.1.4 // indirect
 	github.com/cloudwego/iasm v0.2.0 // indirect
@@ -39,6 +52,7 @@ require (
 	github.com/dlclark/regexp2 v1.11.5 // indirect
 	github.com/dustin/go-humanize v1.0.1 // indirect
 	github.com/emirpasic/gods v1.18.1 // indirect
+	github.com/erikgeiser/coninput v0.0.0-20211004153227-1c3628e74d0f // indirect
 	github.com/gabriel-vasile/mimetype v1.4.3 // indirect
 	github.com/gin-contrib/sse v0.1.0 // indirect
 	github.com/go-git/gcfg/v2 v2.0.2 // indirect
@@ -55,21 +69,28 @@ require (
 	github.com/kevinburke/ssh_config v1.4.0 // indirect
 	github.com/klauspost/cpuid/v2 v2.3.0 // indirect
 	github.com/leodido/go-urn v1.4.0 // indirect
+	github.com/lucasb-eyer/go-colorful v1.3.0 // indirect
 	github.com/mattn/go-isatty v0.0.20 // indirect
+	github.com/mattn/go-localereader v0.0.1 // indirect
+	github.com/mattn/go-runewidth v0.0.19 // indirect
 	github.com/minio/md5-simd v1.1.2 // indirect
 	github.com/minio/sha256-simd v1.0.1 // indirect
 	github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd // indirect
 	github.com/modern-go/reflect2 v1.0.2 // indirect
+	github.com/muesli/ansi v0.0.0-20230316100256-276c6243b2f6 // indirect
+	github.com/muesli/cancelreader v0.2.2 // indirect
+	github.com/muesli/termenv v0.16.0 // indirect
 	github.com/pelletier/go-toml/v2 v2.2.2 // indirect
 	github.com/pjbgf/sha1cd v0.5.0 // indirect
+	github.com/rivo/uniseg v0.4.7 // indirect
 	github.com/rs/xid v1.5.0 // indirect
 	github.com/sergi/go-diff v1.4.0 // indirect
 	github.com/tidwall/match v1.1.1 // indirect
 	github.com/tidwall/pretty v1.2.0 // indirect
 	github.com/twitchyliquid64/golang-asm v0.15.1 // indirect
 	github.com/ugorji/go/codec v1.2.12 // indirect
+	github.com/xo/terminfo v0.0.0-20220910002029-abceb7e1c41e // indirect
 	golang.org/x/arch v0.8.0 // indirect
-	golang.org/x/sync v0.18.0 // indirect
 	golang.org/x/sys v0.38.0 // indirect
 	golang.org/x/text v0.31.0 // indirect
 	google.golang.org/protobuf v1.34.1 // indirect

go.sum

@@ -10,10 +10,34 @@ github.com/anmitsu/go-shlex v0.0.0-20200514113438-38f4b401e2be h1:9AeTilPcZAjCFI
 github.com/anmitsu/go-shlex v0.0.0-20200514113438-38f4b401e2be/go.mod h1:ySMOLuWl6zY27l47sB3qLNK6tF2fkHG55UZxx8oIVo4=
 github.com/armon/go-socks5 v0.0.0-20160902184237-e75332964ef5 h1:0CwZNZbxp69SHPdPJAN/hZIm0C4OItdklCFmMRWYpio=
 github.com/armon/go-socks5 v0.0.0-20160902184237-e75332964ef5/go.mod h1:wHh0iHkYZB8zMSxRWpUBQtwG5a7fFgvEO+odwuTv2gs=
+github.com/atotto/clipboard v0.1.4 h1:EH0zSVneZPSuFR11BlR9YppQTVDbh5+16AmcJi4g1z4=
+github.com/atotto/clipboard v0.1.4/go.mod h1:ZY9tmq7sm5xIbd9bOK4onWV4S6X0u6GY7Vn0Yu86PYI=
+github.com/aymanbagabas/go-osc52/v2 v2.0.1 h1:HwpRHbFMcZLEVr42D4p7XBqjyuxQH5SMiErDT4WkJ2k=
+github.com/aymanbagabas/go-osc52/v2 v2.0.1/go.mod h1:uYgXzlJ7ZpABp8OJ+exZzJJhRNQ2ASbcXHWsFqH8hp8=
 github.com/bytedance/sonic v1.11.6 h1:oUp34TzMlL+OY1OUWxHqsdkgC/Zfc85zGqw9siXjrc0=
 github.com/bytedance/sonic v1.11.6/go.mod h1:LysEHSvpvDySVdC2f87zGWf6CIKJcAvqab1ZaiQtds4=
 github.com/bytedance/sonic/loader v0.1.1 h1:c+e5Pt1k/cy5wMveRDyk2X4B9hF4g7an8N3zCYjJFNM=
 github.com/bytedance/sonic/loader v0.1.1/go.mod h1:ncP89zfokxS5LZrJxl5z0UJcsk4M4yY2JpfqGeCtNLU=
+github.com/charmbracelet/bubbles v1.0.0 h1:12J8/ak/uCZEMQ6KU7pcfwceyjLlWsDLAxB5fXonfvc=
+github.com/charmbracelet/bubbles v1.0.0/go.mod h1:9d/Zd5GdnauMI5ivUIVisuEm3ave1XwXtD1ckyV6r3E=
+github.com/charmbracelet/bubbletea v1.3.10 h1:otUDHWMMzQSB0Pkc87rm691KZ3SWa4KUlvF9nRvCICw=
+github.com/charmbracelet/bubbletea v1.3.10/go.mod h1:ORQfo0fk8U+po9VaNvnV95UPWA1BitP1E0N6xJPlHr4=
+github.com/charmbracelet/colorprofile v0.4.1 h1:a1lO03qTrSIRaK8c3JRxJDZOvhvIeSco3ej+ngLk1kk=
+github.com/charmbracelet/colorprofile v0.4.1/go.mod h1:U1d9Dljmdf9DLegaJ0nGZNJvoXAhayhmidOdcBwAvKk=
+github.com/charmbracelet/lipgloss v1.1.0 h1:vYXsiLHVkK7fp74RkV7b2kq9+zDLoEU4MZoFqR/noCY=
+github.com/charmbracelet/lipgloss v1.1.0/go.mod h1:/6Q8FR2o+kj8rz4Dq0zQc3vYf7X+B0binUUBwA0aL30=
+github.com/charmbracelet/x/ansi v0.11.6 h1:GhV21SiDz/45W9AnV2R61xZMRri5NlLnl6CVF7ihZW8=
+github.com/charmbracelet/x/ansi v0.11.6/go.mod h1:2JNYLgQUsyqaiLovhU2Rv/pb8r6ydXKS3NIttu3VGZQ=
+github.com/charmbracelet/x/cellbuf v0.0.15 h1:ur3pZy0o6z/R7EylET877CBxaiE1Sp1GMxoFPAIztPI=
+github.com/charmbracelet/x/cellbuf v0.0.15/go.mod h1:J1YVbR7MUuEGIFPCaaZ96KDl5NoS0DAWkskup+mOY+Q=
+github.com/charmbracelet/x/term v0.2.2 h1:xVRT/S2ZcKdhhOuSP4t5cLi5o+JxklsoEObBSgfgZRk=
+github.com/charmbracelet/x/term v0.2.2/go.mod h1:kF8CY5RddLWrsgVwpw4kAa6TESp6EB5y3uxGLeCqzAI=
+github.com/clipperhouse/displaywidth v0.9.0 h1:Qb4KOhYwRiN3viMv1v/3cTBlz3AcAZX3+y9OLhMtAtA=
+github.com/clipperhouse/displaywidth v0.9.0/go.mod h1:aCAAqTlh4GIVkhQnJpbL0T/WfcrJXHcj8C0yjYcjOZA=
+github.com/clipperhouse/stringish v0.1.1 h1:+NSqMOr3GR6k1FdRhhnXrLfztGzuG+VuFDfatpWHKCs=
+github.com/clipperhouse/stringish v0.1.1/go.mod h1:v/WhFtE1q0ovMta2+m+UbpZ+2/HEXNWYXQgCt4hdOzA=
+github.com/clipperhouse/uax29/v2 v2.5.0 h1:x7T0T4eTHDONxFJsL94uKNKPHrclyFI0lm7+w94cO8U=
+github.com/clipperhouse/uax29/v2 v2.5.0/go.mod h1:Wn1g7MK6OoeDT0vL+Q0SQLDz/KpfsVRgg6W7ihQeh4g=
 github.com/cloudflare/circl v1.6.1 h1:zqIqSPIndyBh1bjLVVDHMPpVKqp8Su/V+6MeDzzQBQ0=
 github.com/cloudflare/circl v1.6.1/go.mod h1:uddAzsPgqdMAYatqJ0lsjX1oECcQLIlRpzZh3pJrofs=
 github.com/cloudwego/base64x v0.1.4 h1:jwCgWpFanWmN8xoIUHa2rtzmkd5J2plF/dnLS6Xd/0Y=
@@ -33,6 +57,8 @@ github.com/elazarl/goproxy v1.7.2 h1:Y2o6urb7Eule09PjlhQRGNsqRfPmYI3KKQLFpCAV3+o
 github.com/elazarl/goproxy v1.7.2/go.mod h1:82vkLNir0ALaW14Rc399OTTjyNREgmdL2cVoIbS6XaE=
 github.com/emirpasic/gods v1.18.1 h1:FXtiHYKDGKCW2KzwZKx0iC0PQmdlorYgdFG9jPXJ1Bc=
 github.com/emirpasic/gods v1.18.1/go.mod h1:8tpGGwCnJ5H4r6BWwaV6OrWmMoPhUl5jm/FMNAnJvWQ=
+github.com/erikgeiser/coninput v0.0.0-20211004153227-1c3628e74d0f h1:Y/CXytFA4m6baUTXGLOoWe4PQhGxaX0KpnayAqC48p4=
+github.com/erikgeiser/coninput v0.0.0-20211004153227-1c3628e74d0f/go.mod h1:vw97MGsxSvLiUE2X8qFplwetxpGLQrlU1Q9AUEIzCaM=
 github.com/fsnotify/fsnotify v1.9.0 h1:2Ml+OJNzbYCTzsxtv8vKSFD9PbJjmhYF14k/jKC7S9k=
 github.com/fsnotify/fsnotify v1.9.0/go.mod h1:8jBTzvmWwFyi3Pb8djgCCO5IBqzKJ/Jwo8TRcHyHii0=
 github.com/gabriel-vasile/mimetype v1.4.3 h1:in2uUcidCuFcDKtdcBxlR0rJ1+fsokWf+uqxgUFjbI0=
@@ -99,8 +125,14 @@ github.com/kr/text v0.1.0 h1:45sCR5RtlFHMR4UwH9sdQ5TC8v0qDQCHnXt+kaKSTVE=
 github.com/kr/text v0.1.0/go.mod h1:4Jbv+DJW3UT/LiOwJeYQe1efqtUx/iVham/4vfdArNI=
 github.com/leodido/go-urn v1.4.0 h1:WT9HwE9SGECu3lg4d/dIA+jxlljEa1/ffXKmRjqdmIQ=
 github.com/leodido/go-urn v1.4.0/go.mod h1:bvxc+MVxLKB4z00jd1z+Dvzr47oO32F/QSNjSBOlFxI=
+github.com/lucasb-eyer/go-colorful v1.3.0 h1:2/yBRLdWBZKrf7gB40FoiKfAWYQ0lqNcbuQwVHXptag=
+github.com/lucasb-eyer/go-colorful v1.3.0/go.mod h1:R4dSotOR9KMtayYi1e77YzuveK+i7ruzyGqttikkLy0=
 github.com/mattn/go-isatty v0.0.20 h1:xfD0iDuEKnDkl03q4limB+vH+GxLEtL/jb4xVJSWWEY=
 github.com/mattn/go-isatty v0.0.20/go.mod h1:W+V8PltTTMOvKvAeJH7IuucS94S2C6jfK/D7dTCTo3Y=
+github.com/mattn/go-localereader v0.0.1 h1:ygSAOl7ZXTx4RdPYinUpg6W99U8jWvWi9Ye2JC/oIi4=
+github.com/mattn/go-localereader v0.0.1/go.mod h1:8fBrzywKY7BI3czFoHkuzRoWE9C+EiG4R1k4Cjx5p88=
+github.com/mattn/go-runewidth v0.0.19 h1:v++JhqYnZuu5jSKrk9RbgF5v4CGUjqRfBm05byFGLdw=
+github.com/mattn/go-runewidth v0.0.19/go.mod h1:XBkDxAl56ILZc9knddidhrOlY5R/pDhgLpndooCuJAs=
 github.com/minio/md5-simd v1.1.2 h1:Gdi1DZK69+ZVMoNHRXJyNcxrMA4dSxoYHZSQbirFg34=
 github.com/minio/md5-simd v1.1.2/go.mod h1:MzdKDxYpY2BT9XQFocsiZf/NKVtR7nkE4RoEpN+20RM=
 github.com/minio/minio-go/v7 v7.0.66 h1:bnTOXOHjOqv/gcMuiVbN9o2ngRItvqE774dG9nq0Dzw=
@@ -112,6 +144,12 @@ github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd h1:TRLaZ9cD/w
 github.com/modern-go/concurrent v0.0.0-20180306012644-bacd9c7ef1dd/go.mod h1:6dJC0mAP4ikYIbvyc7fijjWJddQyLn8Ig3JB5CqoB9Q=
 github.com/modern-go/reflect2 v1.0.2 h1:xBagoLtFs94CBntxluKeaWgTMpvLxC4ur3nMaC9Gz0M=
 github.com/modern-go/reflect2 v1.0.2/go.mod h1:yWuevngMOJpCy52FWWMvUC8ws7m/LJsjYzDa0/r8luk=
+github.com/muesli/ansi v0.0.0-20230316100256-276c6243b2f6 h1:ZK8zHtRHOkbHy6Mmr5D264iyp3TiX5OmNcI5cIARiQI=
+github.com/muesli/ansi v0.0.0-20230316100256-276c6243b2f6/go.mod h1:CJlz5H+gyd6CUWT45Oy4q24RdLyn7Md9Vj2/ldJBSIo=
+github.com/muesli/cancelreader v0.2.2 h1:3I4Kt4BQjOR54NavqnDogx/MIoWBFa0StPA8ELUXHmA=
+github.com/muesli/cancelreader v0.2.2/go.mod h1:3XuTXfFS2VjM+HTLZY9Ak0l6eUKfijIfMUZ4EgX0QYo=
+github.com/muesli/termenv v0.16.0 h1:S5AlUN9dENB57rsbnkPyfdGuWIlkmzJjbFf0Tf5FWUc=
+github.com/muesli/termenv v0.16.0/go.mod h1:ZRfOIKPFDYQoDFF4Olj7/QJbW60Ol/kL1pU3VfY/Cnk=
 github.com/pelletier/go-toml/v2 v2.2.2 h1:aYUidT7k73Pcl9nb2gScu7NSrKCSHIDE89b3+6Wq+LM=
 github.com/pelletier/go-toml/v2 v2.2.2/go.mod h1:1t835xjRzz80PqgE6HHgN2JOsmgYu/h4qDAS4n929Rs=
 github.com/pjbgf/sha1cd v0.5.0 h1:a+UkboSi1znleCDUNT3M5YxjOnN1fz2FhN48FlwCxs0=
@@ -120,6 +158,8 @@ github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZb
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/refraction-networking/utls v1.8.2 h1:j4Q1gJj0xngdeH+Ox/qND11aEfhpgoEvV+S9iJ2IdQo=
 github.com/refraction-networking/utls v1.8.2/go.mod h1:jkSOEkLqn+S/jtpEHPOsVv/4V4EVnelwbMQl4vCWXAM=
+github.com/rivo/uniseg v0.4.7 h1:WUdvkW8uEhrYfLC4ZzdpI2ztxP1I582+49Oc5Mq64VQ=
+github.com/rivo/uniseg v0.4.7/go.mod h1:FN3SvrM+Zdj16jyLfmOkMNblXMcoc8DfTHruCPUcx88=
 github.com/rogpeppe/go-internal v1.14.1 h1:UQB4HGPB6osV0SQTLymcB4TgvyWu6ZyliaW0tI/otEQ=
 github.com/rogpeppe/go-internal v1.14.1/go.mod h1:MaRKkUm5W0goXpeCfT7UZI6fk/L7L7so1lCWt35ZSgc=
 github.com/rs/xid v1.5.0 h1:mKX4bl4iPYJtEIxp6CYiUuLQ/8DYMoz0PUdtGgMFRVc=
@@ -159,17 +199,22 @@ github.com/twitchyliquid64/golang-asm v0.15.1 h1:SU5vSMR7hnwNxj24w34ZyCi/FmDZTkS
 github.com/twitchyliquid64/golang-asm v0.15.1/go.mod h1:a1lVb/DtPvCB8fslRZhAngC2+aY1QWCk3Cedj/Gdt08=
 github.com/ugorji/go/codec v1.2.12 h1:9LC83zGrHhuUA9l16C9AHXAqEV/2wBQ4nkvumAE65EE=
 github.com/ugorji/go/codec v1.2.12/go.mod h1:UNopzCgEMSXjBc6AOMqYvWC1ktqTAfzJZUZgYf6w6lg=
+github.com/xo/terminfo v0.0.0-20220910002029-abceb7e1c41e h1:JVG44RsyaB9T2KIHavMF/ppJZNG9ZpyihvCd0w101no=
+github.com/xo/terminfo v0.0.0-20220910002029-abceb7e1c41e/go.mod h1:RbqR21r5mrJuqunuUZ/Dhy/avygyECGrLceyNeo4LiM=
 golang.org/x/arch v0.0.0-20210923205945-b76863e36670/go.mod h1:5om86z9Hs0C8fWVUuoMHwpExlXzs5Tkyp9hOrfG7pp8=
 golang.org/x/arch v0.8.0 h1:3wRIsP3pM4yUptoR96otTUOXI367OS0+c9eeRi9doIc=
 golang.org/x/arch v0.8.0/go.mod h1:FEVrYAQjsQXMVJ1nsMoVVXPZg6p2JE2mx8psSWTDQys=
 golang.org/x/crypto v0.45.0 h1:jMBrvKuj23MTlT0bQEOBcAE0mjg8mK9RXFhRH6nyF3Q=
 golang.org/x/crypto v0.45.0/go.mod h1:XTGrrkGJve7CYK7J8PEww4aY7gM3qMCElcJQ8n8JdX4=
+golang.org/x/exp v0.0.0-20231006140011-7918f672742d h1:jtJma62tbqLibJ5sFQz8bKtEM8rJBtfilJ2qTU199MI=
+golang.org/x/exp v0.0.0-20231006140011-7918f672742d/go.mod h1:ldy0pHrwJyGW56pPQzzkH36rKxoZW1tw7ZJpeKx+hdo=
 golang.org/x/net v0.47.0 h1:Mx+4dIFzqraBXUugkia1OOvlD6LemFo1ALMHjrXDOhY=
 golang.org/x/net v0.47.0/go.mod h1:/jNxtkgq5yWUGYkaZGqo27cfGZ1c5Nen03aYrrKpVRU=
 golang.org/x/oauth2 v0.30.0 h1:dnDm7JmhM45NNpd8FDDeLhK6FwqbOf4MLCM9zb1BOHI=
 golang.org/x/oauth2 v0.30.0/go.mod h1:B++QgG3ZKulg6sRPGD/mqlHQs5rB3Ml9erfeDY7xKlU=
 golang.org/x/sync v0.18.0 h1:kr88TuHDroi+UVf+0hZnirlk8o8T+4MrK6mr60WkH/I=
 golang.org/x/sync v0.18.0/go.mod h1:9KTHXmSnoGruLpwFjVSX0lNNA75CykiMECbovNTZqGI=
+golang.org/x/sys v0.0.0-20210809222454-d867a43fc93e/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20220715151400-c0bba94af5f8/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.38.0 h1:3yZWxaJjBmCWXqhN1qh02AkOnCQ1poK6oF+a7xWL6Gc=

internal/access/config_access/provider.go

@@ -4,19 +4,28 @@ import (
 	"context"
 	"net/http"
 	"strings"
-	"sync"
 
 	sdkaccess "github.com/router-for-me/CLIProxyAPI/v6/sdk/access"
 	sdkconfig "github.com/router-for-me/CLIProxyAPI/v6/sdk/config"
 )
 
-var registerOnce sync.Once
-
 // Register ensures the config-access provider is available to the access manager.
-func Register() {
-	registerOnce.Do(func() {
-		sdkaccess.RegisterProvider(sdkconfig.AccessProviderTypeConfigAPIKey, newProvider)
-	})
+func Register(cfg *sdkconfig.SDKConfig) {
+	if cfg == nil {
+		sdkaccess.UnregisterProvider(sdkaccess.AccessProviderTypeConfigAPIKey)
+		return
+	}
+
+	keys := normalizeKeys(cfg.APIKeys)
+	if len(keys) == 0 {
+		sdkaccess.UnregisterProvider(sdkaccess.AccessProviderTypeConfigAPIKey)
+		return
+	}
+
+	sdkaccess.RegisterProvider(
+		sdkaccess.AccessProviderTypeConfigAPIKey,
+		newProvider(sdkaccess.DefaultAccessProviderName, keys),
+	)
 }
 
 type provider struct {
@@ -24,34 +33,31 @@ type provider struct {
 	keys map[string]struct{}
 }
 
-func newProvider(cfg *sdkconfig.AccessProvider, _ *sdkconfig.SDKConfig) (sdkaccess.Provider, error) {
-	name := cfg.Name
-	if name == "" {
-		name = sdkconfig.DefaultAccessProviderName
+func newProvider(name string, keys []string) *provider {
+	providerName := strings.TrimSpace(name)
+	if providerName == "" {
+		providerName = sdkaccess.DefaultAccessProviderName
 	}
-	keys := make(map[string]struct{}, len(cfg.APIKeys))
-	for _, key := range cfg.APIKeys {
-		if key == "" {
-			continue
-		}
-		keys[key] = struct{}{}
+	keySet := make(map[string]struct{}, len(keys))
+	for _, key := range keys {
+		keySet[key] = struct{}{}
 	}
-	return &provider{name: name, keys: keys}, nil
+	return &provider{name: providerName, keys: keySet}
 }
 
 func (p *provider) Identifier() string {
 	if p == nil || p.name == "" {
-		return sdkconfig.DefaultAccessProviderName
+		return sdkaccess.DefaultAccessProviderName
 	}
 	return p.name
 }
 
-func (p *provider) Authenticate(_ context.Context, r *http.Request) (*sdkaccess.Result, error) {
+func (p *provider) Authenticate(_ context.Context, r *http.Request) (*sdkaccess.Result, *sdkaccess.AuthError) {
 	if p == nil {
-		return nil, sdkaccess.ErrNotHandled
+		return nil, sdkaccess.NewNotHandledError()
 	}
 	if len(p.keys) == 0 {
-		return nil, sdkaccess.ErrNotHandled
+		return nil, sdkaccess.NewNotHandledError()
 	}
 	authHeader := r.Header.Get("Authorization")
 	authHeaderGoogle := r.Header.Get("X-Goog-Api-Key")
@@ -63,7 +69,7 @@ func (p *provider) Authenticate(_ context.Context, r *http.Request) (*sdkaccess.
 		queryAuthToken = r.URL.Query().Get("auth_token")
 	}
 	if authHeader == "" && authHeaderGoogle == "" && authHeaderAnthropic == "" && queryKey == "" && queryAuthToken == "" {
-		return nil, sdkaccess.ErrNoCredentials
+		return nil, sdkaccess.NewNoCredentialsError()
 	}
 
 	apiKey := extractBearerToken(authHeader)
@@ -94,7 +100,7 @@ func (p *provider) Authenticate(_ context.Context, r *http.Request) (*sdkaccess.
 		}
 	}
 
-	return nil, sdkaccess.ErrInvalidCredential
+	return nil, sdkaccess.NewInvalidCredentialError()
 }
 
 func extractBearerToken(header string) string {
@@ -110,3 +116,26 @@ func extractBearerToken(header string) string {
 	}
 	return strings.TrimSpace(parts[1])
 }
+
+func normalizeKeys(keys []string) []string {
+	if len(keys) == 0 {
+		return nil
+	}
+	normalized := make([]string, 0, len(keys))
+	seen := make(map[string]struct{}, len(keys))
+	for _, key := range keys {
+		trimmedKey := strings.TrimSpace(key)
+		if trimmedKey == "" {
+			continue
+		}
+		if _, exists := seen[trimmedKey]; exists {
+			continue
+		}
+		seen[trimmedKey] = struct{}{}
+		normalized = append(normalized, trimmedKey)
+	}
+	if len(normalized) == 0 {
+		return nil
+	}
+	return normalized
+}

internal/access/reconcile.go

@@ -6,9 +6,9 @@ import (
 	"sort"
 	"strings"
 
+	configaccess "github.com/router-for-me/CLIProxyAPI/v6/internal/access/config_access"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
 	sdkaccess "github.com/router-for-me/CLIProxyAPI/v6/sdk/access"
-	sdkConfig "github.com/router-for-me/CLIProxyAPI/v6/sdk/config"
 	log "github.com/sirupsen/logrus"
 )
 
@@ -17,26 +17,26 @@ import (
 // ordered provider slice along with the identifiers of providers that were added, updated, or
 // removed compared to the previous configuration.
 func ReconcileProviders(oldCfg, newCfg *config.Config, existing []sdkaccess.Provider) (result []sdkaccess.Provider, added, updated, removed []string, err error) {
+	_ = oldCfg
 	if newCfg == nil {
 		return nil, nil, nil, nil, nil
 	}
 
+	result = sdkaccess.RegisteredProviders()
+
 	existingMap := make(map[string]sdkaccess.Provider, len(existing))
 	for _, provider := range existing {
-		if provider == nil {
+		providerID := identifierFromProvider(provider)
+		if providerID == "" {
 			continue
 		}
-		existingMap[provider.Identifier()] = provider
+		existingMap[providerID] = provider
 	}
 
-	oldCfgMap := accessProviderMap(oldCfg)
-	newEntries := collectProviderEntries(newCfg)
-
-	result = make([]sdkaccess.Provider, 0, len(newEntries))
-	finalIDs := make(map[string]struct{}, len(newEntries))
+	finalIDs := make(map[string]struct{}, len(result))
 
 	isInlineProvider := func(id string) bool {
-		return strings.EqualFold(id, sdkConfig.DefaultAccessProviderName)
+		return strings.EqualFold(id, sdkaccess.DefaultAccessProviderName)
 	}
 	appendChange := func(list *[]string, id string) {
 		if isInlineProvider(id) {
@@ -45,85 +45,28 @@ func ReconcileProviders(oldCfg, newCfg *config.Config, existing []sdkaccess.Prov
 		*list = append(*list, id)
 	}
 
-	for _, providerCfg := range newEntries {
-		key := providerIdentifier(providerCfg)
-		if key == "" {
+	for _, provider := range result {
+		providerID := identifierFromProvider(provider)
+		if providerID == "" {
 			continue
 		}
+		finalIDs[providerID] = struct{}{}
 
-		forceRebuild := strings.EqualFold(strings.TrimSpace(providerCfg.Type), sdkConfig.AccessProviderTypeConfigAPIKey)
-		if oldCfgProvider, ok := oldCfgMap[key]; ok {
-			isAliased := oldCfgProvider == providerCfg
-			if !forceRebuild && !isAliased && providerConfigEqual(oldCfgProvider, providerCfg) {
-				if existingProvider, okExisting := existingMap[key]; okExisting {
-					result = append(result, existingProvider)
-					finalIDs[key] = struct{}{}
-					continue
-				}
-			}
+		existingProvider, exists := existingMap[providerID]
+		if !exists {
+			appendChange(&added, providerID)
+			continue
 		}
-
-		provider, buildErr := sdkaccess.BuildProvider(providerCfg, &newCfg.SDKConfig)
-		if buildErr != nil {
-			return nil, nil, nil, nil, buildErr
-		}
-		if _, ok := oldCfgMap[key]; ok {
-			if _, existed := existingMap[key]; existed {
-				appendChange(&updated, key)
-			} else {
-				appendChange(&added, key)
-			}
-		} else {
-			appendChange(&added, key)
-		}
-		result = append(result, provider)
-		finalIDs[key] = struct{}{}
-	}
-
-	if len(result) == 0 {
-		if inline := sdkConfig.MakeInlineAPIKeyProvider(newCfg.APIKeys); inline != nil {
-			key := providerIdentifier(inline)
-			if key != "" {
-				if oldCfgProvider, ok := oldCfgMap[key]; ok {
-					if providerConfigEqual(oldCfgProvider, inline) {
-						if existingProvider, okExisting := existingMap[key]; okExisting {
-							result = append(result, existingProvider)
-							finalIDs[key] = struct{}{}
-							goto inlineDone
-						}
-					}
-				}
-				provider, buildErr := sdkaccess.BuildProvider(inline, &newCfg.SDKConfig)
-				if buildErr != nil {
-					return nil, nil, nil, nil, buildErr
-				}
-				if _, existed := existingMap[key]; existed {
-					appendChange(&updated, key)
-				} else if _, hadOld := oldCfgMap[key]; hadOld {
-					appendChange(&updated, key)
-				} else {
-					appendChange(&added, key)
-				}
-				result = append(result, provider)
-				finalIDs[key] = struct{}{}
-			}
-		}
-	inlineDone:
-	}
-
-	removedSet := make(map[string]struct{})
-	for id := range existingMap {
-		if _, ok := finalIDs[id]; !ok {
-			if isInlineProvider(id) {
-				continue
-			}
-			removedSet[id] = struct{}{}
+		if !providerInstanceEqual(existingProvider, provider) {
+			appendChange(&updated, providerID)
 		}
 	}
 
-	removed = make([]string, 0, len(removedSet))
-	for id := range removedSet {
-		removed = append(removed, id)
+	for providerID := range existingMap {
+		if _, exists := finalIDs[providerID]; exists {
+			continue
+		}
+		appendChange(&removed, providerID)
 	}
 
 	sort.Strings(added)
@@ -142,6 +85,7 @@ func ApplyAccessProviders(manager *sdkaccess.Manager, oldCfg, newCfg *config.Con
 	}
 
 	existing := manager.Providers()
+	configaccess.Register(&newCfg.SDKConfig)
 	providers, added, updated, removed, err := ReconcileProviders(oldCfg, newCfg, existing)
 	if err != nil {
 		log.Errorf("failed to reconcile request auth providers: %v", err)
@@ -160,111 +104,24 @@ func ApplyAccessProviders(manager *sdkaccess.Manager, oldCfg, newCfg *config.Con
 	return false, nil
 }
 
-func accessProviderMap(cfg *config.Config) map[string]*sdkConfig.AccessProvider {
-	result := make(map[string]*sdkConfig.AccessProvider)
-	if cfg == nil {
-		return result
-	}
-	for i := range cfg.Access.Providers {
-		providerCfg := &cfg.Access.Providers[i]
-		if providerCfg.Type == "" {
-			continue
-		}
-		key := providerIdentifier(providerCfg)
-		if key == "" {
-			continue
-		}
-		result[key] = providerCfg
-	}
-	if len(result) == 0 && len(cfg.APIKeys) > 0 {
-		if provider := sdkConfig.MakeInlineAPIKeyProvider(cfg.APIKeys); provider != nil {
-			if key := providerIdentifier(provider); key != "" {
-				result[key] = provider
-			}
-		}
-	}
-	return result
-}
-
-func collectProviderEntries(cfg *config.Config) []*sdkConfig.AccessProvider {
-	entries := make([]*sdkConfig.AccessProvider, 0, len(cfg.Access.Providers))
-	for i := range cfg.Access.Providers {
-		providerCfg := &cfg.Access.Providers[i]
-		if providerCfg.Type == "" {
-			continue
-		}
-		if key := providerIdentifier(providerCfg); key != "" {
-			entries = append(entries, providerCfg)
-		}
-	}
-	if len(entries) == 0 && len(cfg.APIKeys) > 0 {
-		if inline := sdkConfig.MakeInlineAPIKeyProvider(cfg.APIKeys); inline != nil {
-			entries = append(entries, inline)
-		}
-	}
-	return entries
-}
-
-func providerIdentifier(provider *sdkConfig.AccessProvider) string {
+func identifierFromProvider(provider sdkaccess.Provider) string {
 	if provider == nil {
 		return ""
 	}
-	if name := strings.TrimSpace(provider.Name); name != "" {
-		return name
-	}
-	typ := strings.TrimSpace(provider.Type)
-	if typ == "" {
-		return ""
-	}
-	if strings.EqualFold(typ, sdkConfig.AccessProviderTypeConfigAPIKey) {
-		return sdkConfig.DefaultAccessProviderName
-	}
-	return typ
+	return strings.TrimSpace(provider.Identifier())
 }
 
-func providerConfigEqual(a, b *sdkConfig.AccessProvider) bool {
+func providerInstanceEqual(a, b sdkaccess.Provider) bool {
 	if a == nil || b == nil {
 		return a == nil && b == nil
 	}
-	if !strings.EqualFold(strings.TrimSpace(a.Type), strings.TrimSpace(b.Type)) {
+	if reflect.TypeOf(a) != reflect.TypeOf(b) {
 		return false
 	}
-	if strings.TrimSpace(a.SDK) != strings.TrimSpace(b.SDK) {
-		return false
+	valueA := reflect.ValueOf(a)
+	valueB := reflect.ValueOf(b)
+	if valueA.Kind() == reflect.Pointer && valueB.Kind() == reflect.Pointer {
+		return valueA.Pointer() == valueB.Pointer()
 	}
-	if !stringSetEqual(a.APIKeys, b.APIKeys) {
-		return false
-	}
-	if len(a.Config) != len(b.Config) {
-		return false
-	}
-	if len(a.Config) > 0 && !reflect.DeepEqual(a.Config, b.Config) {
-		return false
-	}
-	return true
-}
-
-func stringSetEqual(a, b []string) bool {
-	if len(a) != len(b) {
-		return false
-	}
-	if len(a) == 0 {
-		return true
-	}
-	seen := make(map[string]int, len(a))
-	for _, val := range a {
-		seen[val]++
-	}
-	for _, val := range b {
-		count := seen[val]
-		if count == 0 {
-			return false
-		}
-		if count == 1 {
-			delete(seen, val)
-		} else {
-			seen[val] = count - 1
-		}
-	}
-	return len(seen) == 0
+	return reflect.DeepEqual(a, b)
 }

internal/api/handlers/management/auth_files.go

@@ -25,6 +25,7 @@ import (
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/auth/codex"
 	geminiAuth "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/gemini"
 	iflowauth "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/iflow"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/auth/kimi"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/auth/qwen"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/interfaces"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/misc"
@@ -807,6 +808,87 @@ func (h *Handler) PatchAuthFileStatus(c *gin.Context) {
 	c.JSON(http.StatusOK, gin.H{"status": "ok", "disabled": *req.Disabled})
 }
 
+// PatchAuthFileFields updates editable fields (prefix, proxy_url, priority) of an auth file.
+func (h *Handler) PatchAuthFileFields(c *gin.Context) {
+	if h.authManager == nil {
+		c.JSON(http.StatusServiceUnavailable, gin.H{"error": "core auth manager unavailable"})
+		return
+	}
+
+	var req struct {
+		Name     string  `json:"name"`
+		Prefix   *string `json:"prefix"`
+		ProxyURL *string `json:"proxy_url"`
+		Priority *int    `json:"priority"`
+	}
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid request body"})
+		return
+	}
+
+	name := strings.TrimSpace(req.Name)
+	if name == "" {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "name is required"})
+		return
+	}
+
+	ctx := c.Request.Context()
+
+	// Find auth by name or ID
+	var targetAuth *coreauth.Auth
+	if auth, ok := h.authManager.GetByID(name); ok {
+		targetAuth = auth
+	} else {
+		auths := h.authManager.List()
+		for _, auth := range auths {
+			if auth.FileName == name {
+				targetAuth = auth
+				break
+			}
+		}
+	}
+
+	if targetAuth == nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "auth file not found"})
+		return
+	}
+
+	changed := false
+	if req.Prefix != nil {
+		targetAuth.Prefix = *req.Prefix
+		changed = true
+	}
+	if req.ProxyURL != nil {
+		targetAuth.ProxyURL = *req.ProxyURL
+		changed = true
+	}
+	if req.Priority != nil {
+		if targetAuth.Metadata == nil {
+			targetAuth.Metadata = make(map[string]any)
+		}
+		if *req.Priority == 0 {
+			delete(targetAuth.Metadata, "priority")
+		} else {
+			targetAuth.Metadata["priority"] = *req.Priority
+		}
+		changed = true
+	}
+
+	if !changed {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "no fields to update"})
+		return
+	}
+
+	targetAuth.UpdatedAt = time.Now()
+
+	if _, err := h.authManager.Update(ctx, targetAuth); err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": fmt.Sprintf("failed to update auth: %v", err)})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{"status": "ok"})
+}
+
 func (h *Handler) disableAuth(ctx context.Context, id string) {
 	if h == nil || h.authManager == nil {
 		return
@@ -863,11 +945,17 @@ func (h *Handler) saveTokenRecord(ctx context.Context, record *coreauth.Auth) (s
 	if store == nil {
 		return "", fmt.Errorf("token store unavailable")
 	}
+	if h.postAuthHook != nil {
+		if err := h.postAuthHook(ctx, record); err != nil {
+			return "", fmt.Errorf("post-auth hook failed: %w", err)
+		}
+	}
 	return store.Save(ctx, record)
 }
 
 func (h *Handler) RequestAnthropicToken(c *gin.Context) {
 	ctx := context.Background()
+	ctx = PopulateAuthContext(ctx, c)
 
 	fmt.Println("Initializing Claude authentication...")
 
@@ -1012,6 +1100,7 @@ func (h *Handler) RequestAnthropicToken(c *gin.Context) {
 
 func (h *Handler) RequestGeminiCLIToken(c *gin.Context) {
 	ctx := context.Background()
+	ctx = PopulateAuthContext(ctx, c)
 	proxyHTTPClient := util.SetProxy(&h.cfg.SDKConfig, &http.Client{})
 	ctx = context.WithValue(ctx, oauth2.HTTPClient, proxyHTTPClient)
 
@@ -1187,6 +1276,30 @@ func (h *Handler) RequestGeminiCLIToken(c *gin.Context) {
 			}
 			ts.ProjectID = strings.Join(projects, ",")
 			ts.Checked = true
+		} else if strings.EqualFold(requestedProjectID, "GOOGLE_ONE") {
+			ts.Auto = false
+			if errSetup := performGeminiCLISetup(ctx, gemClient, &ts, ""); errSetup != nil {
+				log.Errorf("Google One auto-discovery failed: %v", errSetup)
+				SetOAuthSessionError(state, "Google One auto-discovery failed")
+				return
+			}
+			if strings.TrimSpace(ts.ProjectID) == "" {
+				log.Error("Google One auto-discovery returned empty project ID")
+				SetOAuthSessionError(state, "Google One auto-discovery returned empty project ID")
+				return
+			}
+			isChecked, errCheck := checkCloudAPIIsEnabled(ctx, gemClient, ts.ProjectID)
+			if errCheck != nil {
+				log.Errorf("Failed to verify Cloud AI API status: %v", errCheck)
+				SetOAuthSessionError(state, "Failed to verify Cloud AI API status")
+				return
+			}
+			ts.Checked = isChecked
+			if !isChecked {
+				log.Error("Cloud AI API is not enabled for the auto-discovered project")
+				SetOAuthSessionError(state, "Cloud AI API not enabled")
+				return
+			}
 		} else {
 			if errEnsure := ensureGeminiProjectAndOnboard(ctx, gemClient, &ts, requestedProjectID); errEnsure != nil {
 				log.Errorf("Failed to complete Gemini CLI onboarding: %v", errEnsure)
@@ -1246,6 +1359,7 @@ func (h *Handler) RequestGeminiCLIToken(c *gin.Context) {
 
 func (h *Handler) RequestCodexToken(c *gin.Context) {
 	ctx := context.Background()
+	ctx = PopulateAuthContext(ctx, c)
 
 	fmt.Println("Initializing Codex authentication...")
 
@@ -1391,6 +1505,7 @@ func (h *Handler) RequestCodexToken(c *gin.Context) {
 
 func (h *Handler) RequestAntigravityToken(c *gin.Context) {
 	ctx := context.Background()
+	ctx = PopulateAuthContext(ctx, c)
 
 	fmt.Println("Initializing Antigravity authentication...")
 
@@ -1555,6 +1670,7 @@ func (h *Handler) RequestAntigravityToken(c *gin.Context) {
 
 func (h *Handler) RequestQwenToken(c *gin.Context) {
 	ctx := context.Background()
+	ctx = PopulateAuthContext(ctx, c)
 
 	fmt.Println("Initializing Qwen authentication...")
 
@@ -1608,8 +1724,86 @@ func (h *Handler) RequestQwenToken(c *gin.Context) {
 	c.JSON(200, gin.H{"status": "ok", "url": authURL, "state": state})
 }
 
+func (h *Handler) RequestKimiToken(c *gin.Context) {
+	ctx := context.Background()
+	ctx = PopulateAuthContext(ctx, c)
+
+	fmt.Println("Initializing Kimi authentication...")
+
+	state := fmt.Sprintf("kmi-%d", time.Now().UnixNano())
+	// Initialize Kimi auth service
+	kimiAuth := kimi.NewKimiAuth(h.cfg)
+
+	// Generate authorization URL
+	deviceFlow, errStartDeviceFlow := kimiAuth.StartDeviceFlow(ctx)
+	if errStartDeviceFlow != nil {
+		log.Errorf("Failed to generate authorization URL: %v", errStartDeviceFlow)
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to generate authorization url"})
+		return
+	}
+	authURL := deviceFlow.VerificationURIComplete
+	if authURL == "" {
+		authURL = deviceFlow.VerificationURI
+	}
+
+	RegisterOAuthSession(state, "kimi")
+
+	go func() {
+		fmt.Println("Waiting for authentication...")
+		authBundle, errWaitForAuthorization := kimiAuth.WaitForAuthorization(ctx, deviceFlow)
+		if errWaitForAuthorization != nil {
+			SetOAuthSessionError(state, "Authentication failed")
+			fmt.Printf("Authentication failed: %v\n", errWaitForAuthorization)
+			return
+		}
+
+		// Create token storage
+		tokenStorage := kimiAuth.CreateTokenStorage(authBundle)
+
+		metadata := map[string]any{
+			"type":          "kimi",
+			"access_token":  authBundle.TokenData.AccessToken,
+			"refresh_token": authBundle.TokenData.RefreshToken,
+			"token_type":    authBundle.TokenData.TokenType,
+			"scope":         authBundle.TokenData.Scope,
+			"timestamp":     time.Now().UnixMilli(),
+		}
+		if authBundle.TokenData.ExpiresAt > 0 {
+			expired := time.Unix(authBundle.TokenData.ExpiresAt, 0).UTC().Format(time.RFC3339)
+			metadata["expired"] = expired
+		}
+		if strings.TrimSpace(authBundle.DeviceID) != "" {
+			metadata["device_id"] = strings.TrimSpace(authBundle.DeviceID)
+		}
+
+		fileName := fmt.Sprintf("kimi-%d.json", time.Now().UnixMilli())
+		record := &coreauth.Auth{
+			ID:       fileName,
+			Provider: "kimi",
+			FileName: fileName,
+			Label:    "Kimi User",
+			Storage:  tokenStorage,
+			Metadata: metadata,
+		}
+		savedPath, errSave := h.saveTokenRecord(ctx, record)
+		if errSave != nil {
+			log.Errorf("Failed to save authentication tokens: %v", errSave)
+			SetOAuthSessionError(state, "Failed to save authentication tokens")
+			return
+		}
+
+		fmt.Printf("Authentication successful! Token saved to %s\n", savedPath)
+		fmt.Println("You can now use Kimi services through this CLI")
+		CompleteOAuthSession(state)
+		CompleteOAuthSessionsByProvider("kimi")
+	}()
+
+	c.JSON(200, gin.H{"status": "ok", "url": authURL, "state": state})
+}
+
 func (h *Handler) RequestIFlowToken(c *gin.Context) {
 	ctx := context.Background()
+	ctx = PopulateAuthContext(ctx, c)
 
 	fmt.Println("Initializing iFlow authentication...")
 
@@ -1959,7 +2153,48 @@ func performGeminiCLISetup(ctx context.Context, httpClient *http.Client, storage
 		}
 	}
 	if projectID == "" {
-		return &projectSelectionRequiredError{}
+		// Auto-discovery: try onboardUser without specifying a project
+		// to let Google auto-provision one (matches Gemini CLI headless behavior
+		// and Antigravity's FetchProjectID pattern).
+		autoOnboardReq := map[string]any{
+			"tierId":   tierID,
+			"metadata": metadata,
+		}
+
+		autoCtx, autoCancel := context.WithTimeout(ctx, 30*time.Second)
+		defer autoCancel()
+		for attempt := 1; ; attempt++ {
+			var onboardResp map[string]any
+			if errOnboard := callGeminiCLI(autoCtx, httpClient, "onboardUser", autoOnboardReq, &onboardResp); errOnboard != nil {
+				return fmt.Errorf("auto-discovery onboardUser: %w", errOnboard)
+			}
+
+			if done, okDone := onboardResp["done"].(bool); okDone && done {
+				if resp, okResp := onboardResp["response"].(map[string]any); okResp {
+					switch v := resp["cloudaicompanionProject"].(type) {
+					case string:
+						projectID = strings.TrimSpace(v)
+					case map[string]any:
+						if id, okID := v["id"].(string); okID {
+							projectID = strings.TrimSpace(id)
+						}
+					}
+				}
+				break
+			}
+
+			log.Debugf("Auto-discovery: onboarding in progress, attempt %d...", attempt)
+			select {
+			case <-autoCtx.Done():
+				return &projectSelectionRequiredError{}
+			case <-time.After(2 * time.Second):
+			}
+		}
+
+		if projectID == "" {
+			return &projectSelectionRequiredError{}
+		}
+		log.Infof("Auto-discovered project ID via onboarding: %s", projectID)
 	}
 
 	onboardReqBody := map[string]any{
@@ -2189,3 +2424,12 @@ func (h *Handler) GetAuthStatus(c *gin.Context) {
 	}
 	c.JSON(http.StatusOK, gin.H{"status": "wait"})
 }
+
+// PopulateAuthContext extracts request info and adds it to the context
+func PopulateAuthContext(ctx context.Context, c *gin.Context) context.Context {
+	info := &coreauth.RequestInfo{
+		Query:   c.Request.URL.Query(),
+		Headers: c.Request.Header,
+	}
+	return coreauth.WithRequestInfo(ctx, info)
+}

internal/api/handlers/management/config_basic.go

@@ -28,8 +28,7 @@ func (h *Handler) GetConfig(c *gin.Context) {
 		c.JSON(200, gin.H{})
 		return
 	}
-	cfgCopy := *h.cfg
-	c.JSON(200, &cfgCopy)
+	c.JSON(200, new(*h.cfg))
 }
 
 type releaseInfo struct {

internal/api/handlers/management/config_lists.go

@@ -109,14 +109,13 @@ func (h *Handler) GetAPIKeys(c *gin.Context) { c.JSON(200, gin.H{"api-keys": h.c
 func (h *Handler) PutAPIKeys(c *gin.Context) {
 	h.putStringList(c, func(v []string) {
 		h.cfg.APIKeys = append([]string(nil), v...)
-		h.cfg.Access.Providers = nil
 	}, nil)
 }
 func (h *Handler) PatchAPIKeys(c *gin.Context) {
-	h.patchStringList(c, &h.cfg.APIKeys, func() { h.cfg.Access.Providers = nil })
+	h.patchStringList(c, &h.cfg.APIKeys, func() {})
 }
 func (h *Handler) DeleteAPIKeys(c *gin.Context) {
-	h.deleteFromStringList(c, &h.cfg.APIKeys, func() { h.cfg.Access.Providers = nil })
+	h.deleteFromStringList(c, &h.cfg.APIKeys, func() {})
 }
 
 // gemini-api-key: []GeminiKey

internal/api/handlers/management/handler.go

@@ -47,6 +47,7 @@ type Handler struct {
 	allowRemoteOverride bool
 	envSecret           string
 	logDir              string
+	postAuthHook        coreauth.PostAuthHook
 }
 
 // NewHandler creates a new management handler instance.
@@ -128,6 +129,11 @@ func (h *Handler) SetLogDirectory(dir string) {
 	h.logDir = dir
 }
 
+// SetPostAuthHook registers a hook to be called after auth record creation but before persistence.
+func (h *Handler) SetPostAuthHook(hook coreauth.PostAuthHook) {
+	h.postAuthHook = hook
+}
+
 // Middleware enforces access control for management endpoints.
 // All requests (local and remote) require a valid management key.
 // Additionally, remote access requires allow-remote-management=true.

internal/api/middleware/request_logging.go

@@ -15,10 +15,12 @@ import (
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 )
 
+const maxErrorOnlyCapturedRequestBodyBytes int64 = 1 << 20 // 1 MiB
+
 // 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. When logging is disabled in the
-// logger, it still captures data so that upstream errors can be persisted.
+// and uses the provided RequestLogger to record this data. When full request logging is disabled,
+// body capture is limited to small known-size payloads to avoid large per-request memory spikes.
 func RequestLoggingMiddleware(logger logging.RequestLogger) gin.HandlerFunc {
 	return func(c *gin.Context) {
 		if logger == nil {
@@ -26,7 +28,7 @@ func RequestLoggingMiddleware(logger logging.RequestLogger) gin.HandlerFunc {
 			return
 		}
 
-		if c.Request.Method == http.MethodGet {
+		if shouldSkipMethodForRequestLogging(c.Request) {
 			c.Next()
 			return
 		}
@@ -37,8 +39,10 @@ func RequestLoggingMiddleware(logger logging.RequestLogger) gin.HandlerFunc {
 			return
 		}
 
+		loggerEnabled := logger.IsEnabled()
+
 		// Capture request information
-		requestInfo, err := captureRequestInfo(c)
+		requestInfo, err := captureRequestInfo(c, shouldCaptureRequestBody(loggerEnabled, c.Request))
 		if err != nil {
 			// Log error but continue processing
 			// In a real implementation, you might want to use a proper logger here
@@ -48,7 +52,7 @@ func RequestLoggingMiddleware(logger logging.RequestLogger) gin.HandlerFunc {
 
 		// Create response writer wrapper
 		wrapper := NewResponseWriterWrapper(c.Writer, logger, requestInfo)
-		if !logger.IsEnabled() {
+		if !loggerEnabled {
 			wrapper.logOnErrorOnly = true
 		}
 		c.Writer = wrapper
@@ -64,10 +68,47 @@ func RequestLoggingMiddleware(logger logging.RequestLogger) gin.HandlerFunc {
 	}
 }
 
+func shouldSkipMethodForRequestLogging(req *http.Request) bool {
+	if req == nil {
+		return true
+	}
+	if req.Method != http.MethodGet {
+		return false
+	}
+	return !isResponsesWebsocketUpgrade(req)
+}
+
+func isResponsesWebsocketUpgrade(req *http.Request) bool {
+	if req == nil || req.URL == nil {
+		return false
+	}
+	if req.URL.Path != "/v1/responses" {
+		return false
+	}
+	return strings.EqualFold(strings.TrimSpace(req.Header.Get("Upgrade")), "websocket")
+}
+
+func shouldCaptureRequestBody(loggerEnabled bool, req *http.Request) bool {
+	if loggerEnabled {
+		return true
+	}
+	if req == nil || req.Body == nil {
+		return false
+	}
+	contentType := strings.ToLower(strings.TrimSpace(req.Header.Get("Content-Type")))
+	if strings.HasPrefix(contentType, "multipart/form-data") {
+		return false
+	}
+	if req.ContentLength <= 0 {
+		return false
+	}
+	return req.ContentLength <= maxErrorOnlyCapturedRequestBodyBytes
+}
+
 // captureRequestInfo extracts relevant information from the incoming HTTP request.
 // It captures the URL, method, headers, and body. The request body is read and then
 // restored so that it can be processed by subsequent handlers.
-func captureRequestInfo(c *gin.Context) (*RequestInfo, error) {
+func captureRequestInfo(c *gin.Context, captureBody bool) (*RequestInfo, error) {
 	// Capture URL with sensitive query parameters masked
 	maskedQuery := util.MaskSensitiveQuery(c.Request.URL.RawQuery)
 	url := c.Request.URL.Path
@@ -86,7 +127,7 @@ func captureRequestInfo(c *gin.Context) (*RequestInfo, error) {
 
 	// Capture request body
 	var body []byte
-	if c.Request.Body != nil {
+	if captureBody && c.Request.Body != nil {
 		// Read the body
 		bodyBytes, err := io.ReadAll(c.Request.Body)
 		if err != nil {

internal/api/middleware/request_logging_test.go

@@ -0,0 +1,138 @@
+package middleware
+
+import (
+	"io"
+	"net/http"
+	"net/url"
+	"strings"
+	"testing"
+)
+
+func TestShouldSkipMethodForRequestLogging(t *testing.T) {
+	tests := []struct {
+		name string
+		req  *http.Request
+		skip bool
+	}{
+		{
+			name: "nil request",
+			req:  nil,
+			skip: true,
+		},
+		{
+			name: "post request should not skip",
+			req: &http.Request{
+				Method: http.MethodPost,
+				URL:    &url.URL{Path: "/v1/responses"},
+			},
+			skip: false,
+		},
+		{
+			name: "plain get should skip",
+			req: &http.Request{
+				Method: http.MethodGet,
+				URL:    &url.URL{Path: "/v1/models"},
+				Header: http.Header{},
+			},
+			skip: true,
+		},
+		{
+			name: "responses websocket upgrade should not skip",
+			req: &http.Request{
+				Method: http.MethodGet,
+				URL:    &url.URL{Path: "/v1/responses"},
+				Header: http.Header{"Upgrade": []string{"websocket"}},
+			},
+			skip: false,
+		},
+		{
+			name: "responses get without upgrade should skip",
+			req: &http.Request{
+				Method: http.MethodGet,
+				URL:    &url.URL{Path: "/v1/responses"},
+				Header: http.Header{},
+			},
+			skip: true,
+		},
+	}
+
+	for i := range tests {
+		got := shouldSkipMethodForRequestLogging(tests[i].req)
+		if got != tests[i].skip {
+			t.Fatalf("%s: got skip=%t, want %t", tests[i].name, got, tests[i].skip)
+		}
+	}
+}
+
+func TestShouldCaptureRequestBody(t *testing.T) {
+	tests := []struct {
+		name          string
+		loggerEnabled bool
+		req           *http.Request
+		want          bool
+	}{
+		{
+			name:          "logger enabled always captures",
+			loggerEnabled: true,
+			req: &http.Request{
+				Body:          io.NopCloser(strings.NewReader("{}")),
+				ContentLength: -1,
+				Header:        http.Header{"Content-Type": []string{"application/json"}},
+			},
+			want: true,
+		},
+		{
+			name:          "nil request",
+			loggerEnabled: false,
+			req:           nil,
+			want:          false,
+		},
+		{
+			name:          "small known size json in error-only mode",
+			loggerEnabled: false,
+			req: &http.Request{
+				Body:          io.NopCloser(strings.NewReader("{}")),
+				ContentLength: 2,
+				Header:        http.Header{"Content-Type": []string{"application/json"}},
+			},
+			want: true,
+		},
+		{
+			name:          "large known size skipped in error-only mode",
+			loggerEnabled: false,
+			req: &http.Request{
+				Body:          io.NopCloser(strings.NewReader("x")),
+				ContentLength: maxErrorOnlyCapturedRequestBodyBytes + 1,
+				Header:        http.Header{"Content-Type": []string{"application/json"}},
+			},
+			want: false,
+		},
+		{
+			name:          "unknown size skipped in error-only mode",
+			loggerEnabled: false,
+			req: &http.Request{
+				Body:          io.NopCloser(strings.NewReader("x")),
+				ContentLength: -1,
+				Header:        http.Header{"Content-Type": []string{"application/json"}},
+			},
+			want: false,
+		},
+		{
+			name:          "multipart skipped in error-only mode",
+			loggerEnabled: false,
+			req: &http.Request{
+				Body:          io.NopCloser(strings.NewReader("x")),
+				ContentLength: 1,
+				Header:        http.Header{"Content-Type": []string{"multipart/form-data; boundary=abc"}},
+			},
+			want: false,
+		},
+	}
+
+	for i := range tests {
+		got := shouldCaptureRequestBody(tests[i].loggerEnabled, tests[i].req)
+		if got != tests[i].want {
+			t.Fatalf("%s: got %t, want %t", tests[i].name, got, tests[i].want)
+		}
+	}
+}

internal/api/middleware/response_writer.go

@@ -14,6 +14,8 @@ import (
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/logging"
 )
 
+const requestBodyOverrideContextKey = "REQUEST_BODY_OVERRIDE"
+
 // RequestInfo holds essential details of an incoming HTTP request for logging purposes.
 type RequestInfo struct {
 	URL       string              // URL is the request URL.
@@ -223,8 +225,8 @@ func (w *ResponseWriterWrapper) detectStreaming(contentType string) bool {
 
 	// Only fall back to request payload hints when Content-Type is not set yet.
 	if w.requestInfo != nil && len(w.requestInfo.Body) > 0 {
-		bodyStr := string(w.requestInfo.Body)
-		return strings.Contains(bodyStr, `"stream": true`) || strings.Contains(bodyStr, `"stream":true`)
+		return bytes.Contains(w.requestInfo.Body, []byte(`"stream": true`)) ||
+			bytes.Contains(w.requestInfo.Body, []byte(`"stream":true`))
 	}
 
 	return false
@@ -310,7 +312,7 @@ func (w *ResponseWriterWrapper) Finalize(c *gin.Context) error {
 		return nil
 	}
 
-	return w.logRequest(finalStatusCode, w.cloneHeaders(), w.body.Bytes(), w.extractAPIRequest(c), w.extractAPIResponse(c), w.extractAPIResponseTimestamp(c), slicesAPIResponseError, forceLog)
+	return w.logRequest(w.extractRequestBody(c), finalStatusCode, w.cloneHeaders(), w.body.Bytes(), w.extractAPIRequest(c), w.extractAPIResponse(c), w.extractAPIResponseTimestamp(c), slicesAPIResponseError, forceLog)
 }
 
 func (w *ResponseWriterWrapper) cloneHeaders() map[string][]string {
@@ -361,16 +363,32 @@ func (w *ResponseWriterWrapper) extractAPIResponseTimestamp(c *gin.Context) time
 	return time.Time{}
 }
 
-func (w *ResponseWriterWrapper) logRequest(statusCode int, headers map[string][]string, body []byte, apiRequestBody, apiResponseBody []byte, apiResponseTimestamp time.Time, apiResponseErrors []*interfaces.ErrorMessage, forceLog bool) error {
+func (w *ResponseWriterWrapper) extractRequestBody(c *gin.Context) []byte {
+	if c != nil {
+		if bodyOverride, isExist := c.Get(requestBodyOverrideContextKey); isExist {
+			switch value := bodyOverride.(type) {
+			case []byte:
+				if len(value) > 0 {
+					return bytes.Clone(value)
+				}
+			case string:
+				if strings.TrimSpace(value) != "" {
+					return []byte(value)
+				}
+			}
+		}
+	}
+	if w.requestInfo != nil && len(w.requestInfo.Body) > 0 {
+		return w.requestInfo.Body
+	}
+	return nil
+}
+
+func (w *ResponseWriterWrapper) logRequest(requestBody []byte, statusCode int, headers map[string][]string, body []byte, apiRequestBody, apiResponseBody []byte, apiResponseTimestamp time.Time, 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, string, time.Time, time.Time) error
 	}); ok {

internal/api/middleware/response_writer_test.go

@@ -0,0 +1,43 @@
+package middleware
+
+import (
+	"net/http/httptest"
+	"testing"
+
+	"github.com/gin-gonic/gin"
+)
+
+func TestExtractRequestBodyPrefersOverride(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	recorder := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(recorder)
+
+	wrapper := &ResponseWriterWrapper{
+		requestInfo: &RequestInfo{Body: []byte("original-body")},
+	}
+
+	body := wrapper.extractRequestBody(c)
+	if string(body) != "original-body" {
+		t.Fatalf("request body = %q, want %q", string(body), "original-body")
+	}
+
+	c.Set(requestBodyOverrideContextKey, []byte("override-body"))
+	body = wrapper.extractRequestBody(c)
+	if string(body) != "override-body" {
+		t.Fatalf("request body = %q, want %q", string(body), "override-body")
+	}
+}
+
+func TestExtractRequestBodySupportsStringOverride(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	recorder := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(recorder)
+
+	wrapper := &ResponseWriterWrapper{}
+	c.Set(requestBodyOverrideContextKey, "override-as-string")
+
+	body := wrapper.extractRequestBody(c)
+	if string(body) != "override-as-string" {
+		t.Fatalf("request body = %q, want %q", string(body), "override-as-string")
+	}
+}

internal/api/modules/amp/amp.go

@@ -127,8 +127,7 @@ func (m *AmpModule) Register(ctx modules.Context) error {
 		m.modelMapper = NewModelMapper(settings.ModelMappings)
 
 		// Store initial config for partial reload comparison
-		settingsCopy := settings
-		m.lastConfig = &settingsCopy
+		m.lastConfig = new(settings)
 
 		// Initialize localhost restriction setting (hot-reloadable)
 		m.setRestrictToLocalhost(settings.RestrictManagementToLocalhost)

internal/api/modules/amp/proxy.go

@@ -3,6 +3,8 @@ package amp
 import (
 	"bytes"
 	"compress/gzip"
+	"context"
+	"errors"
 	"fmt"
 	"io"
 	"net/http"
@@ -188,6 +190,10 @@ func createReverseProxy(upstreamURL string, secretSource SecretSource) (*httputi
 
 	// Error handler for proxy failures
 	proxy.ErrorHandler = func(rw http.ResponseWriter, req *http.Request, err error) {
+		// Client-side cancellations are common during polling; suppress logging in this case
+		if errors.Is(err, context.Canceled) {
+			return
+		}
 		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)

internal/api/modules/amp/proxy_test.go

@@ -493,6 +493,30 @@ func TestReverseProxy_ErrorHandler(t *testing.T) {
 	}
 }
 
+func TestReverseProxy_ErrorHandler_ContextCanceled(t *testing.T) {
+	// Test that context.Canceled errors return 499 without generic error response
+	proxy, err := createReverseProxy("http://example.com", NewStaticSecretSource(""))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// Create a canceled context to trigger the cancellation path
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel() // Cancel immediately
+
+	req := httptest.NewRequest(http.MethodGet, "/test", nil).WithContext(ctx)
+	rr := httptest.NewRecorder()
+
+	// Directly invoke the ErrorHandler with context.Canceled
+	proxy.ErrorHandler(rr, req, context.Canceled)
+
+	// Body should be empty for canceled requests (no JSON error response)
+	body := rr.Body.Bytes()
+	if len(body) > 0 {
+		t.Fatalf("expected empty body for canceled context, got: %s", body)
+	}
+}
+
 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) {

internal/api/modules/amp/response_rewriter.go

@@ -66,7 +66,7 @@ func (rw *ResponseRewriter) Flush() {
 }
 
 // modelFieldPaths lists all JSON paths where model name may appear
-var modelFieldPaths = []string{"model", "modelVersion", "response.modelVersion", "message.model"}
+var modelFieldPaths = []string{"message.model", "model", "modelVersion", "response.model", "response.modelVersion"}
 
 // rewriteModelInResponse replaces all occurrences of the mapped model with the original model in JSON
 // It also suppresses "thinking" blocks if "tool_use" is present to ensure Amp client compatibility

internal/api/modules/amp/response_rewriter_test.go

@@ -0,0 +1,110 @@
+package amp
+
+import (
+	"testing"
+)
+
+func TestRewriteModelInResponse_TopLevel(t *testing.T) {
+	rw := &ResponseRewriter{originalModel: "gpt-5.2-codex"}
+
+	input := []byte(`{"id":"resp_1","model":"gpt-5.3-codex","output":[]}`)
+	result := rw.rewriteModelInResponse(input)
+
+	expected := `{"id":"resp_1","model":"gpt-5.2-codex","output":[]}`
+	if string(result) != expected {
+		t.Errorf("expected %s, got %s", expected, string(result))
+	}
+}
+
+func TestRewriteModelInResponse_ResponseModel(t *testing.T) {
+	rw := &ResponseRewriter{originalModel: "gpt-5.2-codex"}
+
+	input := []byte(`{"type":"response.completed","response":{"id":"resp_1","model":"gpt-5.3-codex","status":"completed"}}`)
+	result := rw.rewriteModelInResponse(input)
+
+	expected := `{"type":"response.completed","response":{"id":"resp_1","model":"gpt-5.2-codex","status":"completed"}}`
+	if string(result) != expected {
+		t.Errorf("expected %s, got %s", expected, string(result))
+	}
+}
+
+func TestRewriteModelInResponse_ResponseCreated(t *testing.T) {
+	rw := &ResponseRewriter{originalModel: "gpt-5.2-codex"}
+
+	input := []byte(`{"type":"response.created","response":{"id":"resp_1","model":"gpt-5.3-codex","status":"in_progress"}}`)
+	result := rw.rewriteModelInResponse(input)
+
+	expected := `{"type":"response.created","response":{"id":"resp_1","model":"gpt-5.2-codex","status":"in_progress"}}`
+	if string(result) != expected {
+		t.Errorf("expected %s, got %s", expected, string(result))
+	}
+}
+
+func TestRewriteModelInResponse_NoModelField(t *testing.T) {
+	rw := &ResponseRewriter{originalModel: "gpt-5.2-codex"}
+
+	input := []byte(`{"type":"response.output_item.added","item":{"id":"item_1","type":"message"}}`)
+	result := rw.rewriteModelInResponse(input)
+
+	if string(result) != string(input) {
+		t.Errorf("expected no modification, got %s", string(result))
+	}
+}
+
+func TestRewriteModelInResponse_EmptyOriginalModel(t *testing.T) {
+	rw := &ResponseRewriter{originalModel: ""}
+
+	input := []byte(`{"model":"gpt-5.3-codex"}`)
+	result := rw.rewriteModelInResponse(input)
+
+	if string(result) != string(input) {
+		t.Errorf("expected no modification when originalModel is empty, got %s", string(result))
+	}
+}
+
+func TestRewriteStreamChunk_SSEWithResponseModel(t *testing.T) {
+	rw := &ResponseRewriter{originalModel: "gpt-5.2-codex"}
+
+	chunk := []byte("data: {\"type\":\"response.completed\",\"response\":{\"id\":\"resp_1\",\"model\":\"gpt-5.3-codex\",\"status\":\"completed\"}}\n\n")
+	result := rw.rewriteStreamChunk(chunk)
+
+	expected := "data: {\"type\":\"response.completed\",\"response\":{\"id\":\"resp_1\",\"model\":\"gpt-5.2-codex\",\"status\":\"completed\"}}\n\n"
+	if string(result) != expected {
+		t.Errorf("expected %s, got %s", expected, string(result))
+	}
+}
+
+func TestRewriteStreamChunk_MultipleEvents(t *testing.T) {
+	rw := &ResponseRewriter{originalModel: "gpt-5.2-codex"}
+
+	chunk := []byte("data: {\"type\":\"response.created\",\"response\":{\"model\":\"gpt-5.3-codex\"}}\n\ndata: {\"type\":\"response.output_item.added\",\"item\":{\"id\":\"item_1\"}}\n\n")
+	result := rw.rewriteStreamChunk(chunk)
+
+	if string(result) == string(chunk) {
+		t.Error("expected response.model to be rewritten in SSE stream")
+	}
+	if !contains(result, []byte(`"model":"gpt-5.2-codex"`)) {
+		t.Errorf("expected rewritten model in output, got %s", string(result))
+	}
+}
+
+func TestRewriteStreamChunk_MessageModel(t *testing.T) {
+	rw := &ResponseRewriter{originalModel: "claude-opus-4.5"}
+
+	chunk := []byte("data: {\"message\":{\"model\":\"claude-sonnet-4\",\"role\":\"assistant\"}}\n\n")
+	result := rw.rewriteStreamChunk(chunk)
+
+	expected := "data: {\"message\":{\"model\":\"claude-opus-4.5\",\"role\":\"assistant\"}}\n\n"
+	if string(result) != expected {
+		t.Errorf("expected %s, got %s", expected, string(result))
+	}
+}
+
+func contains(data, substr []byte) bool {
+	for i := 0; i <= len(data)-len(substr); i++ {
+		if string(data[i:i+len(substr)]) == string(substr) {
+			return true
+		}
+	}
+	return false
+}

internal/api/server.go

@@ -51,6 +51,7 @@ type serverOptionConfig struct {
 	keepAliveEnabled     bool
 	keepAliveTimeout     time.Duration
 	keepAliveOnTimeout   func()
+	postAuthHook         auth.PostAuthHook
 }
 
 // ServerOption customises HTTP server construction.
@@ -111,6 +112,13 @@ func WithRequestLoggerFactory(factory func(*config.Config, string) logging.Reque
 	}
 }
 
+// WithPostAuthHook registers a hook to be called after auth record creation.
+func WithPostAuthHook(hook auth.PostAuthHook) ServerOption {
+	return func(cfg *serverOptionConfig) {
+		cfg.postAuthHook = hook
+	}
+}
+
 // Server represents the main API server.
 // It encapsulates the Gin engine, HTTP server, handlers, and configuration.
 type Server struct {
@@ -262,6 +270,9 @@ func NewServer(cfg *config.Config, authManager *auth.Manager, accessManager *sdk
 	}
 	logDir := logging.ResolveLogDirectory(cfg)
 	s.mgmt.SetLogDirectory(logDir)
+	if optionState.postAuthHook != nil {
+		s.mgmt.SetPostAuthHook(optionState.postAuthHook)
+	}
 	s.localPassword = optionState.localPassword
 
 	// Setup routes
@@ -284,8 +295,9 @@ func NewServer(cfg *config.Config, authManager *auth.Manager, accessManager *sdk
 		optionState.routerConfigurator(engine, s.handlers, cfg)
 	}
 
-	// Register management routes when configuration or environment secrets are available.
-	hasManagementSecret := cfg.RemoteManagement.SecretKey != "" || envManagementSecret
+	// Register management routes when configuration or environment secrets are available,
+	// or when a local management password is provided (e.g. TUI mode).
+	hasManagementSecret := cfg.RemoteManagement.SecretKey != "" || envManagementSecret || s.localPassword != ""
 	s.managementRoutesEnabled.Store(hasManagementSecret)
 	if hasManagementSecret {
 		s.registerManagementRoutes()
@@ -323,6 +335,7 @@ func (s *Server) setupRoutes() {
 		v1.POST("/completions", openaiHandlers.Completions)
 		v1.POST("/messages", claudeCodeHandlers.ClaudeMessages)
 		v1.POST("/messages/count_tokens", claudeCodeHandlers.ClaudeCountTokens)
+		v1.GET("/responses", openaiResponsesHandlers.ResponsesWebsocket)
 		v1.POST("/responses", openaiResponsesHandlers.Responses)
 		v1.POST("/responses/compact", openaiResponsesHandlers.Compact)
 	}
@@ -616,6 +629,7 @@ func (s *Server) registerManagementRoutes() {
 		mgmt.POST("/auth-files", s.mgmt.UploadAuthFile)
 		mgmt.DELETE("/auth-files", s.mgmt.DeleteAuthFile)
 		mgmt.PATCH("/auth-files/status", s.mgmt.PatchAuthFileStatus)
+		mgmt.PATCH("/auth-files/fields", s.mgmt.PatchAuthFileFields)
 		mgmt.POST("/vertex/import", s.mgmt.ImportVertexCredential)
 
 		mgmt.GET("/anthropic-auth-url", s.mgmt.RequestAnthropicToken)
@@ -623,6 +637,7 @@ func (s *Server) registerManagementRoutes() {
 		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("/kimi-auth-url", s.mgmt.RequestKimiToken)
 		mgmt.GET("/iflow-auth-url", s.mgmt.RequestIFlowToken)
 		mgmt.POST("/iflow-auth-url", s.mgmt.RequestIFlowCookieToken)
 		mgmt.POST("/oauth-callback", s.mgmt.PostOAuthCallback)
@@ -654,14 +669,17 @@ func (s *Server) serveManagementControlPanel(c *gin.Context) {
 
 	if _, err := os.Stat(filePath); err != nil {
 		if os.IsNotExist(err) {
-			go managementasset.EnsureLatestManagementHTML(context.Background(), managementasset.StaticDir(s.configFilePath), cfg.ProxyURL, cfg.RemoteManagement.PanelGitHubRepository)
-			c.AbortWithStatus(http.StatusNotFound)
+			// Synchronously ensure management.html is available with a detached context.
+			// Control panel bootstrap should not be canceled by client disconnects.
+			if !managementasset.EnsureLatestManagementHTML(context.Background(), managementasset.StaticDir(s.configFilePath), cfg.ProxyURL, cfg.RemoteManagement.PanelGitHubRepository) {
+				c.AbortWithStatus(http.StatusNotFound)
+				return
+			}
+		} else {
+			log.WithError(err).Error("failed to stat management control panel asset")
+			c.AbortWithStatus(http.StatusInternalServerError)
 			return
 		}
-
-		log.WithError(err).Error("failed to stat management control panel asset")
-		c.AbortWithStatus(http.StatusInternalServerError)
-		return
 	}
 
 	c.File(filePath)
@@ -951,10 +969,6 @@ func (s *Server) UpdateClients(cfg *config.Config) {
 
 	s.handlers.UpdateClients(&cfg.SDKConfig)
 
-	if !cfg.RemoteManagement.DisableControlPanel {
-		staticDir := managementasset.StaticDir(s.configFilePath)
-		go managementasset.EnsureLatestManagementHTML(context.Background(), staticDir, cfg.ProxyURL, cfg.RemoteManagement.PanelGitHubRepository)
-	}
 	if s.mgmt != nil {
 		s.mgmt.SetConfig(cfg)
 		s.mgmt.SetAuthManager(s.handlers.AuthManager)
@@ -1033,14 +1047,10 @@ func AuthMiddleware(manager *sdkaccess.Manager) gin.HandlerFunc {
 			return
 		}
 
-		switch {
-		case errors.Is(err, sdkaccess.ErrNoCredentials):
-			c.AbortWithStatusJSON(http.StatusUnauthorized, gin.H{"error": "Missing API key"})
-		case errors.Is(err, sdkaccess.ErrInvalidCredential):
-			c.AbortWithStatusJSON(http.StatusUnauthorized, gin.H{"error": "Invalid API key"})
-		default:
+		statusCode := err.HTTPStatusCode()
+		if statusCode >= http.StatusInternalServerError {
 			log.Errorf("authentication middleware error: %v", err)
-			c.AbortWithStatusJSON(http.StatusInternalServerError, gin.H{"error": "Authentication service error"})
 		}
+		c.AbortWithStatusJSON(statusCode, gin.H{"error": err.Message})
 	}
 }

internal/auth/claude/anthropic_auth.go

@@ -20,7 +20,7 @@ import (
 // OAuth configuration constants for Claude/Anthropic
 const (
 	AuthURL     = "https://claude.ai/oauth/authorize"
-	TokenURL    = "https://console.anthropic.com/v1/oauth/token"
+	TokenURL    = "https://api.anthropic.com/v1/oauth/token"
 	ClientID    = "9d1c250a-e61b-44d9-88ed-5944d1962f5e"
 	RedirectURI = "http://localhost:54545/callback"
 )

internal/auth/claude/token.go

@@ -36,11 +36,21 @@ type ClaudeTokenStorage struct {
 
 	// Expire is the timestamp when the current access token expires.
 	Expire string `json:"expired"`
+
+	// Metadata holds arbitrary key-value pairs injected via hooks.
+	// It is not exported to JSON directly to allow flattening during serialization.
+	Metadata map[string]any `json:"-"`
+}
+
+// SetMetadata allows external callers to inject metadata into the storage before saving.
+func (ts *ClaudeTokenStorage) SetMetadata(meta map[string]any) {
+	ts.Metadata = meta
 }
 
 // SaveTokenToFile serializes the Claude token storage to a JSON file.
 // This method creates the necessary directory structure and writes the token
 // data in JSON format to the specified file path for persistent storage.
+// It merges any injected metadata into the top-level JSON object.
 //
 // Parameters:
 //   - authFilePath: The full path where the token file should be saved
@@ -65,8 +75,14 @@ func (ts *ClaudeTokenStorage) SaveTokenToFile(authFilePath string) error {
 		_ = f.Close()
 	}()
 
+	// Merge metadata using helper
+	data, errMerge := misc.MergeMetadata(ts, ts.Metadata)
+	if errMerge != nil {
+		return fmt.Errorf("failed to merge metadata: %w", errMerge)
+	}
+
 	// Encode and write the token data as JSON
-	if err = json.NewEncoder(f).Encode(ts); err != nil {
+	if err = json.NewEncoder(f).Encode(data); err != nil {
 		return fmt.Errorf("failed to write token to file: %w", err)
 	}
 	return nil

internal/auth/codex/openai_auth.go

@@ -71,16 +71,26 @@ func (o *CodexAuth) GenerateAuthURL(state string, pkceCodes *PKCECodes) (string,
 // It performs an HTTP POST request to the OpenAI token endpoint with the provided
 // authorization code and PKCE verifier.
 func (o *CodexAuth) ExchangeCodeForTokens(ctx context.Context, code string, pkceCodes *PKCECodes) (*CodexAuthBundle, error) {
+	return o.ExchangeCodeForTokensWithRedirect(ctx, code, RedirectURI, pkceCodes)
+}
+
+// ExchangeCodeForTokensWithRedirect exchanges an authorization code for tokens using
+// a caller-provided redirect URI. This supports alternate auth flows such as device
+// login while preserving the existing token parsing and storage behavior.
+func (o *CodexAuth) ExchangeCodeForTokensWithRedirect(ctx context.Context, code, redirectURI string, pkceCodes *PKCECodes) (*CodexAuthBundle, error) {
 	if pkceCodes == nil {
 		return nil, fmt.Errorf("PKCE codes are required for token exchange")
 	}
+	if strings.TrimSpace(redirectURI) == "" {
+		return nil, fmt.Errorf("redirect URI is required for token exchange")
+	}
 
 	// Prepare token exchange request
 	data := url.Values{
 		"grant_type":    {"authorization_code"},
 		"client_id":     {ClientID},
 		"code":          {code},
-		"redirect_uri":  {RedirectURI},
+		"redirect_uri":  {strings.TrimSpace(redirectURI)},
 		"code_verifier": {pkceCodes.CodeVerifier},
 	}
 

internal/auth/codex/token.go

@@ -32,11 +32,21 @@ type CodexTokenStorage struct {
 	Type string `json:"type"`
 	// Expire is the timestamp when the current access token expires.
 	Expire string `json:"expired"`
+
+	// Metadata holds arbitrary key-value pairs injected via hooks.
+	// It is not exported to JSON directly to allow flattening during serialization.
+	Metadata map[string]any `json:"-"`
+}
+
+// SetMetadata allows external callers to inject metadata into the storage before saving.
+func (ts *CodexTokenStorage) SetMetadata(meta map[string]any) {
+	ts.Metadata = meta
 }
 
 // SaveTokenToFile serializes the Codex token storage to a JSON file.
 // This method creates the necessary directory structure and writes the token
 // data in JSON format to the specified file path for persistent storage.
+// It merges any injected metadata into the top-level JSON object.
 //
 // Parameters:
 //   - authFilePath: The full path where the token file should be saved
@@ -58,7 +68,13 @@ func (ts *CodexTokenStorage) SaveTokenToFile(authFilePath string) error {
 		_ = f.Close()
 	}()
 
-	if err = json.NewEncoder(f).Encode(ts); err != nil {
+	// Merge metadata using helper
+	data, errMerge := misc.MergeMetadata(ts, ts.Metadata)
+	if errMerge != nil {
+		return fmt.Errorf("failed to merge metadata: %w", errMerge)
+	}
+
+	if err = json.NewEncoder(f).Encode(data); err != nil {
 		return fmt.Errorf("failed to write token to file: %w", err)
 	}
 	return nil

internal/auth/gemini/gemini_token.go

@@ -35,11 +35,21 @@ type GeminiTokenStorage struct {
 
 	// Type indicates the authentication provider type, always "gemini" for this storage.
 	Type string `json:"type"`
+
+	// Metadata holds arbitrary key-value pairs injected via hooks.
+	// It is not exported to JSON directly to allow flattening during serialization.
+	Metadata map[string]any `json:"-"`
+}
+
+// SetMetadata allows external callers to inject metadata into the storage before saving.
+func (ts *GeminiTokenStorage) SetMetadata(meta map[string]any) {
+	ts.Metadata = meta
 }
 
 // SaveTokenToFile serializes the Gemini token storage to a JSON file.
 // This method creates the necessary directory structure and writes the token
 // data in JSON format to the specified file path for persistent storage.
+// It merges any injected metadata into the top-level JSON object.
 //
 // Parameters:
 //   - authFilePath: The full path where the token file should be saved
@@ -49,6 +59,11 @@ type GeminiTokenStorage struct {
 func (ts *GeminiTokenStorage) SaveTokenToFile(authFilePath string) error {
 	misc.LogSavingCredentials(authFilePath)
 	ts.Type = "gemini"
+	// Merge metadata using helper
+	data, errMerge := misc.MergeMetadata(ts, ts.Metadata)
+	if errMerge != nil {
+		return fmt.Errorf("failed to merge metadata: %w", errMerge)
+	}
 	if err := os.MkdirAll(filepath.Dir(authFilePath), 0700); err != nil {
 		return fmt.Errorf("failed to create directory: %v", err)
 	}
@@ -63,7 +78,9 @@ func (ts *GeminiTokenStorage) SaveTokenToFile(authFilePath string) error {
 		}
 	}()
 
-	if err = json.NewEncoder(f).Encode(ts); err != nil {
+	enc := json.NewEncoder(f)
+	enc.SetIndent("", "  ")
+	if err := enc.Encode(data); err != nil {
 		return fmt.Errorf("failed to write token to file: %w", err)
 	}
 	return nil

internal/auth/iflow/iflow_token.go

@@ -21,6 +21,15 @@ type IFlowTokenStorage struct {
 	Scope        string `json:"scope"`
 	Cookie       string `json:"cookie"`
 	Type         string `json:"type"`
+
+	// Metadata holds arbitrary key-value pairs injected via hooks.
+	// It is not exported to JSON directly to allow flattening during serialization.
+	Metadata map[string]any `json:"-"`
+}
+
+// SetMetadata allows external callers to inject metadata into the storage before saving.
+func (ts *IFlowTokenStorage) SetMetadata(meta map[string]any) {
+	ts.Metadata = meta
 }
 
 // SaveTokenToFile serialises the token storage to disk.
@@ -37,7 +46,13 @@ func (ts *IFlowTokenStorage) SaveTokenToFile(authFilePath string) error {
 	}
 	defer func() { _ = f.Close() }()
 
-	if err = json.NewEncoder(f).Encode(ts); err != nil {
+	// Merge metadata using helper
+	data, errMerge := misc.MergeMetadata(ts, ts.Metadata)
+	if errMerge != nil {
+		return fmt.Errorf("failed to merge metadata: %w", errMerge)
+	}
+
+	if err = json.NewEncoder(f).Encode(data); err != nil {
 		return fmt.Errorf("iflow token: encode token failed: %w", err)
 	}
 	return nil

internal/auth/kimi/kimi.go

@@ -0,0 +1,396 @@
+// Package kimi provides authentication and token management for Kimi (Moonshot AI) API.
+// It handles the RFC 8628 OAuth2 Device Authorization Grant flow for secure authentication.
+package kimi
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+	"os"
+	"runtime"
+	"strings"
+	"time"
+
+	"github.com/google/uuid"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
+	log "github.com/sirupsen/logrus"
+)
+
+const (
+	// kimiClientID is Kimi Code's OAuth client ID.
+	kimiClientID = "17e5f671-d194-4dfb-9706-5516cb48c098"
+	// kimiOAuthHost is the OAuth server endpoint.
+	kimiOAuthHost = "https://auth.kimi.com"
+	// kimiDeviceCodeURL is the endpoint for requesting device codes.
+	kimiDeviceCodeURL = kimiOAuthHost + "/api/oauth/device_authorization"
+	// kimiTokenURL is the endpoint for exchanging device codes for tokens.
+	kimiTokenURL = kimiOAuthHost + "/api/oauth/token"
+	// KimiAPIBaseURL is the base URL for Kimi API requests.
+	KimiAPIBaseURL = "https://api.kimi.com/coding"
+	// defaultPollInterval is the default interval for polling token endpoint.
+	defaultPollInterval = 5 * time.Second
+	// maxPollDuration is the maximum time to wait for user authorization.
+	maxPollDuration = 15 * time.Minute
+	// refreshThresholdSeconds is when to refresh token before expiry (5 minutes).
+	refreshThresholdSeconds = 300
+)
+
+// KimiAuth handles Kimi authentication flow.
+type KimiAuth struct {
+	deviceClient *DeviceFlowClient
+	cfg          *config.Config
+}
+
+// NewKimiAuth creates a new KimiAuth service instance.
+func NewKimiAuth(cfg *config.Config) *KimiAuth {
+	return &KimiAuth{
+		deviceClient: NewDeviceFlowClient(cfg),
+		cfg:          cfg,
+	}
+}
+
+// StartDeviceFlow initiates the device flow authentication.
+func (k *KimiAuth) StartDeviceFlow(ctx context.Context) (*DeviceCodeResponse, error) {
+	return k.deviceClient.RequestDeviceCode(ctx)
+}
+
+// WaitForAuthorization polls for user authorization and returns the auth bundle.
+func (k *KimiAuth) WaitForAuthorization(ctx context.Context, deviceCode *DeviceCodeResponse) (*KimiAuthBundle, error) {
+	tokenData, err := k.deviceClient.PollForToken(ctx, deviceCode)
+	if err != nil {
+		return nil, err
+	}
+
+	return &KimiAuthBundle{
+		TokenData: tokenData,
+		DeviceID:  k.deviceClient.deviceID,
+	}, nil
+}
+
+// CreateTokenStorage creates a new KimiTokenStorage from auth bundle.
+func (k *KimiAuth) CreateTokenStorage(bundle *KimiAuthBundle) *KimiTokenStorage {
+	expired := ""
+	if bundle.TokenData.ExpiresAt > 0 {
+		expired = time.Unix(bundle.TokenData.ExpiresAt, 0).UTC().Format(time.RFC3339)
+	}
+	return &KimiTokenStorage{
+		AccessToken:  bundle.TokenData.AccessToken,
+		RefreshToken: bundle.TokenData.RefreshToken,
+		TokenType:    bundle.TokenData.TokenType,
+		Scope:        bundle.TokenData.Scope,
+		DeviceID:     strings.TrimSpace(bundle.DeviceID),
+		Expired:      expired,
+		Type:         "kimi",
+	}
+}
+
+// DeviceFlowClient handles the OAuth2 device flow for Kimi.
+type DeviceFlowClient struct {
+	httpClient *http.Client
+	cfg        *config.Config
+	deviceID   string
+}
+
+// NewDeviceFlowClient creates a new device flow client.
+func NewDeviceFlowClient(cfg *config.Config) *DeviceFlowClient {
+	return NewDeviceFlowClientWithDeviceID(cfg, "")
+}
+
+// NewDeviceFlowClientWithDeviceID creates a new device flow client with the specified device ID.
+func NewDeviceFlowClientWithDeviceID(cfg *config.Config, deviceID string) *DeviceFlowClient {
+	client := &http.Client{Timeout: 30 * time.Second}
+	if cfg != nil {
+		client = util.SetProxy(&cfg.SDKConfig, client)
+	}
+	resolvedDeviceID := strings.TrimSpace(deviceID)
+	if resolvedDeviceID == "" {
+		resolvedDeviceID = getOrCreateDeviceID()
+	}
+	return &DeviceFlowClient{
+		httpClient: client,
+		cfg:        cfg,
+		deviceID:   resolvedDeviceID,
+	}
+}
+
+// getOrCreateDeviceID returns an in-memory device ID for the current authentication flow.
+func getOrCreateDeviceID() string {
+	return uuid.New().String()
+}
+
+// getDeviceModel returns a device model string.
+func getDeviceModel() string {
+	osName := runtime.GOOS
+	arch := runtime.GOARCH
+
+	switch osName {
+	case "darwin":
+		return fmt.Sprintf("macOS %s", arch)
+	case "windows":
+		return fmt.Sprintf("Windows %s", arch)
+	case "linux":
+		return fmt.Sprintf("Linux %s", arch)
+	default:
+		return fmt.Sprintf("%s %s", osName, arch)
+	}
+}
+
+// getHostname returns the machine hostname.
+func getHostname() string {
+	hostname, err := os.Hostname()
+	if err != nil {
+		return "unknown"
+	}
+	return hostname
+}
+
+// commonHeaders returns headers required for Kimi API requests.
+func (c *DeviceFlowClient) commonHeaders() map[string]string {
+	return map[string]string{
+		"X-Msh-Platform":     "cli-proxy-api",
+		"X-Msh-Version":      "1.0.0",
+		"X-Msh-Device-Name":  getHostname(),
+		"X-Msh-Device-Model": getDeviceModel(),
+		"X-Msh-Device-Id":    c.deviceID,
+	}
+}
+
+// RequestDeviceCode initiates the device flow by requesting a device code from Kimi.
+func (c *DeviceFlowClient) RequestDeviceCode(ctx context.Context) (*DeviceCodeResponse, error) {
+	data := url.Values{}
+	data.Set("client_id", kimiClientID)
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, kimiDeviceCodeURL, strings.NewReader(data.Encode()))
+	if err != nil {
+		return nil, fmt.Errorf("kimi: failed to create device code request: %w", err)
+	}
+	req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
+	req.Header.Set("Accept", "application/json")
+	for k, v := range c.commonHeaders() {
+		req.Header.Set(k, v)
+	}
+
+	resp, err := c.httpClient.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("kimi: device code request failed: %w", err)
+	}
+	defer func() {
+		if errClose := resp.Body.Close(); errClose != nil {
+			log.Errorf("kimi device code: close body error: %v", errClose)
+		}
+	}()
+
+	bodyBytes, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, fmt.Errorf("kimi: failed to read device code response: %w", err)
+	}
+
+	if resp.StatusCode != http.StatusOK {
+		return nil, fmt.Errorf("kimi: device code request failed with status %d: %s", resp.StatusCode, string(bodyBytes))
+	}
+
+	var deviceCode DeviceCodeResponse
+	if err = json.Unmarshal(bodyBytes, &deviceCode); err != nil {
+		return nil, fmt.Errorf("kimi: failed to parse device code response: %w", err)
+	}
+
+	return &deviceCode, nil
+}
+
+// PollForToken polls the token endpoint until the user authorizes or the device code expires.
+func (c *DeviceFlowClient) PollForToken(ctx context.Context, deviceCode *DeviceCodeResponse) (*KimiTokenData, error) {
+	if deviceCode == nil {
+		return nil, fmt.Errorf("kimi: device code is nil")
+	}
+
+	interval := time.Duration(deviceCode.Interval) * time.Second
+	if interval < defaultPollInterval {
+		interval = defaultPollInterval
+	}
+
+	deadline := time.Now().Add(maxPollDuration)
+	if deviceCode.ExpiresIn > 0 {
+		codeDeadline := time.Now().Add(time.Duration(deviceCode.ExpiresIn) * time.Second)
+		if codeDeadline.Before(deadline) {
+			deadline = codeDeadline
+		}
+	}
+
+	ticker := time.NewTicker(interval)
+	defer ticker.Stop()
+
+	for {
+		select {
+		case <-ctx.Done():
+			return nil, fmt.Errorf("kimi: context cancelled: %w", ctx.Err())
+		case <-ticker.C:
+			if time.Now().After(deadline) {
+				return nil, fmt.Errorf("kimi: device code expired")
+			}
+
+			token, pollErr, shouldContinue := c.exchangeDeviceCode(ctx, deviceCode.DeviceCode)
+			if token != nil {
+				return token, nil
+			}
+			if !shouldContinue {
+				return nil, pollErr
+			}
+			// Continue polling
+		}
+	}
+}
+
+// exchangeDeviceCode attempts to exchange the device code for an access token.
+// Returns (token, error, shouldContinue).
+func (c *DeviceFlowClient) exchangeDeviceCode(ctx context.Context, deviceCode string) (*KimiTokenData, error, bool) {
+	data := url.Values{}
+	data.Set("client_id", kimiClientID)
+	data.Set("device_code", deviceCode)
+	data.Set("grant_type", "urn:ietf:params:oauth:grant-type:device_code")
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, kimiTokenURL, strings.NewReader(data.Encode()))
+	if err != nil {
+		return nil, fmt.Errorf("kimi: failed to create token request: %w", err), false
+	}
+	req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
+	req.Header.Set("Accept", "application/json")
+	for k, v := range c.commonHeaders() {
+		req.Header.Set(k, v)
+	}
+
+	resp, err := c.httpClient.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("kimi: token request failed: %w", err), false
+	}
+	defer func() {
+		if errClose := resp.Body.Close(); errClose != nil {
+			log.Errorf("kimi token exchange: close body error: %v", errClose)
+		}
+	}()
+
+	bodyBytes, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, fmt.Errorf("kimi: failed to read token response: %w", err), false
+	}
+
+	// Parse response - Kimi returns 200 for both success and pending states
+	var oauthResp struct {
+		Error            string  `json:"error"`
+		ErrorDescription string  `json:"error_description"`
+		AccessToken      string  `json:"access_token"`
+		RefreshToken     string  `json:"refresh_token"`
+		TokenType        string  `json:"token_type"`
+		ExpiresIn        float64 `json:"expires_in"`
+		Scope            string  `json:"scope"`
+	}
+
+	if err = json.Unmarshal(bodyBytes, &oauthResp); err != nil {
+		return nil, fmt.Errorf("kimi: failed to parse token response: %w", err), false
+	}
+
+	if oauthResp.Error != "" {
+		switch oauthResp.Error {
+		case "authorization_pending":
+			return nil, nil, true // Continue polling
+		case "slow_down":
+			return nil, nil, true // Continue polling (with increased interval handled by caller)
+		case "expired_token":
+			return nil, fmt.Errorf("kimi: device code expired"), false
+		case "access_denied":
+			return nil, fmt.Errorf("kimi: access denied by user"), false
+		default:
+			return nil, fmt.Errorf("kimi: OAuth error: %s - %s", oauthResp.Error, oauthResp.ErrorDescription), false
+		}
+	}
+
+	if oauthResp.AccessToken == "" {
+		return nil, fmt.Errorf("kimi: empty access token in response"), false
+	}
+
+	var expiresAt int64
+	if oauthResp.ExpiresIn > 0 {
+		expiresAt = time.Now().Unix() + int64(oauthResp.ExpiresIn)
+	}
+
+	return &KimiTokenData{
+		AccessToken:  oauthResp.AccessToken,
+		RefreshToken: oauthResp.RefreshToken,
+		TokenType:    oauthResp.TokenType,
+		ExpiresAt:    expiresAt,
+		Scope:        oauthResp.Scope,
+	}, nil, false
+}
+
+// RefreshToken exchanges a refresh token for a new access token.
+func (c *DeviceFlowClient) RefreshToken(ctx context.Context, refreshToken string) (*KimiTokenData, error) {
+	data := url.Values{}
+	data.Set("client_id", kimiClientID)
+	data.Set("grant_type", "refresh_token")
+	data.Set("refresh_token", refreshToken)
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, kimiTokenURL, strings.NewReader(data.Encode()))
+	if err != nil {
+		return nil, fmt.Errorf("kimi: failed to create refresh request: %w", err)
+	}
+	req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
+	req.Header.Set("Accept", "application/json")
+	for k, v := range c.commonHeaders() {
+		req.Header.Set(k, v)
+	}
+
+	resp, err := c.httpClient.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("kimi: refresh request failed: %w", err)
+	}
+	defer func() {
+		if errClose := resp.Body.Close(); errClose != nil {
+			log.Errorf("kimi refresh token: close body error: %v", errClose)
+		}
+	}()
+
+	bodyBytes, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, fmt.Errorf("kimi: failed to read refresh response: %w", err)
+	}
+
+	if resp.StatusCode == http.StatusUnauthorized || resp.StatusCode == http.StatusForbidden {
+		return nil, fmt.Errorf("kimi: refresh token rejected (status %d)", resp.StatusCode)
+	}
+
+	if resp.StatusCode != http.StatusOK {
+		return nil, fmt.Errorf("kimi: refresh failed with status %d: %s", resp.StatusCode, string(bodyBytes))
+	}
+
+	var tokenResp struct {
+		AccessToken  string  `json:"access_token"`
+		RefreshToken string  `json:"refresh_token"`
+		TokenType    string  `json:"token_type"`
+		ExpiresIn    float64 `json:"expires_in"`
+		Scope        string  `json:"scope"`
+	}
+
+	if err = json.Unmarshal(bodyBytes, &tokenResp); err != nil {
+		return nil, fmt.Errorf("kimi: failed to parse refresh response: %w", err)
+	}
+
+	if tokenResp.AccessToken == "" {
+		return nil, fmt.Errorf("kimi: empty access token in refresh response")
+	}
+
+	var expiresAt int64
+	if tokenResp.ExpiresIn > 0 {
+		expiresAt = time.Now().Unix() + int64(tokenResp.ExpiresIn)
+	}
+
+	return &KimiTokenData{
+		AccessToken:  tokenResp.AccessToken,
+		RefreshToken: tokenResp.RefreshToken,
+		TokenType:    tokenResp.TokenType,
+		ExpiresAt:    expiresAt,
+		Scope:        tokenResp.Scope,
+	}, nil
+}

internal/auth/kimi/token.go

@@ -0,0 +1,131 @@
+// Package kimi provides authentication and token management functionality
+// for Kimi (Moonshot AI) services. It handles OAuth2 device flow token storage,
+// serialization, and retrieval for maintaining authenticated sessions with the Kimi API.
+package kimi
+
+import (
+	"encoding/json"
+	"fmt"
+	"os"
+	"path/filepath"
+	"time"
+
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/misc"
+)
+
+// KimiTokenStorage stores OAuth2 token information for Kimi API authentication.
+type KimiTokenStorage struct {
+	// AccessToken is the OAuth2 access token used for authenticating API requests.
+	AccessToken string `json:"access_token"`
+	// RefreshToken is the OAuth2 refresh token used to obtain new access tokens.
+	RefreshToken string `json:"refresh_token"`
+	// TokenType is the type of token, typically "Bearer".
+	TokenType string `json:"token_type"`
+	// Scope is the OAuth2 scope granted to the token.
+	Scope string `json:"scope,omitempty"`
+	// DeviceID is the OAuth device flow identifier used for Kimi requests.
+	DeviceID string `json:"device_id,omitempty"`
+	// Expired is the RFC3339 timestamp when the access token expires.
+	Expired string `json:"expired,omitempty"`
+	// Type indicates the authentication provider type, always "kimi" for this storage.
+	Type string `json:"type"`
+
+	// Metadata holds arbitrary key-value pairs injected via hooks.
+	// It is not exported to JSON directly to allow flattening during serialization.
+	Metadata map[string]any `json:"-"`
+}
+
+// SetMetadata allows external callers to inject metadata into the storage before saving.
+func (ts *KimiTokenStorage) SetMetadata(meta map[string]any) {
+	ts.Metadata = meta
+}
+
+// KimiTokenData holds the raw OAuth token response from Kimi.
+type KimiTokenData struct {
+	// AccessToken is the OAuth2 access token.
+	AccessToken string `json:"access_token"`
+	// RefreshToken is the OAuth2 refresh token.
+	RefreshToken string `json:"refresh_token"`
+	// TokenType is the type of token, typically "Bearer".
+	TokenType string `json:"token_type"`
+	// ExpiresAt is the Unix timestamp when the token expires.
+	ExpiresAt int64 `json:"expires_at"`
+	// Scope is the OAuth2 scope granted to the token.
+	Scope string `json:"scope"`
+}
+
+// KimiAuthBundle bundles authentication data for storage.
+type KimiAuthBundle struct {
+	// TokenData contains the OAuth token information.
+	TokenData *KimiTokenData
+	// DeviceID is the device identifier used during OAuth device flow.
+	DeviceID string
+}
+
+// DeviceCodeResponse represents Kimi's device code response.
+type DeviceCodeResponse struct {
+	// DeviceCode is the device verification code.
+	DeviceCode string `json:"device_code"`
+	// UserCode is the code the user must enter at the verification URI.
+	UserCode string `json:"user_code"`
+	// VerificationURI is the URL where the user should enter the code.
+	VerificationURI string `json:"verification_uri,omitempty"`
+	// VerificationURIComplete is the URL with the code pre-filled.
+	VerificationURIComplete string `json:"verification_uri_complete"`
+	// ExpiresIn is the number of seconds until the device code expires.
+	ExpiresIn int `json:"expires_in"`
+	// Interval is the minimum number of seconds to wait between polling requests.
+	Interval int `json:"interval"`
+}
+
+// SaveTokenToFile serializes the Kimi token storage to a JSON file.
+func (ts *KimiTokenStorage) SaveTokenToFile(authFilePath string) error {
+	misc.LogSavingCredentials(authFilePath)
+	ts.Type = "kimi"
+
+	if err := os.MkdirAll(filepath.Dir(authFilePath), 0700); err != nil {
+		return fmt.Errorf("failed to create directory: %v", err)
+	}
+
+	f, err := os.Create(authFilePath)
+	if err != nil {
+		return fmt.Errorf("failed to create token file: %w", err)
+	}
+	defer func() {
+		_ = f.Close()
+	}()
+
+	// Merge metadata using helper
+	data, errMerge := misc.MergeMetadata(ts, ts.Metadata)
+	if errMerge != nil {
+		return fmt.Errorf("failed to merge metadata: %w", errMerge)
+	}
+
+	encoder := json.NewEncoder(f)
+	encoder.SetIndent("", "  ")
+	if err = encoder.Encode(data); err != nil {
+		return fmt.Errorf("failed to write token to file: %w", err)
+	}
+	return nil
+}
+
+// IsExpired checks if the token has expired.
+func (ts *KimiTokenStorage) IsExpired() bool {
+	if ts.Expired == "" {
+		return false // No expiry set, assume valid
+	}
+	t, err := time.Parse(time.RFC3339, ts.Expired)
+	if err != nil {
+		return true // Has expiry string but can't parse
+	}
+	// Consider expired if within refresh threshold
+	return time.Now().Add(time.Duration(refreshThresholdSeconds) * time.Second).After(t)
+}
+
+// NeedsRefresh checks if the token should be refreshed.
+func (ts *KimiTokenStorage) NeedsRefresh() bool {
+	if ts.RefreshToken == "" {
+		return false // Can't refresh without refresh token
+	}
+	return ts.IsExpired()
+}

internal/auth/qwen/qwen_token.go

@@ -30,11 +30,21 @@ type QwenTokenStorage struct {
 	Type string `json:"type"`
 	// Expire is the timestamp when the current access token expires.
 	Expire string `json:"expired"`
+
+	// Metadata holds arbitrary key-value pairs injected via hooks.
+	// It is not exported to JSON directly to allow flattening during serialization.
+	Metadata map[string]any `json:"-"`
+}
+
+// SetMetadata allows external callers to inject metadata into the storage before saving.
+func (ts *QwenTokenStorage) SetMetadata(meta map[string]any) {
+	ts.Metadata = meta
 }
 
 // SaveTokenToFile serializes the Qwen token storage to a JSON file.
 // This method creates the necessary directory structure and writes the token
 // data in JSON format to the specified file path for persistent storage.
+// It merges any injected metadata into the top-level JSON object.
 //
 // Parameters:
 //   - authFilePath: The full path where the token file should be saved
@@ -56,7 +66,13 @@ func (ts *QwenTokenStorage) SaveTokenToFile(authFilePath string) error {
 		_ = f.Close()
 	}()
 
-	if err = json.NewEncoder(f).Encode(ts); err != nil {
+	// Merge metadata using helper
+	data, errMerge := misc.MergeMetadata(ts, ts.Metadata)
+	if errMerge != nil {
+		return fmt.Errorf("failed to merge metadata: %w", errMerge)
+	}
+
+	if err = json.NewEncoder(f).Encode(data); err != nil {
 		return fmt.Errorf("failed to write token to file: %w", err)
 	}
 	return nil

internal/cmd/anthropic_login.go

@@ -40,8 +40,7 @@ func DoClaudeLogin(cfg *config.Config, options *LoginOptions) {
 
 	_, savedPath, err := manager.Login(context.Background(), "claude", cfg, authOpts)
 	if err != nil {
-		var authErr *claude.AuthenticationError
-		if errors.As(err, &authErr) {
+		if authErr, ok := errors.AsType[*claude.AuthenticationError](err); ok {
 			log.Error(claude.GetUserFriendlyMessage(authErr))
 			if authErr.Type == claude.ErrPortInUse.Type {
 				os.Exit(claude.ErrPortInUse.Code)

internal/cmd/auth_manager.go

@@ -19,6 +19,7 @@ func newAuthManager() *sdkAuth.Manager {
 		sdkAuth.NewQwenAuthenticator(),
 		sdkAuth.NewIFlowAuthenticator(),
 		sdkAuth.NewAntigravityAuthenticator(),
+		sdkAuth.NewKimiAuthenticator(),
 	)
 	return manager
 }

internal/cmd/iflow_login.go

@@ -32,8 +32,7 @@ func DoIFlowLogin(cfg *config.Config, options *LoginOptions) {
 
 	_, savedPath, err := manager.Login(context.Background(), "iflow", cfg, authOpts)
 	if err != nil {
-		var emailErr *sdkAuth.EmailRequiredError
-		if errors.As(err, &emailErr) {
+		if emailErr, ok := errors.AsType[*sdkAuth.EmailRequiredError](err); ok {
 			log.Error(emailErr.Error())
 			return
 		}

internal/cmd/kimi_login.go

@@ -0,0 +1,44 @@
+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"
+)
+
+// DoKimiLogin triggers the OAuth device flow for Kimi (Moonshot AI) and saves tokens.
+// It initiates the device flow authentication, displays the verification URL for the user,
+// and waits for authorization before saving the tokens.
+//
+// Parameters:
+//   - cfg: The application configuration containing proxy and auth directory settings
+//   - options: Login options including browser behavior settings
+func DoKimiLogin(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(), "kimi", cfg, authOpts)
+	if err != nil {
+		log.Errorf("Kimi 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("Kimi authentication successful!")
+}

internal/cmd/login.go

@@ -100,49 +100,74 @@ func DoLogin(cfg *config.Config, projectID string, options *LoginOptions) {
 
 	log.Info("Authentication successful.")
 
-	projects, errProjects := fetchGCPProjects(ctx, httpClient)
-	if errProjects != nil {
-		log.Errorf("Failed to get project list: %v", errProjects)
-		return
+	var activatedProjects []string
+
+	useGoogleOne := false
+	if trimmedProjectID == "" && promptFn != nil {
+		fmt.Println("\nSelect login mode:")
+		fmt.Println("  1. Code Assist  (GCP project, manual selection)")
+		fmt.Println("  2. Google One   (personal account, auto-discover project)")
+		choice, errPrompt := promptFn("Enter choice [1/2] (default: 1): ")
+		if errPrompt == nil && strings.TrimSpace(choice) == "2" {
+			useGoogleOne = true
+		}
 	}
 
-	selectedProjectID := promptForProjectSelection(projects, trimmedProjectID, promptFn)
-	projectSelections, errSelection := resolveProjectSelections(selectedProjectID, projects)
-	if errSelection != nil {
-		log.Errorf("Invalid project selection: %v", errSelection)
-		return
-	}
-	if len(projectSelections) == 0 {
-		log.Error("No project selected; aborting login.")
-		return
-	}
-
-	activatedProjects := make([]string, 0, len(projectSelections))
-	seenProjects := make(map[string]bool)
-	for _, candidateID := range projectSelections {
-		log.Infof("Activating project %s", candidateID)
-		if errSetup := performGeminiCLISetup(ctx, httpClient, storage, candidateID); errSetup != nil {
-			var projectErr *projectSelectionRequiredError
-			if errors.As(errSetup, &projectErr) {
-				log.Error("Failed to start user onboarding: A project ID is required.")
-				showProjectSelectionHelp(storage.Email, projects)
-				return
-			}
-			log.Errorf("Failed to complete user setup: %v", errSetup)
+	if useGoogleOne {
+		log.Info("Google One mode: auto-discovering project...")
+		if errSetup := performGeminiCLISetup(ctx, httpClient, storage, ""); errSetup != nil {
+			log.Errorf("Google One auto-discovery failed: %v", errSetup)
 			return
 		}
-		finalID := strings.TrimSpace(storage.ProjectID)
-		if finalID == "" {
-			finalID = candidateID
+		autoProject := strings.TrimSpace(storage.ProjectID)
+		if autoProject == "" {
+			log.Error("Google One auto-discovery returned empty project ID")
+			return
+		}
+		log.Infof("Auto-discovered project: %s", autoProject)
+		activatedProjects = []string{autoProject}
+	} else {
+		projects, errProjects := fetchGCPProjects(ctx, httpClient)
+		if errProjects != nil {
+			log.Errorf("Failed to get project list: %v", errProjects)
+			return
 		}
 
-		// Skip duplicates
-		if seenProjects[finalID] {
-			log.Infof("Project %s already activated, skipping", finalID)
-			continue
+		selectedProjectID := promptForProjectSelection(projects, trimmedProjectID, promptFn)
+		projectSelections, errSelection := resolveProjectSelections(selectedProjectID, projects)
+		if errSelection != nil {
+			log.Errorf("Invalid project selection: %v", errSelection)
+			return
+		}
+		if len(projectSelections) == 0 {
+			log.Error("No project selected; aborting login.")
+			return
+		}
+
+		seenProjects := make(map[string]bool)
+		for _, candidateID := range projectSelections {
+			log.Infof("Activating project %s", candidateID)
+			if errSetup := performGeminiCLISetup(ctx, httpClient, storage, candidateID); errSetup != nil {
+				if _, ok := errors.AsType[*projectSelectionRequiredError](errSetup); ok {
+					log.Error("Failed to start user onboarding: A project ID is required.")
+					showProjectSelectionHelp(storage.Email, projects)
+					return
+				}
+				log.Errorf("Failed to complete user setup: %v", errSetup)
+				return
+			}
+			finalID := strings.TrimSpace(storage.ProjectID)
+			if finalID == "" {
+				finalID = candidateID
+			}
+
+			if seenProjects[finalID] {
+				log.Infof("Project %s already activated, skipping", finalID)
+				continue
+			}
+			seenProjects[finalID] = true
+			activatedProjects = append(activatedProjects, finalID)
 		}
-		seenProjects[finalID] = true
-		activatedProjects = append(activatedProjects, finalID)
 	}
 
 	storage.Auto = false
@@ -235,7 +260,48 @@ func performGeminiCLISetup(ctx context.Context, httpClient *http.Client, storage
 		}
 	}
 	if projectID == "" {
-		return &projectSelectionRequiredError{}
+		// Auto-discovery: try onboardUser without specifying a project
+		// to let Google auto-provision one (matches Gemini CLI headless behavior
+		// and Antigravity's FetchProjectID pattern).
+		autoOnboardReq := map[string]any{
+			"tierId":   tierID,
+			"metadata": metadata,
+		}
+
+		autoCtx, autoCancel := context.WithTimeout(ctx, 30*time.Second)
+		defer autoCancel()
+		for attempt := 1; ; attempt++ {
+			var onboardResp map[string]any
+			if errOnboard := callGeminiCLI(autoCtx, httpClient, "onboardUser", autoOnboardReq, &onboardResp); errOnboard != nil {
+				return fmt.Errorf("auto-discovery onboardUser: %w", errOnboard)
+			}
+
+			if done, okDone := onboardResp["done"].(bool); okDone && done {
+				if resp, okResp := onboardResp["response"].(map[string]any); okResp {
+					switch v := resp["cloudaicompanionProject"].(type) {
+					case string:
+						projectID = strings.TrimSpace(v)
+					case map[string]any:
+						if id, okID := v["id"].(string); okID {
+							projectID = strings.TrimSpace(id)
+						}
+					}
+				}
+				break
+			}
+
+			log.Debugf("Auto-discovery: onboarding in progress, attempt %d...", attempt)
+			select {
+			case <-autoCtx.Done():
+				return &projectSelectionRequiredError{}
+			case <-time.After(2 * time.Second):
+			}
+		}
+
+		if projectID == "" {
+			return &projectSelectionRequiredError{}
+		}
+		log.Infof("Auto-discovered project ID via onboarding: %s", projectID)
 	}
 
 	onboardReqBody := map[string]any{
@@ -617,7 +683,7 @@ func updateAuthRecord(record *cliproxyauth.Auth, storage *gemini.GeminiTokenStor
 		return
 	}
 
-	finalName := gemini.CredentialFileName(storage.Email, storage.ProjectID, false)
+	finalName := gemini.CredentialFileName(storage.Email, storage.ProjectID, true)
 
 	if record.Metadata == nil {
 		record.Metadata = make(map[string]any)

internal/cmd/openai_device_login.go

@@ -0,0 +1,60 @@
+package cmd
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"os"
+
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/auth/codex"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	sdkAuth "github.com/router-for-me/CLIProxyAPI/v6/sdk/auth"
+	log "github.com/sirupsen/logrus"
+)
+
+const (
+	codexLoginModeMetadataKey = "codex_login_mode"
+	codexLoginModeDevice      = "device"
+)
+
+// DoCodexDeviceLogin triggers the Codex device-code flow while keeping the
+// existing codex-login OAuth callback flow intact.
+func DoCodexDeviceLogin(cfg *config.Config, options *LoginOptions) {
+	if options == nil {
+		options = &LoginOptions{}
+	}
+
+	promptFn := options.Prompt
+	if promptFn == nil {
+		promptFn = defaultProjectPrompt()
+	}
+
+	manager := newAuthManager()
+
+	authOpts := &sdkAuth.LoginOptions{
+		NoBrowser:    options.NoBrowser,
+		CallbackPort: options.CallbackPort,
+		Metadata: map[string]string{
+			codexLoginModeMetadataKey: codexLoginModeDevice,
+		},
+		Prompt: promptFn,
+	}
+
+	_, savedPath, err := manager.Login(context.Background(), "codex", cfg, authOpts)
+	if err != nil {
+		if authErr, ok := errors.AsType[*codex.AuthenticationError](err); ok {
+			log.Error(codex.GetUserFriendlyMessage(authErr))
+			if authErr.Type == codex.ErrPortInUse.Type {
+				os.Exit(codex.ErrPortInUse.Code)
+			}
+			return
+		}
+		fmt.Printf("Codex device authentication failed: %v\n", err)
+		return
+	}
+
+	if savedPath != "" {
+		fmt.Printf("Authentication saved to %s\n", savedPath)
+	}
+	fmt.Println("Codex device authentication successful!")
+}

internal/cmd/openai_login.go

@@ -54,8 +54,7 @@ func DoCodexLogin(cfg *config.Config, options *LoginOptions) {
 
 	_, savedPath, err := manager.Login(context.Background(), "codex", cfg, authOpts)
 	if err != nil {
-		var authErr *codex.AuthenticationError
-		if errors.As(err, &authErr) {
+		if authErr, ok := errors.AsType[*codex.AuthenticationError](err); ok {
 			log.Error(codex.GetUserFriendlyMessage(authErr))
 			if authErr.Type == codex.ErrPortInUse.Type {
 				os.Exit(codex.ErrPortInUse.Code)

internal/cmd/qwen_login.go

@@ -44,8 +44,7 @@ func DoQwenLogin(cfg *config.Config, options *LoginOptions) {
 
 	_, savedPath, err := manager.Login(context.Background(), "qwen", cfg, authOpts)
 	if err != nil {
-		var emailErr *sdkAuth.EmailRequiredError
-		if errors.As(err, &emailErr) {
+		if emailErr, ok := errors.AsType[*sdkAuth.EmailRequiredError](err); ok {
 			log.Error(emailErr.Error())
 			return
 		}

internal/cmd/run.go

@@ -55,6 +55,34 @@ func StartService(cfg *config.Config, configPath string, localPassword string) {
 	}
 }
 
+// StartServiceBackground starts the proxy service in a background goroutine
+// and returns a cancel function for shutdown and a done channel.
+func StartServiceBackground(cfg *config.Config, configPath string, localPassword string) (cancel func(), done <-chan struct{}) {
+	builder := cliproxy.NewBuilder().
+		WithConfig(cfg).
+		WithConfigPath(configPath).
+		WithLocalManagementPassword(localPassword)
+
+	ctx, cancelFn := context.WithCancel(context.Background())
+	doneCh := make(chan struct{})
+
+	service, err := builder.Build()
+	if err != nil {
+		log.Errorf("failed to build proxy service: %v", err)
+		close(doneCh)
+		return cancelFn, doneCh
+	}
+
+	go func() {
+		defer close(doneCh)
+		if err := service.Run(ctx); err != nil && !errors.Is(err, context.Canceled) {
+			log.Errorf("proxy service exited with error: %v", err)
+		}
+	}()
+
+	return cancelFn, doneCh
+}
+
 // WaitForCloudDeploy waits indefinitely for shutdown signals in cloud deploy mode
 // when no configuration file is available.
 func WaitForCloudDeploy() {

internal/config/config.go

@@ -90,6 +90,10 @@ type Config struct {
 	// 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"`
 
+	// ClaudeHeaderDefaults configures default header values for Claude API requests.
+	// These are used as fallbacks when the client does not send its own headers.
+	ClaudeHeaderDefaults ClaudeHeaderDefaults `yaml:"claude-header-defaults" json:"claude-header-defaults"`
+
 	// OpenAICompatibility defines OpenAI API compatibility configurations for external providers.
 	OpenAICompatibility []OpenAICompatibility `yaml:"openai-compatibility" json:"openai-compatibility"`
 
@@ -117,6 +121,15 @@ type Config struct {
 	legacyMigrationPending bool `yaml:"-" json:"-"`
 }
 
+// ClaudeHeaderDefaults configures default header values injected into Claude API requests
+// when the client does not send them. Update these when Claude Code releases a new version.
+type ClaudeHeaderDefaults struct {
+	UserAgent      string `yaml:"user-agent" json:"user-agent"`
+	PackageVersion string `yaml:"package-version" json:"package-version"`
+	RuntimeVersion string `yaml:"runtime-version" json:"runtime-version"`
+	Timeout        string `yaml:"timeout" json:"timeout"`
+}
+
 // TLSConfig holds HTTPS server settings.
 type TLSConfig struct {
 	// Enable toggles HTTPS server mode.
@@ -288,6 +301,10 @@ type CloakConfig struct {
 	// SensitiveWords is a list of words to obfuscate with zero-width characters.
 	// This can help bypass certain content filters.
 	SensitiveWords []string `yaml:"sensitive-words,omitempty" json:"sensitive-words,omitempty"`
+
+	// CacheUserID controls whether Claude user_id values are cached per API key.
+	// When false, a fresh random user_id is generated for every request.
+	CacheUserID *bool `yaml:"cache-user-id,omitempty" json:"cache-user-id,omitempty"`
 }
 
 // ClaudeKey represents the configuration for a Claude API key,
@@ -355,6 +372,9 @@ type CodexKey struct {
 	// If empty, the default Codex API URL will be used.
 	BaseURL string `yaml:"base-url" json:"base-url"`
 
+	// Websockets enables the Responses API websocket transport for this credential.
+	Websockets bool `yaml:"websockets,omitempty" json:"websockets,omitempty"`
+
 	// ProxyURL overrides the global proxy setting for this API key if provided.
 	ProxyURL string `yaml:"proxy-url" json:"proxy-url"`
 
@@ -493,14 +513,15 @@ func LoadConfig(configFile string) (*Config, error) {
 // If optional is true and the file is missing, it returns an empty Config.
 // If optional is true and the file is empty or invalid, it returns an empty Config.
 func LoadConfigOptional(configFile string, optional bool) (*Config, error) {
-	// Perform oauth-model-alias migration before loading config.
-	// This migrates oauth-model-mappings to oauth-model-alias if needed.
-	if migrated, err := MigrateOAuthModelAlias(configFile); err != nil {
-		// Log warning but don't fail - config loading should still work
-		fmt.Printf("Warning: oauth-model-alias migration failed: %v\n", err)
-	} else if migrated {
-		fmt.Println("Migrated oauth-model-mappings to oauth-model-alias")
-	}
+	// NOTE: Startup oauth-model-alias migration is intentionally disabled.
+	// Reason: avoid mutating config.yaml during server startup.
+	// Re-enable the block below if automatic startup migration is needed again.
+	// if migrated, err := MigrateOAuthModelAlias(configFile); err != nil {
+	// 	// Log warning but don't fail - config loading should still work
+	// 	fmt.Printf("Warning: oauth-model-alias migration failed: %v\n", err)
+	// } else if migrated {
+	// 	fmt.Println("Migrated oauth-model-mappings to oauth-model-alias")
+	// }
 
 	// Read the entire configuration file into memory.
 	data, err := os.ReadFile(configFile)
@@ -540,18 +561,21 @@ 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
-		}
-	}
+	// NOTE: Startup legacy key migration is intentionally disabled.
+	// Reason: avoid mutating config.yaml during server startup.
+	// Re-enable the block below if automatic startup migration is needed again.
+	// 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).
@@ -585,9 +609,6 @@ func LoadConfigOptional(configFile string, optional bool) (*Config, error) {
 		cfg.ErrorLogsMaxFiles = 10
 	}
 
-	// Sync request authentication providers with inline API keys for backwards compatibility.
-	syncInlineAccessProvider(&cfg)
-
 	// Sanitize Gemini API key configuration and migrate legacy entries.
 	cfg.SanitizeGeminiKeys()
 
@@ -612,17 +633,20 @@ func LoadConfigOptional(configFile string, optional bool) (*Config, error) {
 	// Validate raw payload rules and drop invalid entries.
 	cfg.SanitizePayloadRules()
 
-	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.")
-		}
-	}
+	// NOTE: Legacy migration persistence is intentionally disabled together with
+	// startup legacy migration to keep startup read-only for config.yaml.
+	// Re-enable the block below if automatic startup migration is needed again.
+	// 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
@@ -818,18 +842,6 @@ func normalizeModelPrefix(prefix string) string {
 	return trimmed
 }
 
-func syncInlineAccessProvider(cfg *Config) {
-	if cfg == nil {
-		return
-	}
-	if len(cfg.APIKeys) == 0 {
-		if provider := cfg.ConfigAPIKeyProvider(); provider != nil && len(provider.APIKeys) > 0 {
-			cfg.APIKeys = append([]string(nil), provider.APIKeys...)
-		}
-	}
-	cfg.Access.Providers = nil
-}
-
 // looksLikeBcrypt returns true if the provided string appears to be a bcrypt hash.
 func looksLikeBcrypt(s string) bool {
 	return len(s) > 4 && (s[:4] == "$2a$" || s[:4] == "$2b$" || s[:4] == "$2y$")
@@ -917,7 +929,7 @@ func hashSecret(secret string) (string, error) {
 // SaveConfigPreserveComments writes the config back to YAML while preserving existing comments
 // and key ordering by loading the original file into a yaml.Node tree and updating values in-place.
 func SaveConfigPreserveComments(configFile string, cfg *Config) error {
-	persistCfg := sanitizeConfigForPersist(cfg)
+	persistCfg := cfg
 	// Load original YAML as a node tree to preserve comments and ordering.
 	data, err := os.ReadFile(configFile)
 	if err != nil {
@@ -985,16 +997,6 @@ func SaveConfigPreserveComments(configFile string, cfg *Config) error {
 	return err
 }
 
-func sanitizeConfigForPersist(cfg *Config) *Config {
-	if cfg == nil {
-		return nil
-	}
-	clone := *cfg
-	clone.SDKConfig = cfg.SDKConfig
-	clone.SDKConfig.Access = AccessConfig{}
-	return &clone
-}
-
 // SaveConfigPreserveCommentsUpdateNestedScalar updates a nested scalar key path like ["a","b"]
 // while preserving comments and positions.
 func SaveConfigPreserveCommentsUpdateNestedScalar(configFile string, path []string, value string) error {
@@ -1091,8 +1093,13 @@ func getOrCreateMapValue(mapNode *yaml.Node, key string) *yaml.Node {
 
 // mergeMappingPreserve merges keys from src into dst mapping node while preserving
 // key order and comments of existing keys in dst. New keys are only added if their
-// value is non-zero to avoid polluting the config with defaults.
-func mergeMappingPreserve(dst, src *yaml.Node) {
+// value is non-zero and not a known default to avoid polluting the config with defaults.
+func mergeMappingPreserve(dst, src *yaml.Node, path ...[]string) {
+	var currentPath []string
+	if len(path) > 0 {
+		currentPath = path[0]
+	}
+
 	if dst == nil || src == nil {
 		return
 	}
@@ -1106,16 +1113,19 @@ func mergeMappingPreserve(dst, src *yaml.Node) {
 		sk := src.Content[i]
 		sv := src.Content[i+1]
 		idx := findMapKeyIndex(dst, sk.Value)
+		childPath := appendPath(currentPath, sk.Value)
 		if idx >= 0 {
 			// Merge into existing value node (always update, even to zero values)
 			dv := dst.Content[idx+1]
-			mergeNodePreserve(dv, sv)
+			mergeNodePreserve(dv, sv, childPath)
 		} else {
-			// New key: only add if value is non-zero to avoid polluting config with defaults
-			if isZeroValueNode(sv) {
+			// New key: only add if value is non-zero and not a known default
+			candidate := deepCopyNode(sv)
+			pruneKnownDefaultsInNewNode(childPath, candidate)
+			if isKnownDefaultValue(childPath, candidate) {
 				continue
 			}
-			dst.Content = append(dst.Content, deepCopyNode(sk), deepCopyNode(sv))
+			dst.Content = append(dst.Content, deepCopyNode(sk), candidate)
 		}
 	}
 }
@@ -1123,7 +1133,12 @@ func mergeMappingPreserve(dst, src *yaml.Node) {
 // mergeNodePreserve merges src into dst for scalars, mappings and sequences while
 // reusing destination nodes to keep comments and anchors. For sequences, it updates
 // in-place by index.
-func mergeNodePreserve(dst, src *yaml.Node) {
+func mergeNodePreserve(dst, src *yaml.Node, path ...[]string) {
+	var currentPath []string
+	if len(path) > 0 {
+		currentPath = path[0]
+	}
+
 	if dst == nil || src == nil {
 		return
 	}
@@ -1132,7 +1147,7 @@ func mergeNodePreserve(dst, src *yaml.Node) {
 		if dst.Kind != yaml.MappingNode {
 			copyNodeShallow(dst, src)
 		}
-		mergeMappingPreserve(dst, src)
+		mergeMappingPreserve(dst, src, currentPath)
 	case yaml.SequenceNode:
 		// Preserve explicit null style if dst was null and src is empty sequence
 		if dst.Kind == yaml.ScalarNode && dst.Tag == "!!null" && len(src.Content) == 0 {
@@ -1155,7 +1170,7 @@ func mergeNodePreserve(dst, src *yaml.Node) {
 				dst.Content[i] = deepCopyNode(src.Content[i])
 				continue
 			}
-			mergeNodePreserve(dst.Content[i], src.Content[i])
+			mergeNodePreserve(dst.Content[i], src.Content[i], currentPath)
 			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])
@@ -1197,6 +1212,94 @@ func findMapKeyIndex(mapNode *yaml.Node, key string) int {
 	return -1
 }
 
+// appendPath appends a key to the path, returning a new slice to avoid modifying the original.
+func appendPath(path []string, key string) []string {
+	if len(path) == 0 {
+		return []string{key}
+	}
+	newPath := make([]string, len(path)+1)
+	copy(newPath, path)
+	newPath[len(path)] = key
+	return newPath
+}
+
+// isKnownDefaultValue returns true if the given node at the specified path
+// represents a known default value that should not be written to the config file.
+// This prevents non-zero defaults from polluting the config.
+func isKnownDefaultValue(path []string, node *yaml.Node) bool {
+	// First check if it's a zero value
+	if isZeroValueNode(node) {
+		return true
+	}
+
+	// Match known non-zero defaults by exact dotted path.
+	if len(path) == 0 {
+		return false
+	}
+
+	fullPath := strings.Join(path, ".")
+
+	// Check string defaults
+	if node.Kind == yaml.ScalarNode && node.Tag == "!!str" {
+		switch fullPath {
+		case "pprof.addr":
+			return node.Value == DefaultPprofAddr
+		case "remote-management.panel-github-repository":
+			return node.Value == DefaultPanelGitHubRepository
+		case "routing.strategy":
+			return node.Value == "round-robin"
+		}
+	}
+
+	// Check integer defaults
+	if node.Kind == yaml.ScalarNode && node.Tag == "!!int" {
+		switch fullPath {
+		case "error-logs-max-files":
+			return node.Value == "10"
+		}
+	}
+
+	return false
+}
+
+// pruneKnownDefaultsInNewNode removes default-valued descendants from a new node
+// before it is appended into the destination YAML tree.
+func pruneKnownDefaultsInNewNode(path []string, node *yaml.Node) {
+	if node == nil {
+		return
+	}
+
+	switch node.Kind {
+	case yaml.MappingNode:
+		filtered := make([]*yaml.Node, 0, len(node.Content))
+		for i := 0; i+1 < len(node.Content); i += 2 {
+			keyNode := node.Content[i]
+			valueNode := node.Content[i+1]
+			if keyNode == nil || valueNode == nil {
+				continue
+			}
+
+			childPath := appendPath(path, keyNode.Value)
+			if isKnownDefaultValue(childPath, valueNode) {
+				continue
+			}
+
+			pruneKnownDefaultsInNewNode(childPath, valueNode)
+			if (valueNode.Kind == yaml.MappingNode || valueNode.Kind == yaml.SequenceNode) &&
+				len(valueNode.Content) == 0 {
+				continue
+			}
+
+			filtered = append(filtered, keyNode, valueNode)
+		}
+		node.Content = filtered
+	case yaml.SequenceNode:
+		for _, child := range node.Content {
+			pruneKnownDefaultsInNewNode(path, child)
+		}
+	}
+}
+
 // isZeroValueNode returns true if the YAML node represents a zero/default value
 // that should not be written as a new key to preserve config cleanliness.
 // For mappings and sequences, recursively checks if all children are zero values.

internal/config/sdk_config.go

@@ -20,8 +20,9 @@ type SDKConfig struct {
 	// APIKeys is a list of keys for authenticating clients to this proxy server.
 	APIKeys []string `yaml:"api-keys" json:"api-keys"`
 
-	// Access holds request authentication provider configuration.
-	Access AccessConfig `yaml:"auth,omitempty" json:"auth,omitempty"`
+	// PassthroughHeaders controls whether upstream response headers are forwarded to downstream clients.
+	// Default is false (disabled).
+	PassthroughHeaders bool `yaml:"passthrough-headers" json:"passthrough-headers"`
 
 	// Streaming configures server-side streaming behavior (keep-alives and safe bootstrap retries).
 	Streaming StreamingConfig `yaml:"streaming" json:"streaming"`
@@ -42,65 +43,3 @@ type StreamingConfig struct {
 	// <= 0 disables bootstrap retries. Default is 0.
 	BootstrapRetries int `yaml:"bootstrap-retries,omitempty" json:"bootstrap-retries,omitempty"`
 }
-
-// AccessConfig groups request authentication providers.
-type AccessConfig struct {
-	// Providers lists configured authentication providers.
-	Providers []AccessProvider `yaml:"providers,omitempty" json:"providers,omitempty"`
-}
-
-// AccessProvider describes a request authentication provider entry.
-type AccessProvider struct {
-	// Name is the instance identifier for the provider.
-	Name string `yaml:"name" json:"name"`
-
-	// Type selects the provider implementation registered via the SDK.
-	Type string `yaml:"type" json:"type"`
-
-	// SDK optionally names a third-party SDK module providing this provider.
-	SDK string `yaml:"sdk,omitempty" json:"sdk,omitempty"`
-
-	// APIKeys lists inline keys for providers that require them.
-	APIKeys []string `yaml:"api-keys,omitempty" json:"api-keys,omitempty"`
-
-	// Config passes provider-specific options to the implementation.
-	Config map[string]any `yaml:"config,omitempty" json:"config,omitempty"`
-}
-
-const (
-	// AccessProviderTypeConfigAPIKey is the built-in provider validating inline API keys.
-	AccessProviderTypeConfigAPIKey = "config-api-key"
-
-	// DefaultAccessProviderName is applied when no provider name is supplied.
-	DefaultAccessProviderName = "config-inline"
-)
-
-// ConfigAPIKeyProvider returns the first inline API key provider if present.
-func (c *SDKConfig) ConfigAPIKeyProvider() *AccessProvider {
-	if c == nil {
-		return nil
-	}
-	for i := range c.Access.Providers {
-		if c.Access.Providers[i].Type == AccessProviderTypeConfigAPIKey {
-			if c.Access.Providers[i].Name == "" {
-				c.Access.Providers[i].Name = DefaultAccessProviderName
-			}
-			return &c.Access.Providers[i]
-		}
-	}
-	return nil
-}
-
-// MakeInlineAPIKeyProvider constructs an inline API key provider configuration.
-// It returns nil when no keys are supplied.
-func MakeInlineAPIKeyProvider(keys []string) *AccessProvider {
-	if len(keys) == 0 {
-		return nil
-	}
-	provider := &AccessProvider{
-		Name:    DefaultAccessProviderName,
-		Type:    AccessProviderTypeConfigAPIKey,
-		APIKeys: append([]string(nil), keys...),
-	}
-	return provider
-}

internal/managementasset/updater.go

@@ -21,6 +21,7 @@ import (
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 	sdkconfig "github.com/router-for-me/CLIProxyAPI/v6/sdk/config"
 	log "github.com/sirupsen/logrus"
+	"golang.org/x/sync/singleflight"
 )
 
 const (
@@ -28,6 +29,7 @@ const (
 	defaultManagementFallbackURL = "https://cpamc.router-for.me/"
 	managementAssetName          = "management.html"
 	httpUserAgent                = "CLIProxyAPI-management-updater"
+	managementSyncMinInterval    = 30 * time.Second
 	updateCheckInterval          = 3 * time.Hour
 )
 
@@ -37,11 +39,10 @@ const ManagementFileName = managementAssetName
 var (
 	lastUpdateCheckMu   sync.Mutex
 	lastUpdateCheckTime time.Time
-
 	currentConfigPtr    atomic.Pointer[config.Config]
-	disableControlPanel atomic.Bool
 	schedulerOnce       sync.Once
 	schedulerConfigPath atomic.Value
+	sfGroup             singleflight.Group
 )
 
 // SetCurrentConfig stores the latest configuration snapshot for management asset decisions.
@@ -50,16 +51,7 @@ func SetCurrentConfig(cfg *config.Config) {
 		currentConfigPtr.Store(nil)
 		return
 	}
-
-	prevDisabled := disableControlPanel.Load()
 	currentConfigPtr.Store(cfg)
-	disableControlPanel.Store(cfg.RemoteManagement.DisableControlPanel)
-
-	if prevDisabled && !cfg.RemoteManagement.DisableControlPanel {
-		lastUpdateCheckMu.Lock()
-		lastUpdateCheckTime = time.Time{}
-		lastUpdateCheckMu.Unlock()
-	}
 }
 
 // StartAutoUpdater launches a background goroutine that periodically ensures the management asset is up to date.
@@ -92,7 +84,7 @@ func runAutoUpdater(ctx context.Context) {
 			log.Debug("management asset auto-updater skipped: config not yet available")
 			return
 		}
-		if disableControlPanel.Load() {
+		if cfg.RemoteManagement.DisableControlPanel {
 			log.Debug("management asset auto-updater skipped: control panel disabled")
 			return
 		}
@@ -181,103 +173,106 @@ func FilePath(configFilePath string) string {
 }
 
 // EnsureLatestManagementHTML checks the latest management.html asset and updates the local copy when needed.
-// The function is designed to run in a background goroutine and will never panic.
-// It enforces a 3-hour rate limit to avoid frequent checks on config/auth file changes.
-func EnsureLatestManagementHTML(ctx context.Context, staticDir string, proxyURL string, panelRepository string) {
+// It coalesces concurrent sync attempts and returns whether the asset exists after the sync attempt.
+func EnsureLatestManagementHTML(ctx context.Context, staticDir string, proxyURL string, panelRepository string) bool {
 	if ctx == nil {
 		ctx = context.Background()
 	}
 
-	if disableControlPanel.Load() {
-		log.Debug("management asset sync skipped: control panel disabled by configuration")
-		return
-	}
-
 	staticDir = strings.TrimSpace(staticDir)
 	if staticDir == "" {
 		log.Debug("management asset sync skipped: empty static directory")
-		return
+		return false
 	}
-
 	localPath := filepath.Join(staticDir, managementAssetName)
-	localFileMissing := false
-	if _, errStat := os.Stat(localPath); errStat != nil {
-		if errors.Is(errStat, os.ErrNotExist) {
-			localFileMissing = true
-		} else {
-			log.WithError(errStat).Debug("failed to stat local management asset")
-		}
-	}
 
-	// Rate limiting: check only once every 3 hours
-	lastUpdateCheckMu.Lock()
-	now := time.Now()
-	timeSinceLastCheck := now.Sub(lastUpdateCheckTime)
-	if timeSinceLastCheck < updateCheckInterval {
+	_, _, _ = sfGroup.Do(localPath, func() (interface{}, error) {
+		lastUpdateCheckMu.Lock()
+		now := time.Now()
+		timeSinceLastAttempt := now.Sub(lastUpdateCheckTime)
+		if !lastUpdateCheckTime.IsZero() && timeSinceLastAttempt < managementSyncMinInterval {
+			lastUpdateCheckMu.Unlock()
+			log.Debugf(
+				"management asset sync skipped by throttle: last attempt %v ago (interval %v)",
+				timeSinceLastAttempt.Round(time.Second),
+				managementSyncMinInterval,
+			)
+			return nil, nil
+		}
+		lastUpdateCheckTime = now
 		lastUpdateCheckMu.Unlock()
-		log.Debugf("management asset update check skipped: last check was %v ago (interval: %v)", timeSinceLastCheck.Round(time.Second), updateCheckInterval)
-		return
-	}
-	lastUpdateCheckTime = now
-	lastUpdateCheckMu.Unlock()
 
-	if errMkdirAll := os.MkdirAll(staticDir, 0o755); errMkdirAll != nil {
-		log.WithError(errMkdirAll).Warn("failed to prepare static directory for management asset")
-		return
-	}
-
-	releaseURL := resolveReleaseURL(panelRepository)
-	client := newHTTPClient(proxyURL)
-
-	localHash, err := fileSHA256(localPath)
-	if err != nil {
-		if !errors.Is(err, os.ErrNotExist) {
-			log.WithError(err).Debug("failed to read local management asset hash")
-		}
-		localHash = ""
-	}
-
-	asset, remoteHash, err := fetchLatestAsset(ctx, client, releaseURL)
-	if err != nil {
-		if localFileMissing {
-			log.WithError(err).Warn("failed to fetch latest management release information, trying fallback page")
-			if ensureFallbackManagementHTML(ctx, client, localPath) {
-				return
+		localFileMissing := false
+		if _, errStat := os.Stat(localPath); errStat != nil {
+			if errors.Is(errStat, os.ErrNotExist) {
+				localFileMissing = true
+			} else {
+				log.WithError(errStat).Debug("failed to stat local management asset")
 			}
-			return
 		}
-		log.WithError(err).Warn("failed to fetch latest management release information")
-		return
-	}
 
-	if remoteHash != "" && localHash != "" && strings.EqualFold(remoteHash, localHash) {
-		log.Debug("management asset is already up to date")
-		return
-	}
+		if errMkdirAll := os.MkdirAll(staticDir, 0o755); errMkdirAll != nil {
+			log.WithError(errMkdirAll).Warn("failed to prepare static directory for management asset")
+			return nil, nil
+		}
 
-	data, downloadedHash, err := downloadAsset(ctx, client, asset.BrowserDownloadURL)
-	if err != nil {
-		if localFileMissing {
-			log.WithError(err).Warn("failed to download management asset, trying fallback page")
-			if ensureFallbackManagementHTML(ctx, client, localPath) {
-				return
+		releaseURL := resolveReleaseURL(panelRepository)
+		client := newHTTPClient(proxyURL)
+
+		localHash, err := fileSHA256(localPath)
+		if err != nil {
+			if !errors.Is(err, os.ErrNotExist) {
+				log.WithError(err).Debug("failed to read local management asset hash")
 			}
-			return
+			localHash = ""
 		}
-		log.WithError(err).Warn("failed to download management asset")
-		return
-	}
 
-	if remoteHash != "" && !strings.EqualFold(remoteHash, downloadedHash) {
-		log.Warnf("remote digest mismatch for management asset: expected %s got %s", remoteHash, downloadedHash)
-	}
+		asset, remoteHash, err := fetchLatestAsset(ctx, client, releaseURL)
+		if err != nil {
+			if localFileMissing {
+				log.WithError(err).Warn("failed to fetch latest management release information, trying fallback page")
+				if ensureFallbackManagementHTML(ctx, client, localPath) {
+					return nil, nil
+				}
+				return nil, nil
+			}
+			log.WithError(err).Warn("failed to fetch latest management release information")
+			return nil, nil
+		}
 
-	if err = atomicWriteFile(localPath, data); err != nil {
-		log.WithError(err).Warn("failed to update management asset on disk")
-		return
-	}
+		if remoteHash != "" && localHash != "" && strings.EqualFold(remoteHash, localHash) {
+			log.Debug("management asset is already up to date")
+			return nil, nil
+		}
 
-	log.Infof("management asset updated successfully (hash=%s)", downloadedHash)
+		data, downloadedHash, err := downloadAsset(ctx, client, asset.BrowserDownloadURL)
+		if err != nil {
+			if localFileMissing {
+				log.WithError(err).Warn("failed to download management asset, trying fallback page")
+				if ensureFallbackManagementHTML(ctx, client, localPath) {
+					return nil, nil
+				}
+				return nil, nil
+			}
+			log.WithError(err).Warn("failed to download management asset")
+			return nil, nil
+		}
+
+		if remoteHash != "" && !strings.EqualFold(remoteHash, downloadedHash) {
+			log.Warnf("remote digest mismatch for management asset: expected %s got %s", remoteHash, downloadedHash)
+		}
+
+		if err = atomicWriteFile(localPath, data); err != nil {
+			log.WithError(err).Warn("failed to update management asset on disk")
+			return nil, nil
+		}
+
+		log.Infof("management asset updated successfully (hash=%s)", downloadedHash)
+		return nil, nil
+	})
+
+	_, err := os.Stat(localPath)
+	return err == nil
 }
 
 func ensureFallbackManagementHTML(ctx context.Context, client *http.Client, localPath string) bool {

internal/misc/credentials.go

@@ -1,6 +1,7 @@
 package misc
 
 import (
+	"encoding/json"
 	"fmt"
 	"path/filepath"
 	"strings"
@@ -24,3 +25,37 @@ func LogSavingCredentials(path string) {
 func LogCredentialSeparator() {
 	log.Debug(credentialSeparator)
 }
+
+// MergeMetadata serializes the source struct into a map and merges the provided metadata into it.
+func MergeMetadata(source any, metadata map[string]any) (map[string]any, error) {
+	var data map[string]any
+
+	// Fast path: if source is already a map, just copy it to avoid mutation of original
+	if srcMap, ok := source.(map[string]any); ok {
+		data = make(map[string]any, len(srcMap)+len(metadata))
+		for k, v := range srcMap {
+			data[k] = v
+		}
+	} else {
+		// Slow path: marshal to JSON and back to map to respect JSON tags
+		temp, err := json.Marshal(source)
+		if err != nil {
+			return nil, fmt.Errorf("failed to marshal source: %w", err)
+		}
+		if err := json.Unmarshal(temp, &data); err != nil {
+			return nil, fmt.Errorf("failed to unmarshal to map: %w", err)
+		}
+	}
+
+	// Merge extra metadata
+	if metadata != nil {
+		if data == nil {
+			data = make(map[string]any)
+		}
+		for k, v := range metadata {
+			data[k] = v
+		}
+	}
+
+	return data, nil
+}

internal/registry/model_definitions.go

@@ -19,6 +19,7 @@ import (
 //   - codex
 //   - qwen
 //   - iflow
+//   - kimi
 //   - antigravity (returns static overrides only)
 func GetStaticModelDefinitionsByChannel(channel string) []*ModelInfo {
 	key := strings.ToLower(strings.TrimSpace(channel))
@@ -39,6 +40,8 @@ func GetStaticModelDefinitionsByChannel(channel string) []*ModelInfo {
 		return GetQwenModels()
 	case "iflow":
 		return GetIFlowModels()
+	case "kimi":
+		return GetKimiModels()
 	case "antigravity":
 		cfg := GetAntigravityModelConfig()
 		if len(cfg) == 0 {
@@ -83,6 +86,7 @@ func LookupStaticModelInfo(modelID string) *ModelInfo {
 		GetOpenAIModels(),
 		GetQwenModels(),
 		GetIFlowModels(),
+		GetKimiModels(),
 	}
 	for _, models := range allModels {
 		for _, m := range models {

internal/registry/model_definitions_static_data.go

@@ -15,7 +15,7 @@ func GetClaudeModels() []*ModelInfo {
 			DisplayName:         "Claude 4.5 Haiku",
 			ContextLength:       200000,
 			MaxCompletionTokens: 64000,
-			// Thinking: not supported for Haiku models
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
 		},
 		{
 			ID:                  "claude-sonnet-4-5-20250929",
@@ -29,15 +29,26 @@ func GetClaudeModels() []*ModelInfo {
 			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
 		},
 		{
-			ID:                  "claude-opus-4-6-20260205",
+			ID:                  "claude-sonnet-4-6",
+			Object:              "model",
+			Created:             1771372800, // 2026-02-17
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.6 Sonnet",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
+		},
+		{
+			ID:                  "claude-opus-4-6",
 			Object:              "model",
 			Created:             1770318000, // 2026-02-05
 			OwnedBy:             "anthropic",
 			Type:                "claude",
 			DisplayName:         "Claude 4.6 Opus",
 			Description:         "Premium model combining maximum intelligence with practical performance",
-			ContextLength:       200000,
-			MaxCompletionTokens: 64000,
+			ContextLength:       1000000,
+			MaxCompletionTokens: 128000,
 			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
 		},
 		{
@@ -173,6 +184,21 @@ func GetGeminiModels() []*ModelInfo {
 			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
 			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}},
 		},
+		{
+			ID:                         "gemini-3.1-pro-preview",
+			Object:                     "model",
+			Created:                    1771459200,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3.1-pro-preview",
+			Version:                    "3.1",
+			DisplayName:                "Gemini 3.1 Pro Preview",
+			Description:                "Gemini 3.1 Pro Preview",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}},
+		},
 		{
 			ID:                         "gemini-3-flash-preview",
 			Object:                     "model",
@@ -283,6 +309,21 @@ func GetGeminiVertexModels() []*ModelInfo {
 			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
 			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"minimal", "low", "medium", "high"}},
 		},
+		{
+			ID:                         "gemini-3.1-pro-preview",
+			Object:                     "model",
+			Created:                    1771459200,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3.1-pro-preview",
+			Version:                    "3.1",
+			DisplayName:                "Gemini 3.1 Pro Preview",
+			Description:                "Gemini 3.1 Pro Preview",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}},
+		},
 		{
 			ID:                         "gemini-3-pro-image-preview",
 			Object:                     "model",
@@ -425,6 +466,21 @@ func GetGeminiCLIModels() []*ModelInfo {
 			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
 			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}},
 		},
+		{
+			ID:                         "gemini-3.1-pro-preview",
+			Object:                     "model",
+			Created:                    1771459200,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3.1-pro-preview",
+			Version:                    "3.1",
+			DisplayName:                "Gemini 3.1 Pro Preview",
+			Description:                "Gemini 3.1 Pro Preview",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}},
+		},
 		{
 			ID:                         "gemini-3-flash-preview",
 			Object:                     "model",
@@ -506,6 +562,21 @@ func GetAIStudioModels() []*ModelInfo {
 			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
 			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
 		},
+		{
+			ID:                         "gemini-3.1-pro-preview",
+			Object:                     "model",
+			Created:                    1771459200,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3.1-pro-preview",
+			Version:                    "3.1",
+			DisplayName:                "Gemini 3.1 Pro Preview",
+			Description:                "Gemini 3.1 Pro Preview",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
+		},
 		{
 			ID:                         "gemini-3-flash-preview",
 			Object:                     "model",
@@ -742,6 +813,20 @@ func GetOpenAIModels() []*ModelInfo {
 			SupportedParameters: []string{"tools"},
 			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high", "xhigh"}},
 		},
+		{
+			ID:                  "gpt-5.3-codex-spark",
+			Object:              "model",
+			Created:             1770912000,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5.3",
+			DisplayName:         "GPT 5.3 Codex Spark",
+			Description:         "Ultra-fast coding model.",
+			ContextLength:       128000,
+			MaxCompletionTokens: 128000,
+			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high", "xhigh"}},
+		},
 	}
 }
 
@@ -774,6 +859,19 @@ func GetQwenModels() []*ModelInfo {
 			MaxCompletionTokens: 2048,
 			SupportedParameters: []string{"temperature", "top_p", "max_tokens", "stream", "stop"},
 		},
+		{
+			ID:                  "coder-model",
+			Object:              "model",
+			Created:             1771171200,
+			OwnedBy:             "qwen",
+			Type:                "qwen",
+			Version:             "3.5",
+			DisplayName:         "Qwen 3.5 Plus",
+			Description:         "efficient hybrid model with leading coding performance",
+			ContextLength:       1048576,
+			MaxCompletionTokens: 65536,
+			SupportedParameters: []string{"temperature", "top_p", "max_tokens", "stream", "stop"},
+		},
 		{
 			ID:                  "vision-model",
 			Object:              "model",
@@ -814,6 +912,7 @@ func GetIFlowModels() []*ModelInfo {
 		{ID: "kimi-k2-0905", DisplayName: "Kimi-K2-Instruct-0905", Description: "Moonshot Kimi K2 instruct 0905", Created: 1757030400},
 		{ID: "glm-4.6", DisplayName: "GLM-4.6", Description: "Zhipu GLM 4.6 general model", Created: 1759190400, Thinking: iFlowThinkingSupport},
 		{ID: "glm-4.7", DisplayName: "GLM-4.7", Description: "Zhipu GLM 4.7 general model", Created: 1766448000, Thinking: iFlowThinkingSupport},
+		{ID: "glm-5", DisplayName: "GLM-5", Description: "Zhipu GLM 5 general model", Created: 1770768000, Thinking: iFlowThinkingSupport},
 		{ID: "kimi-k2", DisplayName: "Kimi-K2", Description: "Moonshot Kimi K2 general model", Created: 1752192000},
 		{ID: "kimi-k2-thinking", DisplayName: "Kimi-K2-Thinking", Description: "Moonshot Kimi K2 thinking model", Created: 1762387200},
 		{ID: "deepseek-v3.2-chat", DisplayName: "DeepSeek-V3.2", Description: "DeepSeek V3.2 Chat", Created: 1764576000},
@@ -828,6 +927,7 @@ func GetIFlowModels() []*ModelInfo {
 		{ID: "qwen3-235b", DisplayName: "Qwen3-235B-A22B", Description: "Qwen3 235B A22B", Created: 1753401600},
 		{ID: "minimax-m2", DisplayName: "MiniMax-M2", Description: "MiniMax M2", Created: 1758672000, Thinking: iFlowThinkingSupport},
 		{ID: "minimax-m2.1", DisplayName: "MiniMax-M2.1", Description: "MiniMax M2.1", Created: 1766448000, Thinking: iFlowThinkingSupport},
+		{ID: "minimax-m2.5", DisplayName: "MiniMax-M2.5", Description: "MiniMax M2.5", Created: 1770825600, Thinking: iFlowThinkingSupport},
 		{ID: "iflow-rome-30ba3b", DisplayName: "iFlow-ROME", Description: "iFlow Rome 30BA3B model", Created: 1736899200},
 		{ID: "kimi-k2.5", DisplayName: "Kimi-K2.5", Description: "Moonshot Kimi K2.5", Created: 1769443200, Thinking: iFlowThinkingSupport},
 	}
@@ -863,12 +963,56 @@ func GetAntigravityModelConfig() map[string]*AntigravityModelConfig {
 		"gemini-2.5-flash-lite":      {Thinking: &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true}},
 		"gemini-3-pro-high":          {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}}},
 		"gemini-3-pro-image":         {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}}},
+		"gemini-3.1-pro-high":        {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"low", "high"}}},
 		"gemini-3-flash":             {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"minimal", "low", "medium", "high"}}},
 		"claude-sonnet-4-5-thinking": {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 64000},
 		"claude-opus-4-5-thinking":   {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 64000},
 		"claude-opus-4-6-thinking":   {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 64000},
 		"claude-sonnet-4-5":          {MaxCompletionTokens: 64000},
+		"claude-sonnet-4-6":          {MaxCompletionTokens: 64000},
+		"claude-sonnet-4-6-thinking": {Thinking: &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: true}, MaxCompletionTokens: 64000},
 		"gpt-oss-120b-medium":        {},
 		"tab_flash_lite_preview":     {},
 	}
 }
+
+// GetKimiModels returns the standard Kimi (Moonshot AI) model definitions
+func GetKimiModels() []*ModelInfo {
+	return []*ModelInfo{
+		{
+			ID:                  "kimi-k2",
+			Object:              "model",
+			Created:             1752192000, // 2025-07-11
+			OwnedBy:             "moonshot",
+			Type:                "kimi",
+			DisplayName:         "Kimi K2",
+			Description:         "Kimi K2 - Moonshot AI's flagship coding model",
+			ContextLength:       131072,
+			MaxCompletionTokens: 32768,
+		},
+		{
+			ID:                  "kimi-k2-thinking",
+			Object:              "model",
+			Created:             1762387200, // 2025-11-06
+			OwnedBy:             "moonshot",
+			Type:                "kimi",
+			DisplayName:         "Kimi K2 Thinking",
+			Description:         "Kimi K2 Thinking - Extended reasoning model",
+			ContextLength:       131072,
+			MaxCompletionTokens: 32768,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 32000, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                  "kimi-k2.5",
+			Object:              "model",
+			Created:             1769472000, // 2026-01-26
+			OwnedBy:             "moonshot",
+			Type:                "kimi",
+			DisplayName:         "Kimi K2.5",
+			Description:         "Kimi K2.5 - Latest Moonshot AI coding model with improved capabilities",
+			ContextLength:       131072,
+			MaxCompletionTokens: 32768,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 32000, ZeroAllowed: true, DynamicAllowed: true},
+		},
+	}
+}

internal/registry/model_registry.go

@@ -596,8 +596,7 @@ func (r *ModelRegistry) SetModelQuotaExceeded(clientID, modelID string) {
 	defer r.mutex.Unlock()
 
 	if registration, exists := r.models[modelID]; exists {
-		now := time.Now()
-		registration.QuotaExceededClients[clientID] = &now
+		registration.QuotaExceededClients[clientID] = new(time.Now())
 		log.Debugf("Marked model %s as quota exceeded for client %s", modelID, clientID)
 	}
 }

internal/runtime/executor/aistudio_executor.go

@@ -164,12 +164,12 @@ func (e *AIStudioExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth,
 	reporter.publish(ctx, parseGeminiUsage(wsResp.Body))
 	var param any
 	out := sdktranslator.TranslateNonStream(ctx, body.toFormat, opts.SourceFormat, req.Model, opts.OriginalRequest, translatedReq, wsResp.Body, &param)
-	resp = cliproxyexecutor.Response{Payload: ensureColonSpacedJSON([]byte(out))}
+	resp = cliproxyexecutor.Response{Payload: ensureColonSpacedJSON([]byte(out)), Headers: wsResp.Headers.Clone()}
 	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) {
+func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
 	if opts.Alt == "responses/compact" {
 		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
 	}
@@ -254,7 +254,6 @@ func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth
 		return nil, statusErr{code: firstEvent.Status, msg: body.String()}
 	}
 	out := make(chan cliproxyexecutor.StreamChunk)
-	stream = out
 	go func(first wsrelay.StreamEvent) {
 		defer close(out)
 		var param any
@@ -318,7 +317,7 @@ func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth
 			}
 		}
 	}(firstEvent)
-	return stream, nil
+	return &cliproxyexecutor.StreamResult{Headers: firstEvent.Headers.Clone(), Chunks: out}, nil
 }
 
 // CountTokens counts tokens for the given request using the AI Studio API.

internal/runtime/executor/antigravity_executor.go

@@ -232,7 +232,7 @@ attemptLoop:
 			reporter.publish(ctx, parseAntigravityUsage(bodyBytes))
 			var param any
 			converted := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, translated, bodyBytes, &param)
-			resp = cliproxyexecutor.Response{Payload: []byte(converted)}
+			resp = cliproxyexecutor.Response{Payload: []byte(converted), Headers: httpResp.Header.Clone()}
 			reporter.ensurePublished(ctx)
 			return resp, nil
 		}
@@ -436,7 +436,7 @@ attemptLoop:
 			reporter.publish(ctx, parseAntigravityUsage(resp.Payload))
 			var param any
 			converted := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, translated, resp.Payload, &param)
-			resp = cliproxyexecutor.Response{Payload: []byte(converted)}
+			resp = cliproxyexecutor.Response{Payload: []byte(converted), Headers: httpResp.Header.Clone()}
 			reporter.ensurePublished(ctx)
 
 			return resp, nil
@@ -645,7 +645,7 @@ func (e *AntigravityExecutor) convertStreamToNonStream(stream []byte) []byte {
 }
 
 // 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) {
+func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
 	if opts.Alt == "responses/compact" {
 		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
 	}
@@ -775,7 +775,6 @@ attemptLoop:
 			}
 
 			out := make(chan cliproxyexecutor.StreamChunk)
-			stream = out
 			go func(resp *http.Response) {
 				defer close(out)
 				defer func() {
@@ -820,7 +819,7 @@ attemptLoop:
 					reporter.ensurePublished(ctx)
 				}
 			}(httpResp)
-			return stream, nil
+			return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
 		}
 
 		switch {
@@ -968,7 +967,7 @@ func (e *AntigravityExecutor) CountTokens(ctx context.Context, auth *cliproxyaut
 		if httpResp.StatusCode >= http.StatusOK && httpResp.StatusCode < http.StatusMultipleChoices {
 			count := gjson.GetBytes(bodyBytes, "totalTokens").Int()
 			translated := sdktranslator.TranslateTokenCount(respCtx, to, from, count, bodyBytes)
-			return cliproxyexecutor.Response{Payload: []byte(translated)}, nil
+			return cliproxyexecutor.Response{Payload: []byte(translated), Headers: httpResp.Header.Clone()}, nil
 		}
 
 		lastStatus = httpResp.StatusCode

internal/runtime/executor/antigravity_executor_buildrequest_test.go

@@ -0,0 +1,159 @@
+package executor
+
+import (
+	"context"
+	"encoding/json"
+	"io"
+	"testing"
+
+	cliproxyauth "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/auth"
+)
+
+func TestAntigravityBuildRequest_SanitizesGeminiToolSchema(t *testing.T) {
+	body := buildRequestBodyFromPayload(t, "gemini-2.5-pro")
+
+	decl := extractFirstFunctionDeclaration(t, body)
+	if _, ok := decl["parametersJsonSchema"]; ok {
+		t.Fatalf("parametersJsonSchema should be renamed to parameters")
+	}
+
+	params, ok := decl["parameters"].(map[string]any)
+	if !ok {
+		t.Fatalf("parameters missing or invalid type")
+	}
+	assertSchemaSanitizedAndPropertyPreserved(t, params)
+}
+
+func TestAntigravityBuildRequest_SanitizesAntigravityToolSchema(t *testing.T) {
+	body := buildRequestBodyFromPayload(t, "claude-opus-4-6")
+
+	decl := extractFirstFunctionDeclaration(t, body)
+	params, ok := decl["parameters"].(map[string]any)
+	if !ok {
+		t.Fatalf("parameters missing or invalid type")
+	}
+	assertSchemaSanitizedAndPropertyPreserved(t, params)
+}
+
+func buildRequestBodyFromPayload(t *testing.T, modelName string) map[string]any {
+	t.Helper()
+
+	executor := &AntigravityExecutor{}
+	auth := &cliproxyauth.Auth{}
+	payload := []byte(`{
+		"request": {
+			"tools": [
+				{
+					"function_declarations": [
+						{
+							"name": "tool_1",
+							"parametersJsonSchema": {
+								"$schema": "http://json-schema.org/draft-07/schema#",
+								"$id": "root-schema",
+								"type": "object",
+								"properties": {
+									"$id": {"type": "string"},
+									"arg": {
+										"type": "object",
+										"prefill": "hello",
+										"properties": {
+											"mode": {
+												"type": "string",
+												"enum": ["a", "b"],
+												"enumTitles": ["A", "B"]
+											}
+										}
+									}
+								},
+								"patternProperties": {
+									"^x-": {"type": "string"}
+								}
+							}
+						}
+					]
+				}
+			]
+		}
+	}`)
+
+	req, err := executor.buildRequest(context.Background(), auth, "token", modelName, payload, false, "", "https://example.com")
+	if err != nil {
+		t.Fatalf("buildRequest error: %v", err)
+	}
+
+	raw, err := io.ReadAll(req.Body)
+	if err != nil {
+		t.Fatalf("read request body error: %v", err)
+	}
+
+	var body map[string]any
+	if err := json.Unmarshal(raw, &body); err != nil {
+		t.Fatalf("unmarshal request body error: %v, body=%s", err, string(raw))
+	}
+	return body
+}
+
+func extractFirstFunctionDeclaration(t *testing.T, body map[string]any) map[string]any {
+	t.Helper()
+
+	request, ok := body["request"].(map[string]any)
+	if !ok {
+		t.Fatalf("request missing or invalid type")
+	}
+	tools, ok := request["tools"].([]any)
+	if !ok || len(tools) == 0 {
+		t.Fatalf("tools missing or empty")
+	}
+	tool, ok := tools[0].(map[string]any)
+	if !ok {
+		t.Fatalf("first tool invalid type")
+	}
+	decls, ok := tool["function_declarations"].([]any)
+	if !ok || len(decls) == 0 {
+		t.Fatalf("function_declarations missing or empty")
+	}
+	decl, ok := decls[0].(map[string]any)
+	if !ok {
+		t.Fatalf("first function declaration invalid type")
+	}
+	return decl
+}
+
+func assertSchemaSanitizedAndPropertyPreserved(t *testing.T, params map[string]any) {
+	t.Helper()
+
+	if _, ok := params["$id"]; ok {
+		t.Fatalf("root $id should be removed from schema")
+	}
+	if _, ok := params["patternProperties"]; ok {
+		t.Fatalf("patternProperties should be removed from schema")
+	}
+
+	props, ok := params["properties"].(map[string]any)
+	if !ok {
+		t.Fatalf("properties missing or invalid type")
+	}
+	if _, ok := props["$id"]; !ok {
+		t.Fatalf("property named $id should be preserved")
+	}
+
+	arg, ok := props["arg"].(map[string]any)
+	if !ok {
+		t.Fatalf("arg property missing or invalid type")
+	}
+	if _, ok := arg["prefill"]; ok {
+		t.Fatalf("prefill should be removed from nested schema")
+	}
+
+	argProps, ok := arg["properties"].(map[string]any)
+	if !ok {
+		t.Fatalf("arg.properties missing or invalid type")
+	}
+	mode, ok := argProps["mode"].(map[string]any)
+	if !ok {
+		t.Fatalf("mode property missing or invalid type")
+	}
+	if _, ok := mode["enumTitles"]; ok {
+		t.Fatalf("enumTitles should be removed from nested schema")
+	}
+}

internal/runtime/executor/claude_executor.go

@@ -9,6 +9,7 @@ import (
 	"fmt"
 	"io"
 	"net/http"
+	"runtime"
 	"strings"
 	"time"
 
@@ -116,7 +117,7 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 
 	// Apply cloaking (system prompt injection, fake user ID, sensitive word obfuscation)
 	// based on client type and configuration.
-	body = applyCloaking(ctx, e.cfg, auth, body, baseModel)
+	body = applyCloaking(ctx, e.cfg, auth, body, baseModel, apiKey)
 
 	requestedModel := payloadRequestedModel(opts, req.Model)
 	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
@@ -134,7 +135,7 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	extraBetas, body = extractAndRemoveBetas(body)
 	bodyForTranslation := body
 	bodyForUpstream := body
-	if isClaudeOAuthToken(apiKey) {
+	if isClaudeOAuthToken(apiKey) && !auth.ToolPrefixDisabled() {
 		bodyForUpstream = applyClaudeToolPrefix(body, claudeToolPrefix)
 	}
 
@@ -143,7 +144,7 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	if err != nil {
 		return resp, err
 	}
-	applyClaudeHeaders(httpReq, auth, apiKey, false, extraBetas)
+	applyClaudeHeaders(httpReq, auth, apiKey, false, extraBetas, e.cfg)
 	var authID, authLabel, authType, authValue string
 	if auth != nil {
 		authID = auth.ID
@@ -208,7 +209,7 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	} else {
 		reporter.publish(ctx, parseClaudeUsage(data))
 	}
-	if isClaudeOAuthToken(apiKey) {
+	if isClaudeOAuthToken(apiKey) && !auth.ToolPrefixDisabled() {
 		data = stripClaudeToolPrefixFromResponse(data, claudeToolPrefix)
 	}
 	var param any
@@ -222,11 +223,11 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 		data,
 		&param,
 	)
-	resp = cliproxyexecutor.Response{Payload: []byte(out)}
+	resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
 	return resp, nil
 }
 
-func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
 	if opts.Alt == "responses/compact" {
 		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
 	}
@@ -257,7 +258,7 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 
 	// Apply cloaking (system prompt injection, fake user ID, sensitive word obfuscation)
 	// based on client type and configuration.
-	body = applyCloaking(ctx, e.cfg, auth, body, baseModel)
+	body = applyCloaking(ctx, e.cfg, auth, body, baseModel, apiKey)
 
 	requestedModel := payloadRequestedModel(opts, req.Model)
 	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
@@ -275,7 +276,7 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	extraBetas, body = extractAndRemoveBetas(body)
 	bodyForTranslation := body
 	bodyForUpstream := body
-	if isClaudeOAuthToken(apiKey) {
+	if isClaudeOAuthToken(apiKey) && !auth.ToolPrefixDisabled() {
 		bodyForUpstream = applyClaudeToolPrefix(body, claudeToolPrefix)
 	}
 
@@ -284,7 +285,7 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	if err != nil {
 		return nil, err
 	}
-	applyClaudeHeaders(httpReq, auth, apiKey, true, extraBetas)
+	applyClaudeHeaders(httpReq, auth, apiKey, true, extraBetas, e.cfg)
 	var authID, authLabel, authType, authValue string
 	if auth != nil {
 		authID = auth.ID
@@ -329,7 +330,6 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 		return nil, err
 	}
 	out := make(chan cliproxyexecutor.StreamChunk)
-	stream = out
 	go func() {
 		defer close(out)
 		defer func() {
@@ -348,7 +348,7 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 				if detail, ok := parseClaudeStreamUsage(line); ok {
 					reporter.publish(ctx, detail)
 				}
-				if isClaudeOAuthToken(apiKey) {
+				if isClaudeOAuthToken(apiKey) && !auth.ToolPrefixDisabled() {
 					line = stripClaudeToolPrefixFromStreamLine(line, claudeToolPrefix)
 				}
 				// Forward the line as-is to preserve SSE format
@@ -375,7 +375,7 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 			if detail, ok := parseClaudeStreamUsage(line); ok {
 				reporter.publish(ctx, detail)
 			}
-			if isClaudeOAuthToken(apiKey) {
+			if isClaudeOAuthToken(apiKey) && !auth.ToolPrefixDisabled() {
 				line = stripClaudeToolPrefixFromStreamLine(line, claudeToolPrefix)
 			}
 			chunks := sdktranslator.TranslateStream(
@@ -398,7 +398,7 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 			out <- cliproxyexecutor.StreamChunk{Err: errScan}
 		}
 	}()
-	return stream, nil
+	return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
 }
 
 func (e *ClaudeExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
@@ -423,7 +423,7 @@ func (e *ClaudeExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
 	// Extract betas from body and convert to header (for count_tokens too)
 	var extraBetas []string
 	extraBetas, body = extractAndRemoveBetas(body)
-	if isClaudeOAuthToken(apiKey) {
+	if isClaudeOAuthToken(apiKey) && !auth.ToolPrefixDisabled() {
 		body = applyClaudeToolPrefix(body, claudeToolPrefix)
 	}
 
@@ -432,7 +432,7 @@ func (e *ClaudeExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
 	if err != nil {
 		return cliproxyexecutor.Response{}, err
 	}
-	applyClaudeHeaders(httpReq, auth, apiKey, false, extraBetas)
+	applyClaudeHeaders(httpReq, auth, apiKey, false, extraBetas, e.cfg)
 	var authID, authLabel, authType, authValue string
 	if auth != nil {
 		authID = auth.ID
@@ -487,7 +487,7 @@ func (e *ClaudeExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
 	appendAPIResponseChunk(ctx, e.cfg, data)
 	count := gjson.GetBytes(data, "input_tokens").Int()
 	out := sdktranslator.TranslateTokenCount(ctx, to, from, count, data)
-	return cliproxyexecutor.Response{Payload: []byte(out)}, nil
+	return cliproxyexecutor.Response{Payload: []byte(out), Headers: resp.Header.Clone()}, nil
 }
 
 func (e *ClaudeExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
@@ -638,7 +638,49 @@ func decodeResponseBody(body io.ReadCloser, contentEncoding string) (io.ReadClos
 	return body, nil
 }
 
-func applyClaudeHeaders(r *http.Request, auth *cliproxyauth.Auth, apiKey string, stream bool, extraBetas []string) {
+// mapStainlessOS maps runtime.GOOS to Stainless SDK OS names.
+func mapStainlessOS() string {
+	switch runtime.GOOS {
+	case "darwin":
+		return "MacOS"
+	case "windows":
+		return "Windows"
+	case "linux":
+		return "Linux"
+	case "freebsd":
+		return "FreeBSD"
+	default:
+		return "Other::" + runtime.GOOS
+	}
+}
+
+// mapStainlessArch maps runtime.GOARCH to Stainless SDK architecture names.
+func mapStainlessArch() string {
+	switch runtime.GOARCH {
+	case "amd64":
+		return "x64"
+	case "arm64":
+		return "arm64"
+	case "386":
+		return "x86"
+	default:
+		return "other::" + runtime.GOARCH
+	}
+}
+
+func applyClaudeHeaders(r *http.Request, auth *cliproxyauth.Auth, apiKey string, stream bool, extraBetas []string, cfg *config.Config) {
+	hdrDefault := func(cfgVal, fallback string) string {
+		if cfgVal != "" {
+			return cfgVal
+		}
+		return fallback
+	}
+
+	var hd config.ClaudeHeaderDefaults
+	if cfg != nil {
+		hd = cfg.ClaudeHeaderDefaults
+	}
+
 	useAPIKey := auth != nil && auth.Attributes != nil && strings.TrimSpace(auth.Attributes["api_key"]) != ""
 	isAnthropicBase := r.URL != nil && strings.EqualFold(r.URL.Scheme, "https") && strings.EqualFold(r.URL.Host, "api.anthropic.com")
 	if isAnthropicBase && useAPIKey {
@@ -685,16 +727,17 @@ func applyClaudeHeaders(r *http.Request, auth *cliproxyauth.Auth, apiKey string,
 	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")
+	// Values below match Claude Code 2.1.44 / @anthropic-ai/sdk 0.74.0 (captured 2026-02-17).
 	misc.EnsureHeader(r.Header, ginHeaders, "X-Stainless-Helper-Method", "stream")
 	misc.EnsureHeader(r.Header, ginHeaders, "X-Stainless-Retry-Count", "0")
-	misc.EnsureHeader(r.Header, ginHeaders, "X-Stainless-Runtime-Version", "v24.3.0")
-	misc.EnsureHeader(r.Header, ginHeaders, "X-Stainless-Package-Version", "0.55.1")
+	misc.EnsureHeader(r.Header, ginHeaders, "X-Stainless-Runtime-Version", hdrDefault(hd.RuntimeVersion, "v24.3.0"))
+	misc.EnsureHeader(r.Header, ginHeaders, "X-Stainless-Package-Version", hdrDefault(hd.PackageVersion, "0.74.0"))
 	misc.EnsureHeader(r.Header, ginHeaders, "X-Stainless-Runtime", "node")
 	misc.EnsureHeader(r.Header, ginHeaders, "X-Stainless-Lang", "js")
-	misc.EnsureHeader(r.Header, ginHeaders, "X-Stainless-Arch", "arm64")
-	misc.EnsureHeader(r.Header, ginHeaders, "X-Stainless-Os", "MacOS")
-	misc.EnsureHeader(r.Header, ginHeaders, "X-Stainless-Timeout", "60")
-	misc.EnsureHeader(r.Header, ginHeaders, "User-Agent", "claude-cli/1.0.83 (external, cli)")
+	misc.EnsureHeader(r.Header, ginHeaders, "X-Stainless-Arch", mapStainlessArch())
+	misc.EnsureHeader(r.Header, ginHeaders, "X-Stainless-Os", mapStainlessOS())
+	misc.EnsureHeader(r.Header, ginHeaders, "X-Stainless-Timeout", hdrDefault(hd.Timeout, "600"))
+	misc.EnsureHeader(r.Header, ginHeaders, "User-Agent", hdrDefault(hd.UserAgent, "claude-cli/2.1.44 (external, sdk-cli)"))
 	r.Header.Set("Connection", "keep-alive")
 	r.Header.Set("Accept-Encoding", "gzip, deflate, br, zstd")
 	if stream {
@@ -702,6 +745,8 @@ func applyClaudeHeaders(r *http.Request, auth *cliproxyauth.Auth, apiKey string,
 	} else {
 		r.Header.Set("Accept", "application/json")
 	}
+	// Keep OS/Arch mapping dynamic (not configurable).
+	// They intentionally continue to derive from runtime.GOOS/runtime.GOARCH.
 	var attrs map[string]string
 	if auth != nil {
 		attrs = auth.Attributes
@@ -753,11 +798,21 @@ func applyClaudeToolPrefix(body []byte, prefix string) []byte {
 		return body
 	}
 
+	// Collect built-in tool names (those with a non-empty "type" field) so we can
+	// skip them consistently in both tools and message history.
+	builtinTools := map[string]bool{}
+	for _, name := range []string{"web_search", "code_execution", "text_editor", "computer"} {
+		builtinTools[name] = true
+	}
+
 	if tools := gjson.GetBytes(body, "tools"); tools.Exists() && tools.IsArray() {
 		tools.ForEach(func(index, tool gjson.Result) bool {
 			// Skip built-in tools (web_search, code_execution, etc.) which have
 			// a "type" field and require their name to remain unchanged.
 			if tool.Get("type").Exists() && tool.Get("type").String() != "" {
+				if n := tool.Get("name").String(); n != "" {
+					builtinTools[n] = true
+				}
 				return true
 			}
 			name := tool.Get("name").String()
@@ -772,7 +827,7 @@ func applyClaudeToolPrefix(body []byte, prefix string) []byte {
 
 	if gjson.GetBytes(body, "tool_choice.type").String() == "tool" {
 		name := gjson.GetBytes(body, "tool_choice.name").String()
-		if name != "" && !strings.HasPrefix(name, prefix) {
+		if name != "" && !strings.HasPrefix(name, prefix) && !builtinTools[name] {
 			body, _ = sjson.SetBytes(body, "tool_choice.name", prefix+name)
 		}
 	}
@@ -784,15 +839,38 @@ func applyClaudeToolPrefix(body []byte, prefix string) []byte {
 				return true
 			}
 			content.ForEach(func(contentIndex, part gjson.Result) bool {
-				if part.Get("type").String() != "tool_use" {
-					return true
+				partType := part.Get("type").String()
+				switch partType {
+				case "tool_use":
+					name := part.Get("name").String()
+					if name == "" || strings.HasPrefix(name, prefix) || builtinTools[name] {
+						return true
+					}
+					path := fmt.Sprintf("messages.%d.content.%d.name", msgIndex.Int(), contentIndex.Int())
+					body, _ = sjson.SetBytes(body, path, prefix+name)
+				case "tool_reference":
+					toolName := part.Get("tool_name").String()
+					if toolName == "" || strings.HasPrefix(toolName, prefix) || builtinTools[toolName] {
+						return true
+					}
+					path := fmt.Sprintf("messages.%d.content.%d.tool_name", msgIndex.Int(), contentIndex.Int())
+					body, _ = sjson.SetBytes(body, path, prefix+toolName)
+				case "tool_result":
+					// Handle nested tool_reference blocks inside tool_result.content[]
+					nestedContent := part.Get("content")
+					if nestedContent.Exists() && nestedContent.IsArray() {
+						nestedContent.ForEach(func(nestedIndex, nestedPart gjson.Result) bool {
+							if nestedPart.Get("type").String() == "tool_reference" {
+								nestedToolName := nestedPart.Get("tool_name").String()
+								if nestedToolName != "" && !strings.HasPrefix(nestedToolName, prefix) && !builtinTools[nestedToolName] {
+									nestedPath := fmt.Sprintf("messages.%d.content.%d.content.%d.tool_name", msgIndex.Int(), contentIndex.Int(), nestedIndex.Int())
+									body, _ = sjson.SetBytes(body, nestedPath, prefix+nestedToolName)
+								}
+							}
+							return true
+						})
+					}
 				}
-				name := part.Get("name").String()
-				if name == "" || strings.HasPrefix(name, prefix) {
-					return true
-				}
-				path := fmt.Sprintf("messages.%d.content.%d.name", msgIndex.Int(), contentIndex.Int())
-				body, _ = sjson.SetBytes(body, path, prefix+name)
 				return true
 			})
 			return true
@@ -811,15 +889,38 @@ func stripClaudeToolPrefixFromResponse(body []byte, prefix string) []byte {
 		return body
 	}
 	content.ForEach(func(index, part gjson.Result) bool {
-		if part.Get("type").String() != "tool_use" {
-			return true
+		partType := part.Get("type").String()
+		switch partType {
+		case "tool_use":
+			name := part.Get("name").String()
+			if !strings.HasPrefix(name, prefix) {
+				return true
+			}
+			path := fmt.Sprintf("content.%d.name", index.Int())
+			body, _ = sjson.SetBytes(body, path, strings.TrimPrefix(name, prefix))
+		case "tool_reference":
+			toolName := part.Get("tool_name").String()
+			if !strings.HasPrefix(toolName, prefix) {
+				return true
+			}
+			path := fmt.Sprintf("content.%d.tool_name", index.Int())
+			body, _ = sjson.SetBytes(body, path, strings.TrimPrefix(toolName, prefix))
+		case "tool_result":
+			// Handle nested tool_reference blocks inside tool_result.content[]
+			nestedContent := part.Get("content")
+			if nestedContent.Exists() && nestedContent.IsArray() {
+				nestedContent.ForEach(func(nestedIndex, nestedPart gjson.Result) bool {
+					if nestedPart.Get("type").String() == "tool_reference" {
+						nestedToolName := nestedPart.Get("tool_name").String()
+						if strings.HasPrefix(nestedToolName, prefix) {
+							nestedPath := fmt.Sprintf("content.%d.content.%d.tool_name", index.Int(), nestedIndex.Int())
+							body, _ = sjson.SetBytes(body, nestedPath, strings.TrimPrefix(nestedToolName, prefix))
+						}
+					}
+					return true
+				})
+			}
 		}
-		name := part.Get("name").String()
-		if !strings.HasPrefix(name, prefix) {
-			return true
-		}
-		path := fmt.Sprintf("content.%d.name", index.Int())
-		body, _ = sjson.SetBytes(body, path, strings.TrimPrefix(name, prefix))
 		return true
 	})
 	return body
@@ -834,15 +935,34 @@ func stripClaudeToolPrefixFromStreamLine(line []byte, prefix string) []byte {
 		return line
 	}
 	contentBlock := gjson.GetBytes(payload, "content_block")
-	if !contentBlock.Exists() || contentBlock.Get("type").String() != "tool_use" {
+	if !contentBlock.Exists() {
 		return line
 	}
-	name := contentBlock.Get("name").String()
-	if !strings.HasPrefix(name, prefix) {
-		return line
-	}
-	updated, err := sjson.SetBytes(payload, "content_block.name", strings.TrimPrefix(name, prefix))
-	if err != nil {
+
+	blockType := contentBlock.Get("type").String()
+	var updated []byte
+	var err error
+
+	switch blockType {
+	case "tool_use":
+		name := contentBlock.Get("name").String()
+		if !strings.HasPrefix(name, prefix) {
+			return line
+		}
+		updated, err = sjson.SetBytes(payload, "content_block.name", strings.TrimPrefix(name, prefix))
+		if err != nil {
+			return line
+		}
+	case "tool_reference":
+		toolName := contentBlock.Get("tool_name").String()
+		if !strings.HasPrefix(toolName, prefix) {
+			return line
+		}
+		updated, err = sjson.SetBytes(payload, "content_block.tool_name", strings.TrimPrefix(toolName, prefix))
+		if err != nil {
+			return line
+		}
+	default:
 		return line
 	}
 
@@ -862,10 +982,10 @@ func getClientUserAgent(ctx context.Context) string {
 }
 
 // getCloakConfigFromAuth extracts cloak configuration from auth attributes.
-// Returns (cloakMode, strictMode, sensitiveWords).
-func getCloakConfigFromAuth(auth *cliproxyauth.Auth) (string, bool, []string) {
+// Returns (cloakMode, strictMode, sensitiveWords, cacheUserID).
+func getCloakConfigFromAuth(auth *cliproxyauth.Auth) (string, bool, []string, bool) {
 	if auth == nil || auth.Attributes == nil {
-		return "auto", false, nil
+		return "auto", false, nil, false
 	}
 
 	cloakMode := auth.Attributes["cloak_mode"]
@@ -883,7 +1003,9 @@ func getCloakConfigFromAuth(auth *cliproxyauth.Auth) (string, bool, []string) {
 		}
 	}
 
-	return cloakMode, strictMode, sensitiveWords
+	cacheUserID := strings.EqualFold(strings.TrimSpace(auth.Attributes["cloak_cache_user_id"]), "true")
+
+	return cloakMode, strictMode, sensitiveWords, cacheUserID
 }
 
 // resolveClaudeKeyCloakConfig finds the matching ClaudeKey config and returns its CloakConfig.
@@ -916,16 +1038,24 @@ func resolveClaudeKeyCloakConfig(cfg *config.Config, auth *cliproxyauth.Auth) *c
 }
 
 // injectFakeUserID generates and injects a fake user ID into the request metadata.
-func injectFakeUserID(payload []byte) []byte {
+// When useCache is false, a new user ID is generated for every call.
+func injectFakeUserID(payload []byte, apiKey string, useCache bool) []byte {
+	generateID := func() string {
+		if useCache {
+			return cachedUserID(apiKey)
+		}
+		return generateFakeUserID()
+	}
+
 	metadata := gjson.GetBytes(payload, "metadata")
 	if !metadata.Exists() {
-		payload, _ = sjson.SetBytes(payload, "metadata.user_id", generateFakeUserID())
+		payload, _ = sjson.SetBytes(payload, "metadata.user_id", generateID())
 		return payload
 	}
 
 	existingUserID := gjson.GetBytes(payload, "metadata.user_id").String()
 	if existingUserID == "" || !isValidUserID(existingUserID) {
-		payload, _ = sjson.SetBytes(payload, "metadata.user_id", generateFakeUserID())
+		payload, _ = sjson.SetBytes(payload, "metadata.user_id", generateID())
 	}
 	return payload
 }
@@ -962,7 +1092,7 @@ func checkSystemInstructionsWithMode(payload []byte, strictMode bool) []byte {
 
 // applyCloaking applies cloaking transformations to the payload based on config and client.
 // Cloaking includes: system prompt injection, fake user ID, and sensitive word obfuscation.
-func applyCloaking(ctx context.Context, cfg *config.Config, auth *cliproxyauth.Auth, payload []byte, model string) []byte {
+func applyCloaking(ctx context.Context, cfg *config.Config, auth *cliproxyauth.Auth, payload []byte, model string, apiKey string) []byte {
 	clientUserAgent := getClientUserAgent(ctx)
 
 	// Get cloak config from ClaudeKey configuration
@@ -972,16 +1102,20 @@ func applyCloaking(ctx context.Context, cfg *config.Config, auth *cliproxyauth.A
 	var cloakMode string
 	var strictMode bool
 	var sensitiveWords []string
+	var cacheUserID bool
 
 	if cloakCfg != nil {
 		cloakMode = cloakCfg.Mode
 		strictMode = cloakCfg.StrictMode
 		sensitiveWords = cloakCfg.SensitiveWords
+		if cloakCfg.CacheUserID != nil {
+			cacheUserID = *cloakCfg.CacheUserID
+		}
 	}
 
 	// Fallback to auth attributes if no config found
 	if cloakMode == "" {
-		attrMode, attrStrict, attrWords := getCloakConfigFromAuth(auth)
+		attrMode, attrStrict, attrWords, attrCache := getCloakConfigFromAuth(auth)
 		cloakMode = attrMode
 		if !strictMode {
 			strictMode = attrStrict
@@ -989,6 +1123,12 @@ func applyCloaking(ctx context.Context, cfg *config.Config, auth *cliproxyauth.A
 		if len(sensitiveWords) == 0 {
 			sensitiveWords = attrWords
 		}
+		if cloakCfg == nil || cloakCfg.CacheUserID == nil {
+			cacheUserID = attrCache
+		}
+	} else if cloakCfg == nil || cloakCfg.CacheUserID == nil {
+		_, _, _, attrCache := getCloakConfigFromAuth(auth)
+		cacheUserID = attrCache
 	}
 
 	// Determine if cloaking should be applied
@@ -1002,7 +1142,7 @@ func applyCloaking(ctx context.Context, cfg *config.Config, auth *cliproxyauth.A
 	}
 
 	// Inject fake user ID
-	payload = injectFakeUserID(payload)
+	payload = injectFakeUserID(payload, apiKey, cacheUserID)
 
 	// Apply sensitive word obfuscation
 	if len(sensitiveWords) > 0 {

internal/runtime/executor/claude_executor_test.go

@@ -2,9 +2,18 @@ package executor
 
 import (
 	"bytes"
+	"context"
+	"io"
+	"net/http"
+	"net/http/httptest"
 	"testing"
 
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	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"
 	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
 )
 
 func TestApplyClaudeToolPrefix(t *testing.T) {
@@ -25,6 +34,18 @@ func TestApplyClaudeToolPrefix(t *testing.T) {
 	}
 }
 
+func TestApplyClaudeToolPrefix_WithToolReference(t *testing.T) {
+	input := []byte(`{"tools":[{"name":"alpha"}],"messages":[{"role":"user","content":[{"type":"tool_reference","tool_name":"beta"},{"type":"tool_reference","tool_name":"proxy_gamma"}]}]}`)
+	out := applyClaudeToolPrefix(input, "proxy_")
+
+	if got := gjson.GetBytes(out, "messages.0.content.0.tool_name").String(); got != "proxy_beta" {
+		t.Fatalf("messages.0.content.0.tool_name = %q, want %q", got, "proxy_beta")
+	}
+	if got := gjson.GetBytes(out, "messages.0.content.1.tool_name").String(); got != "proxy_gamma" {
+		t.Fatalf("messages.0.content.1.tool_name = %q, want %q", got, "proxy_gamma")
+	}
+}
+
 func TestApplyClaudeToolPrefix_SkipsBuiltinTools(t *testing.T) {
 	input := []byte(`{"tools":[{"type":"web_search_20250305","name":"web_search"},{"name":"my_custom_tool","input_schema":{"type":"object"}}]}`)
 	out := applyClaudeToolPrefix(input, "proxy_")
@@ -37,6 +58,97 @@ func TestApplyClaudeToolPrefix_SkipsBuiltinTools(t *testing.T) {
 	}
 }
 
+func TestApplyClaudeToolPrefix_BuiltinToolSkipped(t *testing.T) {
+	body := []byte(`{
+		"tools": [
+			{"type": "web_search_20250305", "name": "web_search", "max_uses": 5},
+			{"name": "Read"}
+		],
+		"messages": [
+			{"role": "user", "content": [
+				{"type": "tool_use", "name": "web_search", "id": "ws1", "input": {}},
+				{"type": "tool_use", "name": "Read", "id": "r1", "input": {}}
+			]}
+		]
+	}`)
+	out := applyClaudeToolPrefix(body, "proxy_")
+
+	if got := gjson.GetBytes(out, "tools.0.name").String(); got != "web_search" {
+		t.Fatalf("tools.0.name = %q, want %q", got, "web_search")
+	}
+	if got := gjson.GetBytes(out, "messages.0.content.0.name").String(); got != "web_search" {
+		t.Fatalf("messages.0.content.0.name = %q, want %q", got, "web_search")
+	}
+	if got := gjson.GetBytes(out, "tools.1.name").String(); got != "proxy_Read" {
+		t.Fatalf("tools.1.name = %q, want %q", got, "proxy_Read")
+	}
+	if got := gjson.GetBytes(out, "messages.0.content.1.name").String(); got != "proxy_Read" {
+		t.Fatalf("messages.0.content.1.name = %q, want %q", got, "proxy_Read")
+	}
+}
+
+func TestApplyClaudeToolPrefix_KnownBuiltinInHistoryOnly(t *testing.T) {
+	body := []byte(`{
+		"tools": [
+			{"name": "Read"}
+		],
+		"messages": [
+			{"role": "user", "content": [
+				{"type": "tool_use", "name": "web_search", "id": "ws1", "input": {}}
+			]}
+		]
+	}`)
+	out := applyClaudeToolPrefix(body, "proxy_")
+
+	if got := gjson.GetBytes(out, "messages.0.content.0.name").String(); got != "web_search" {
+		t.Fatalf("messages.0.content.0.name = %q, want %q", got, "web_search")
+	}
+	if got := gjson.GetBytes(out, "tools.0.name").String(); got != "proxy_Read" {
+		t.Fatalf("tools.0.name = %q, want %q", got, "proxy_Read")
+	}
+}
+
+func TestApplyClaudeToolPrefix_CustomToolsPrefixed(t *testing.T) {
+	body := []byte(`{
+		"tools": [{"name": "Read"}, {"name": "Write"}],
+		"messages": [
+			{"role": "user", "content": [
+				{"type": "tool_use", "name": "Read", "id": "r1", "input": {}},
+				{"type": "tool_use", "name": "Write", "id": "w1", "input": {}}
+			]}
+		]
+	}`)
+	out := applyClaudeToolPrefix(body, "proxy_")
+
+	if got := gjson.GetBytes(out, "tools.0.name").String(); got != "proxy_Read" {
+		t.Fatalf("tools.0.name = %q, want %q", got, "proxy_Read")
+	}
+	if got := gjson.GetBytes(out, "tools.1.name").String(); got != "proxy_Write" {
+		t.Fatalf("tools.1.name = %q, want %q", got, "proxy_Write")
+	}
+	if got := gjson.GetBytes(out, "messages.0.content.0.name").String(); got != "proxy_Read" {
+		t.Fatalf("messages.0.content.0.name = %q, want %q", got, "proxy_Read")
+	}
+	if got := gjson.GetBytes(out, "messages.0.content.1.name").String(); got != "proxy_Write" {
+		t.Fatalf("messages.0.content.1.name = %q, want %q", got, "proxy_Write")
+	}
+}
+
+func TestApplyClaudeToolPrefix_ToolChoiceBuiltin(t *testing.T) {
+	body := []byte(`{
+		"tools": [
+			{"type": "web_search_20250305", "name": "web_search"},
+			{"name": "Read"}
+		],
+		"tool_choice": {"type": "tool", "name": "web_search"}
+	}`)
+	out := applyClaudeToolPrefix(body, "proxy_")
+
+	if got := gjson.GetBytes(out, "tool_choice.name").String(); got != "web_search" {
+		t.Fatalf("tool_choice.name = %q, want %q", got, "web_search")
+	}
+}
+
 func TestStripClaudeToolPrefixFromResponse(t *testing.T) {
 	input := []byte(`{"content":[{"type":"tool_use","name":"proxy_alpha","id":"t1","input":{}},{"type":"tool_use","name":"bravo","id":"t2","input":{}}]}`)
 	out := stripClaudeToolPrefixFromResponse(input, "proxy_")
@@ -49,6 +161,18 @@ func TestStripClaudeToolPrefixFromResponse(t *testing.T) {
 	}
 }
 
+func TestStripClaudeToolPrefixFromResponse_WithToolReference(t *testing.T) {
+	input := []byte(`{"content":[{"type":"tool_reference","tool_name":"proxy_alpha"},{"type":"tool_reference","tool_name":"bravo"}]}`)
+	out := stripClaudeToolPrefixFromResponse(input, "proxy_")
+
+	if got := gjson.GetBytes(out, "content.0.tool_name").String(); got != "alpha" {
+		t.Fatalf("content.0.tool_name = %q, want %q", got, "alpha")
+	}
+	if got := gjson.GetBytes(out, "content.1.tool_name").String(); got != "bravo" {
+		t.Fatalf("content.1.tool_name = %q, want %q", got, "bravo")
+	}
+}
+
 func TestStripClaudeToolPrefixFromStreamLine(t *testing.T) {
 	line := []byte(`data: {"type":"content_block_start","content_block":{"type":"tool_use","name":"proxy_alpha","id":"t1"},"index":0}`)
 	out := stripClaudeToolPrefixFromStreamLine(line, "proxy_")
@@ -61,3 +185,166 @@ func TestStripClaudeToolPrefixFromStreamLine(t *testing.T) {
 		t.Fatalf("content_block.name = %q, want %q", got, "alpha")
 	}
 }
+
+func TestStripClaudeToolPrefixFromStreamLine_WithToolReference(t *testing.T) {
+	line := []byte(`data: {"type":"content_block_start","content_block":{"type":"tool_reference","tool_name":"proxy_beta"},"index":0}`)
+	out := stripClaudeToolPrefixFromStreamLine(line, "proxy_")
+
+	payload := bytes.TrimSpace(out)
+	if bytes.HasPrefix(payload, []byte("data:")) {
+		payload = bytes.TrimSpace(payload[len("data:"):])
+	}
+	if got := gjson.GetBytes(payload, "content_block.tool_name").String(); got != "beta" {
+		t.Fatalf("content_block.tool_name = %q, want %q", got, "beta")
+	}
+}
+
+func TestApplyClaudeToolPrefix_NestedToolReference(t *testing.T) {
+	input := []byte(`{"messages":[{"role":"user","content":[{"type":"tool_result","tool_use_id":"toolu_123","content":[{"type":"tool_reference","tool_name":"mcp__nia__manage_resource"}]}]}]}`)
+	out := applyClaudeToolPrefix(input, "proxy_")
+	got := gjson.GetBytes(out, "messages.0.content.0.content.0.tool_name").String()
+	if got != "proxy_mcp__nia__manage_resource" {
+		t.Fatalf("nested tool_reference tool_name = %q, want %q", got, "proxy_mcp__nia__manage_resource")
+	}
+}
+
+func TestClaudeExecutor_ReusesUserIDAcrossModelsWhenCacheEnabled(t *testing.T) {
+	resetUserIDCache()
+
+	var userIDs []string
+	var requestModels []string
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		body, _ := io.ReadAll(r.Body)
+		userID := gjson.GetBytes(body, "metadata.user_id").String()
+		model := gjson.GetBytes(body, "model").String()
+		userIDs = append(userIDs, userID)
+		requestModels = append(requestModels, model)
+		t.Logf("HTTP Server received request: model=%s, user_id=%s, url=%s", model, userID, r.URL.String())
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write([]byte(`{"id":"msg_1","type":"message","model":"claude-3-5-sonnet","role":"assistant","content":[{"type":"text","text":"ok"}],"usage":{"input_tokens":1,"output_tokens":1}}`))
+	}))
+	defer server.Close()
+
+	t.Logf("End-to-end test: Fake HTTP server started at %s", server.URL)
+
+	cacheEnabled := true
+	executor := NewClaudeExecutor(&config.Config{
+		ClaudeKey: []config.ClaudeKey{
+			{
+				APIKey:  "key-123",
+				BaseURL: server.URL,
+				Cloak: &config.CloakConfig{
+					CacheUserID: &cacheEnabled,
+				},
+			},
+		},
+	})
+	auth := &cliproxyauth.Auth{Attributes: map[string]string{
+		"api_key":  "key-123",
+		"base_url": server.URL,
+	}}
+
+	payload := []byte(`{"messages":[{"role":"user","content":[{"type":"text","text":"hi"}]}]}`)
+	models := []string{"claude-3-5-sonnet", "claude-3-5-haiku"}
+	for _, model := range models {
+		t.Logf("Sending request for model: %s", model)
+		modelPayload, _ := sjson.SetBytes(payload, "model", model)
+		if _, err := executor.Execute(context.Background(), auth, cliproxyexecutor.Request{
+			Model:   model,
+			Payload: modelPayload,
+		}, cliproxyexecutor.Options{
+			SourceFormat: sdktranslator.FromString("claude"),
+		}); err != nil {
+			t.Fatalf("Execute(%s) error: %v", model, err)
+		}
+	}
+
+	if len(userIDs) != 2 {
+		t.Fatalf("expected 2 requests, got %d", len(userIDs))
+	}
+	if userIDs[0] == "" || userIDs[1] == "" {
+		t.Fatal("expected user_id to be populated")
+	}
+	t.Logf("user_id[0] (model=%s): %s", requestModels[0], userIDs[0])
+	t.Logf("user_id[1] (model=%s): %s", requestModels[1], userIDs[1])
+	if userIDs[0] != userIDs[1] {
+		t.Fatalf("expected user_id to be reused across models, got %q and %q", userIDs[0], userIDs[1])
+	}
+	if !isValidUserID(userIDs[0]) {
+		t.Fatalf("user_id %q is not valid", userIDs[0])
+	}
+	t.Logf("✓ End-to-end test passed: Same user_id (%s) was used for both models", userIDs[0])
+}
+
+func TestClaudeExecutor_GeneratesNewUserIDByDefault(t *testing.T) {
+	resetUserIDCache()
+
+	var userIDs []string
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		body, _ := io.ReadAll(r.Body)
+		userIDs = append(userIDs, gjson.GetBytes(body, "metadata.user_id").String())
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write([]byte(`{"id":"msg_1","type":"message","model":"claude-3-5-sonnet","role":"assistant","content":[{"type":"text","text":"ok"}],"usage":{"input_tokens":1,"output_tokens":1}}`))
+	}))
+	defer server.Close()
+
+	executor := NewClaudeExecutor(&config.Config{})
+	auth := &cliproxyauth.Auth{Attributes: map[string]string{
+		"api_key":  "key-123",
+		"base_url": server.URL,
+	}}
+
+	payload := []byte(`{"messages":[{"role":"user","content":[{"type":"text","text":"hi"}]}]}`)
+
+	for i := 0; i < 2; i++ {
+		if _, err := executor.Execute(context.Background(), auth, cliproxyexecutor.Request{
+			Model:   "claude-3-5-sonnet",
+			Payload: payload,
+		}, cliproxyexecutor.Options{
+			SourceFormat: sdktranslator.FromString("claude"),
+		}); err != nil {
+			t.Fatalf("Execute call %d error: %v", i, err)
+		}
+	}
+
+	if len(userIDs) != 2 {
+		t.Fatalf("expected 2 requests, got %d", len(userIDs))
+	}
+	if userIDs[0] == "" || userIDs[1] == "" {
+		t.Fatal("expected user_id to be populated")
+	}
+	if userIDs[0] == userIDs[1] {
+		t.Fatalf("expected user_id to change when caching is not enabled, got identical values %q", userIDs[0])
+	}
+	if !isValidUserID(userIDs[0]) || !isValidUserID(userIDs[1]) {
+		t.Fatalf("user_ids should be valid, got %q and %q", userIDs[0], userIDs[1])
+	}
+}
+
+func TestStripClaudeToolPrefixFromResponse_NestedToolReference(t *testing.T) {
+	input := []byte(`{"content":[{"type":"tool_result","tool_use_id":"toolu_123","content":[{"type":"tool_reference","tool_name":"proxy_mcp__nia__manage_resource"}]}]}`)
+	out := stripClaudeToolPrefixFromResponse(input, "proxy_")
+	got := gjson.GetBytes(out, "content.0.content.0.tool_name").String()
+	if got != "mcp__nia__manage_resource" {
+		t.Fatalf("nested tool_reference tool_name = %q, want %q", got, "mcp__nia__manage_resource")
+	}
+}
+
+func TestApplyClaudeToolPrefix_NestedToolReferenceWithStringContent(t *testing.T) {
+	// tool_result.content can be a string - should not be processed
+	input := []byte(`{"messages":[{"role":"user","content":[{"type":"tool_result","tool_use_id":"toolu_123","content":"plain string result"}]}]}`)
+	out := applyClaudeToolPrefix(input, "proxy_")
+	got := gjson.GetBytes(out, "messages.0.content.0.content").String()
+	if got != "plain string result" {
+		t.Fatalf("string content should remain unchanged = %q", got)
+	}
+}
+
+func TestApplyClaudeToolPrefix_SkipsBuiltinToolReference(t *testing.T) {
+	input := []byte(`{"tools":[{"type":"web_search_20250305","name":"web_search"}],"messages":[{"role":"user","content":[{"type":"tool_result","tool_use_id":"t1","content":[{"type":"tool_reference","tool_name":"web_search"}]}]}]}`)
+	out := applyClaudeToolPrefix(input, "proxy_")
+	got := gjson.GetBytes(out, "messages.0.content.0.content.0.tool_name").String()
+	if got != "web_search" {
+		t.Fatalf("built-in tool_reference should not be prefixed, got %q", got)
+	}
+}

internal/runtime/executor/codex_executor.go

@@ -27,6 +27,11 @@ import (
 	"github.com/google/uuid"
 )
 
+const (
+	codexClientVersion = "0.101.0"
+	codexUserAgent     = "codex_cli_rs/0.101.0 (Mac OS 26.0.1; arm64) Apple_Terminal/464"
+)
+
 var dataTag = []byte("data:")
 
 // CodexExecutor is a stateless executor for Codex (OpenAI Responses API entrypoint).
@@ -178,7 +183,7 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 
 		var param any
 		out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, originalPayload, body, line, &param)
-		resp = cliproxyexecutor.Response{Payload: []byte(out)}
+		resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
 		return resp, nil
 	}
 	err = statusErr{code: 408, msg: "stream error: stream disconnected before completion: stream closed before response.completed"}
@@ -268,11 +273,11 @@ func (e *CodexExecutor) executeCompact(ctx context.Context, auth *cliproxyauth.A
 	reporter.ensurePublished(ctx)
 	var param any
 	out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, originalPayload, body, data, &param)
-	resp = cliproxyexecutor.Response{Payload: []byte(out)}
+	resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
 	return resp, nil
 }
 
-func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
 	if opts.Alt == "responses/compact" {
 		return nil, statusErr{code: http.StatusBadRequest, msg: "streaming not supported for /responses/compact"}
 	}
@@ -357,7 +362,6 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 		return nil, err
 	}
 	out := make(chan cliproxyexecutor.StreamChunk)
-	stream = out
 	go func() {
 		defer close(out)
 		defer func() {
@@ -392,7 +396,7 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 			out <- cliproxyexecutor.StreamChunk{Err: errScan}
 		}
 	}()
-	return stream, nil
+	return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
 }
 
 func (e *CodexExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
@@ -637,10 +641,9 @@ func applyCodexHeaders(r *http.Request, auth *cliproxyauth.Auth, token string, s
 		ginHeaders = ginCtx.Request.Header
 	}
 
-	misc.EnsureHeader(r.Header, ginHeaders, "Version", "0.21.0")
-	misc.EnsureHeader(r.Header, ginHeaders, "Openai-Beta", "responses=experimental")
+	misc.EnsureHeader(r.Header, ginHeaders, "Version", codexClientVersion)
 	misc.EnsureHeader(r.Header, ginHeaders, "Session_id", uuid.NewString())
-	misc.EnsureHeader(r.Header, ginHeaders, "User-Agent", "codex_cli_rs/0.50.0 (Mac OS 26.0.1; arm64) Apple_Terminal/464")
+	misc.EnsureHeader(r.Header, ginHeaders, "User-Agent", codexUserAgent)
 
 	if stream {
 		r.Header.Set("Accept", "text/event-stream")

internal/runtime/executor/gemini_cli_executor.go

@@ -225,7 +225,7 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
 			reporter.publish(ctx, parseGeminiCLIUsage(data))
 			var param any
 			out := sdktranslator.TranslateNonStream(respCtx, to, from, attemptModel, opts.OriginalRequest, payload, data, &param)
-			resp = cliproxyexecutor.Response{Payload: []byte(out)}
+			resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
 			return resp, nil
 		}
 
@@ -256,7 +256,7 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
 }
 
 // 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) {
+func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
 	if opts.Alt == "responses/compact" {
 		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
 	}
@@ -382,7 +382,6 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 		}
 
 		out := make(chan cliproxyexecutor.StreamChunk)
-		stream = out
 		go func(resp *http.Response, reqBody []byte, attemptModel string) {
 			defer close(out)
 			defer func() {
@@ -441,7 +440,7 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 			}
 		}(httpResp, append([]byte(nil), payload...), attemptModel)
 
-		return stream, nil
+		return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
 	}
 
 	if len(lastBody) > 0 {
@@ -546,7 +545,7 @@ func (e *GeminiCLIExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.
 		if resp.StatusCode >= 200 && resp.StatusCode < 300 {
 			count := gjson.GetBytes(data, "totalTokens").Int()
 			translated := sdktranslator.TranslateTokenCount(respCtx, to, from, count, data)
-			return cliproxyexecutor.Response{Payload: []byte(translated)}, nil
+			return cliproxyexecutor.Response{Payload: []byte(translated), Headers: resp.Header.Clone()}, nil
 		}
 		lastStatus = resp.StatusCode
 		lastBody = append([]byte(nil), data...)
@@ -899,8 +898,7 @@ func parseRetryDelay(errorBody []byte) (*time.Duration, error) {
 		if matches := re.FindStringSubmatch(message); len(matches) > 1 {
 			seconds, err := strconv.Atoi(matches[1])
 			if err == nil {
-				duration := time.Duration(seconds) * time.Second
-				return &duration, nil
+				return new(time.Duration(seconds) * time.Second), nil
 			}
 		}
 	}

internal/runtime/executor/gemini_executor.go

@@ -205,12 +205,12 @@ func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	reporter.publish(ctx, parseGeminiUsage(data))
 	var param any
 	out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, body, data, &param)
-	resp = cliproxyexecutor.Response{Payload: []byte(out)}
+	resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
 	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) {
+func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
 	if opts.Alt == "responses/compact" {
 		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
 	}
@@ -298,7 +298,6 @@ func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 		return nil, err
 	}
 	out := make(chan cliproxyexecutor.StreamChunk)
-	stream = out
 	go func() {
 		defer close(out)
 		defer func() {
@@ -335,7 +334,7 @@ func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 			out <- cliproxyexecutor.StreamChunk{Err: errScan}
 		}
 	}()
-	return stream, nil
+	return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
 }
 
 // CountTokens counts tokens for the given request using the Gemini API.
@@ -416,7 +415,7 @@ func (e *GeminiExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
 
 	count := gjson.GetBytes(data, "totalTokens").Int()
 	translated := sdktranslator.TranslateTokenCount(respCtx, to, from, count, data)
-	return cliproxyexecutor.Response{Payload: []byte(translated)}, nil
+	return cliproxyexecutor.Response{Payload: []byte(translated), Headers: resp.Header.Clone()}, nil
 }
 
 // Refresh refreshes the authentication credentials (no-op for Gemini API key).

internal/runtime/executor/gemini_vertex_executor.go

@@ -253,7 +253,7 @@ func (e *GeminiVertexExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 }
 
 // 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) {
+func (e *GeminiVertexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (*cliproxyexecutor.StreamResult, error) {
 	if opts.Alt == "responses/compact" {
 		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
 	}
@@ -419,7 +419,7 @@ func (e *GeminiVertexExecutor) executeWithServiceAccount(ctx context.Context, au
 	to := sdktranslator.FromString("gemini")
 	var param any
 	out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, body, data, &param)
-	resp = cliproxyexecutor.Response{Payload: []byte(out)}
+	resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
 	return resp, nil
 }
 
@@ -524,12 +524,12 @@ func (e *GeminiVertexExecutor) executeWithAPIKey(ctx context.Context, auth *clip
 	reporter.publish(ctx, parseGeminiUsage(data))
 	var param any
 	out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, body, data, &param)
-	resp = cliproxyexecutor.Response{Payload: []byte(out)}
+	resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
 	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) {
+func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, projectID, location string, saJSON []byte) (_ *cliproxyexecutor.StreamResult, err error) {
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	reporter := newUsageReporter(ctx, e.Identifier(), baseModel, auth)
@@ -618,7 +618,6 @@ func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Conte
 	}
 
 	out := make(chan cliproxyexecutor.StreamChunk)
-	stream = out
 	go func() {
 		defer close(out)
 		defer func() {
@@ -650,11 +649,11 @@ func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Conte
 			out <- cliproxyexecutor.StreamChunk{Err: errScan}
 		}
 	}()
-	return stream, nil
+	return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
 }
 
 // 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) {
+func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, apiKey, baseURL string) (_ *cliproxyexecutor.StreamResult, err error) {
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	reporter := newUsageReporter(ctx, e.Identifier(), baseModel, auth)
@@ -743,7 +742,6 @@ func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth
 	}
 
 	out := make(chan cliproxyexecutor.StreamChunk)
-	stream = out
 	go func() {
 		defer close(out)
 		defer func() {
@@ -775,7 +773,7 @@ func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth
 			out <- cliproxyexecutor.StreamChunk{Err: errScan}
 		}
 	}()
-	return stream, nil
+	return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
 }
 
 // countTokensWithServiceAccount counts tokens using service account credentials.
@@ -859,7 +857,7 @@ func (e *GeminiVertexExecutor) countTokensWithServiceAccount(ctx context.Context
 	appendAPIResponseChunk(ctx, e.cfg, data)
 	count := gjson.GetBytes(data, "totalTokens").Int()
 	out := sdktranslator.TranslateTokenCount(ctx, to, from, count, data)
-	return cliproxyexecutor.Response{Payload: []byte(out)}, nil
+	return cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}, nil
 }
 
 // countTokensWithAPIKey handles token counting using API key credentials.
@@ -943,7 +941,7 @@ func (e *GeminiVertexExecutor) countTokensWithAPIKey(ctx context.Context, auth *
 	appendAPIResponseChunk(ctx, e.cfg, data)
 	count := gjson.GetBytes(data, "totalTokens").Int()
 	out := sdktranslator.TranslateTokenCount(ctx, to, from, count, data)
-	return cliproxyexecutor.Response{Payload: []byte(out)}, nil
+	return cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}, nil
 }
 
 // vertexCreds extracts project, location and raw service account JSON from auth metadata.

internal/runtime/executor/iflow_executor.go

@@ -4,12 +4,16 @@ import (
 	"bufio"
 	"bytes"
 	"context"
+	"crypto/hmac"
+	"crypto/sha256"
+	"encoding/hex"
 	"fmt"
 	"io"
 	"net/http"
 	"strings"
 	"time"
 
+	"github.com/google/uuid"
 	iflowauth "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/iflow"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/thinking"
@@ -165,12 +169,12 @@ func (e *IFlowExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	// Note: TranslateNonStream uses req.Model (original with suffix) to preserve
 	// the original model name in the response for client compatibility.
 	out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, body, data, &param)
-	resp = cliproxyexecutor.Response{Payload: []byte(out)}
+	resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
 	return resp, nil
 }
 
 // ExecuteStream performs a streaming chat completion request.
-func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
 	if opts.Alt == "responses/compact" {
 		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
 	}
@@ -258,7 +262,6 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 	}
 
 	out := make(chan cliproxyexecutor.StreamChunk)
-	stream = out
 	go func() {
 		defer close(out)
 		defer func() {
@@ -290,7 +293,7 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 		reporter.ensurePublished(ctx)
 	}()
 
-	return stream, nil
+	return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
 }
 
 func (e *IFlowExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
@@ -453,6 +456,20 @@ func applyIFlowHeaders(r *http.Request, apiKey string, stream bool) {
 	r.Header.Set("Content-Type", "application/json")
 	r.Header.Set("Authorization", "Bearer "+apiKey)
 	r.Header.Set("User-Agent", iflowUserAgent)
+
+	// Generate session-id
+	sessionID := "session-" + generateUUID()
+	r.Header.Set("session-id", sessionID)
+
+	// Generate timestamp and signature
+	timestamp := time.Now().UnixMilli()
+	r.Header.Set("x-iflow-timestamp", fmt.Sprintf("%d", timestamp))
+
+	signature := createIFlowSignature(iflowUserAgent, sessionID, timestamp, apiKey)
+	if signature != "" {
+		r.Header.Set("x-iflow-signature", signature)
+	}
+
 	if stream {
 		r.Header.Set("Accept", "text/event-stream")
 	} else {
@@ -460,6 +477,23 @@ func applyIFlowHeaders(r *http.Request, apiKey string, stream bool) {
 	}
 }
 
+// createIFlowSignature generates HMAC-SHA256 signature for iFlow API requests.
+// The signature payload format is: userAgent:sessionId:timestamp
+func createIFlowSignature(userAgent, sessionID string, timestamp int64, apiKey string) string {
+	if apiKey == "" {
+		return ""
+	}
+	payload := fmt.Sprintf("%s:%s:%d", userAgent, sessionID, timestamp)
+	h := hmac.New(sha256.New, []byte(apiKey))
+	h.Write([]byte(payload))
+	return hex.EncodeToString(h.Sum(nil))
+}
+
+// generateUUID generates a random UUID v4 string.
+func generateUUID() string {
+	return uuid.New().String()
+}
+
 func iflowCreds(a *cliproxyauth.Auth) (apiKey, baseURL string) {
 	if a == nil {
 		return "", ""

internal/runtime/executor/kimi_executor.go

@@ -0,0 +1,617 @@
+package executor
+
+import (
+	"bufio"
+	"bytes"
+	"context"
+	"fmt"
+	"io"
+	"net/http"
+	"os"
+	"path/filepath"
+	"runtime"
+	"strings"
+	"time"
+
+	kimiauth "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/kimi"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/thinking"
+	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"
+)
+
+// KimiExecutor is a stateless executor for Kimi API using OpenAI-compatible chat completions.
+type KimiExecutor struct {
+	ClaudeExecutor
+	cfg *config.Config
+}
+
+// NewKimiExecutor creates a new Kimi executor.
+func NewKimiExecutor(cfg *config.Config) *KimiExecutor { return &KimiExecutor{cfg: cfg} }
+
+// Identifier returns the executor identifier.
+func (e *KimiExecutor) Identifier() string { return "kimi" }
+
+// PrepareRequest injects Kimi credentials into the outgoing HTTP request.
+func (e *KimiExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth) error {
+	if req == nil {
+		return nil
+	}
+	token := kimiCreds(auth)
+	if strings.TrimSpace(token) != "" {
+		req.Header.Set("Authorization", "Bearer "+token)
+	}
+	return nil
+}
+
+// HttpRequest injects Kimi credentials into the request and executes it.
+func (e *KimiExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) {
+	if req == nil {
+		return nil, fmt.Errorf("kimi executor: request is nil")
+	}
+	if ctx == nil {
+		ctx = req.Context()
+	}
+	httpReq := req.WithContext(ctx)
+	if err := e.PrepareRequest(httpReq, auth); err != nil {
+		return nil, err
+	}
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	return httpClient.Do(httpReq)
+}
+
+// Execute performs a non-streaming chat completion request to Kimi.
+func (e *KimiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
+	from := opts.SourceFormat
+	if from.String() == "claude" {
+		auth.Attributes["base_url"] = kimiauth.KimiAPIBaseURL
+		return e.ClaudeExecutor.Execute(ctx, auth, req, opts)
+	}
+
+	baseModel := thinking.ParseSuffix(req.Model).ModelName
+
+	token := kimiCreds(auth)
+
+	reporter := newUsageReporter(ctx, e.Identifier(), baseModel, auth)
+	defer reporter.trackFailure(ctx, &err)
+
+	to := sdktranslator.FromString("openai")
+	originalPayloadSource := req.Payload
+	if len(opts.OriginalRequest) > 0 {
+		originalPayloadSource = opts.OriginalRequest
+	}
+	originalPayload := bytes.Clone(originalPayloadSource)
+	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+
+	// Strip kimi- prefix for upstream API
+	upstreamModel := stripKimiPrefix(baseModel)
+	body, err = sjson.SetBytes(body, "model", upstreamModel)
+	if err != nil {
+		return resp, fmt.Errorf("kimi executor: failed to set model in payload: %w", err)
+	}
+
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), "kimi", e.Identifier())
+	if err != nil {
+		return resp, err
+	}
+
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
+	body, err = normalizeKimiToolMessageLinks(body)
+	if err != nil {
+		return resp, err
+	}
+
+	url := kimiauth.KimiAPIBaseURL + "/v1/chat/completions"
+	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
+	if err != nil {
+		return resp, err
+	}
+	applyKimiHeadersWithAuth(httpReq, token, false, 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, err := httpClient.Do(httpReq)
+	if err != nil {
+		recordAPIResponseError(ctx, e.cfg, err)
+		return resp, err
+	}
+	defer func() {
+		if errClose := httpResp.Body.Close(); errClose != nil {
+			log.Errorf("kimi 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)
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		err = statusErr{code: httpResp.StatusCode, msg: string(b)}
+		return resp, err
+	}
+	data, err := io.ReadAll(httpResp.Body)
+	if err != nil {
+		recordAPIResponseError(ctx, e.cfg, err)
+		return resp, err
+	}
+	appendAPIResponseChunk(ctx, e.cfg, data)
+	reporter.publish(ctx, parseOpenAIUsage(data))
+	var param any
+	// Note: TranslateNonStream uses req.Model (original with suffix) to preserve
+	// the original model name in the response for client compatibility.
+	out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, body, data, &param)
+	resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
+	return resp, nil
+}
+
+// ExecuteStream performs a streaming chat completion request to Kimi.
+func (e *KimiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
+	from := opts.SourceFormat
+	if from.String() == "claude" {
+		auth.Attributes["base_url"] = kimiauth.KimiAPIBaseURL
+		return e.ClaudeExecutor.ExecuteStream(ctx, auth, req, opts)
+	}
+
+	baseModel := thinking.ParseSuffix(req.Model).ModelName
+	token := kimiCreds(auth)
+
+	reporter := newUsageReporter(ctx, e.Identifier(), baseModel, auth)
+	defer reporter.trackFailure(ctx, &err)
+
+	to := sdktranslator.FromString("openai")
+	originalPayloadSource := req.Payload
+	if len(opts.OriginalRequest) > 0 {
+		originalPayloadSource = opts.OriginalRequest
+	}
+	originalPayload := bytes.Clone(originalPayloadSource)
+	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
+
+	// Strip kimi- prefix for upstream API
+	upstreamModel := stripKimiPrefix(baseModel)
+	body, err = sjson.SetBytes(body, "model", upstreamModel)
+	if err != nil {
+		return nil, fmt.Errorf("kimi executor: failed to set model in payload: %w", err)
+	}
+
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), "kimi", e.Identifier())
+	if err != nil {
+		return nil, err
+	}
+
+	body, err = sjson.SetBytes(body, "stream_options.include_usage", true)
+	if err != nil {
+		return nil, fmt.Errorf("kimi executor: failed to set stream_options in payload: %w", err)
+	}
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
+	body, err = normalizeKimiToolMessageLinks(body)
+	if err != nil {
+		return nil, err
+	}
+
+	url := kimiauth.KimiAPIBaseURL + "/v1/chat/completions"
+	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
+	if err != nil {
+		return nil, err
+	}
+	applyKimiHeadersWithAuth(httpReq, token, true, 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, err := httpClient.Do(httpReq)
+	if err != nil {
+		recordAPIResponseError(ctx, e.cfg, err)
+		return nil, err
+	}
+	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)
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		if errClose := httpResp.Body.Close(); errClose != nil {
+			log.Errorf("kimi executor: close response body error: %v", errClose)
+		}
+		err = statusErr{code: httpResp.StatusCode, msg: string(b)}
+		return nil, err
+	}
+	out := make(chan cliproxyexecutor.StreamChunk)
+	go func() {
+		defer close(out)
+		defer func() {
+			if errClose := httpResp.Body.Close(); errClose != nil {
+				log.Errorf("kimi executor: close response body error: %v", errClose)
+			}
+		}()
+		scanner := bufio.NewScanner(httpResp.Body)
+		scanner.Buffer(nil, 1_048_576) // 1MB
+		var param any
+		for scanner.Scan() {
+			line := scanner.Bytes()
+			appendAPIResponseChunk(ctx, e.cfg, line)
+			if detail, ok := parseOpenAIStreamUsage(line); ok {
+				reporter.publish(ctx, detail)
+			}
+			chunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, opts.OriginalRequest, body, bytes.Clone(line), &param)
+			for i := range chunks {
+				out <- cliproxyexecutor.StreamChunk{Payload: []byte(chunks[i])}
+			}
+		}
+		doneChunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, opts.OriginalRequest, body, []byte("[DONE]"), &param)
+		for i := range doneChunks {
+			out <- cliproxyexecutor.StreamChunk{Payload: []byte(doneChunks[i])}
+		}
+		if errScan := scanner.Err(); errScan != nil {
+			recordAPIResponseError(ctx, e.cfg, errScan)
+			reporter.publishFailure(ctx)
+			out <- cliproxyexecutor.StreamChunk{Err: errScan}
+		}
+	}()
+	return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
+}
+
+// CountTokens estimates token count for Kimi requests.
+func (e *KimiExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
+	auth.Attributes["base_url"] = kimiauth.KimiAPIBaseURL
+	return e.ClaudeExecutor.CountTokens(ctx, auth, req, opts)
+}
+
+func normalizeKimiToolMessageLinks(body []byte) ([]byte, error) {
+	if len(body) == 0 || !gjson.ValidBytes(body) {
+		return body, nil
+	}
+
+	messages := gjson.GetBytes(body, "messages")
+	if !messages.Exists() || !messages.IsArray() {
+		return body, nil
+	}
+
+	out := body
+	pending := make([]string, 0)
+	patched := 0
+	patchedReasoning := 0
+	ambiguous := 0
+	latestReasoning := ""
+	hasLatestReasoning := false
+
+	removePending := func(id string) {
+		for idx := range pending {
+			if pending[idx] != id {
+				continue
+			}
+			pending = append(pending[:idx], pending[idx+1:]...)
+			return
+		}
+	}
+
+	msgs := messages.Array()
+	for msgIdx := range msgs {
+		msg := msgs[msgIdx]
+		role := strings.TrimSpace(msg.Get("role").String())
+		switch role {
+		case "assistant":
+			reasoning := msg.Get("reasoning_content")
+			if reasoning.Exists() {
+				reasoningText := reasoning.String()
+				if strings.TrimSpace(reasoningText) != "" {
+					latestReasoning = reasoningText
+					hasLatestReasoning = true
+				}
+			}
+
+			toolCalls := msg.Get("tool_calls")
+			if !toolCalls.Exists() || !toolCalls.IsArray() || len(toolCalls.Array()) == 0 {
+				continue
+			}
+
+			if !reasoning.Exists() || strings.TrimSpace(reasoning.String()) == "" {
+				reasoningText := fallbackAssistantReasoning(msg, hasLatestReasoning, latestReasoning)
+				path := fmt.Sprintf("messages.%d.reasoning_content", msgIdx)
+				next, err := sjson.SetBytes(out, path, reasoningText)
+				if err != nil {
+					return body, fmt.Errorf("kimi executor: failed to set assistant reasoning_content: %w", err)
+				}
+				out = next
+				patchedReasoning++
+			}
+
+			for _, tc := range toolCalls.Array() {
+				id := strings.TrimSpace(tc.Get("id").String())
+				if id == "" {
+					continue
+				}
+				pending = append(pending, id)
+			}
+		case "tool":
+			toolCallID := strings.TrimSpace(msg.Get("tool_call_id").String())
+			if toolCallID == "" {
+				toolCallID = strings.TrimSpace(msg.Get("call_id").String())
+				if toolCallID != "" {
+					path := fmt.Sprintf("messages.%d.tool_call_id", msgIdx)
+					next, err := sjson.SetBytes(out, path, toolCallID)
+					if err != nil {
+						return body, fmt.Errorf("kimi executor: failed to set tool_call_id from call_id: %w", err)
+					}
+					out = next
+					patched++
+				}
+			}
+			if toolCallID == "" {
+				if len(pending) == 1 {
+					toolCallID = pending[0]
+					path := fmt.Sprintf("messages.%d.tool_call_id", msgIdx)
+					next, err := sjson.SetBytes(out, path, toolCallID)
+					if err != nil {
+						return body, fmt.Errorf("kimi executor: failed to infer tool_call_id: %w", err)
+					}
+					out = next
+					patched++
+				} else if len(pending) > 1 {
+					ambiguous++
+				}
+			}
+			if toolCallID != "" {
+				removePending(toolCallID)
+			}
+		}
+	}
+
+	if patched > 0 || patchedReasoning > 0 {
+		log.WithFields(log.Fields{
+			"patched_tool_messages":      patched,
+			"patched_reasoning_messages": patchedReasoning,
+		}).Debug("kimi executor: normalized tool message fields")
+	}
+	if ambiguous > 0 {
+		log.WithFields(log.Fields{
+			"ambiguous_tool_messages": ambiguous,
+			"pending_tool_calls":      len(pending),
+		}).Warn("kimi executor: tool messages missing tool_call_id with ambiguous candidates")
+	}
+
+	return out, nil
+}
+
+func fallbackAssistantReasoning(msg gjson.Result, hasLatest bool, latest string) string {
+	if hasLatest && strings.TrimSpace(latest) != "" {
+		return latest
+	}
+
+	content := msg.Get("content")
+	if content.Type == gjson.String {
+		if text := strings.TrimSpace(content.String()); text != "" {
+			return text
+		}
+	}
+	if content.IsArray() {
+		parts := make([]string, 0, len(content.Array()))
+		for _, item := range content.Array() {
+			text := strings.TrimSpace(item.Get("text").String())
+			if text == "" {
+				continue
+			}
+			parts = append(parts, text)
+		}
+		if len(parts) > 0 {
+			return strings.Join(parts, "\n")
+		}
+	}
+
+	return "[reasoning unavailable]"
+}
+
+// Refresh refreshes the Kimi token using the refresh token.
+func (e *KimiExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
+	log.Debugf("kimi executor: refresh called")
+	if auth == nil {
+		return nil, fmt.Errorf("kimi executor: auth is nil")
+	}
+	// Expect refresh_token in metadata for OAuth-based accounts
+	var refreshToken string
+	if auth.Metadata != nil {
+		if v, ok := auth.Metadata["refresh_token"].(string); ok && strings.TrimSpace(v) != "" {
+			refreshToken = v
+		}
+	}
+	if strings.TrimSpace(refreshToken) == "" {
+		// Nothing to refresh
+		return auth, nil
+	}
+
+	client := kimiauth.NewDeviceFlowClientWithDeviceID(e.cfg, resolveKimiDeviceID(auth))
+	td, err := client.RefreshToken(ctx, refreshToken)
+	if err != nil {
+		return nil, err
+	}
+	if auth.Metadata == nil {
+		auth.Metadata = make(map[string]any)
+	}
+	auth.Metadata["access_token"] = td.AccessToken
+	if td.RefreshToken != "" {
+		auth.Metadata["refresh_token"] = td.RefreshToken
+	}
+	if td.ExpiresAt > 0 {
+		exp := time.Unix(td.ExpiresAt, 0).UTC().Format(time.RFC3339)
+		auth.Metadata["expired"] = exp
+	}
+	auth.Metadata["type"] = "kimi"
+	now := time.Now().Format(time.RFC3339)
+	auth.Metadata["last_refresh"] = now
+	return auth, nil
+}
+
+// applyKimiHeaders sets required headers for Kimi API requests.
+// Headers match kimi-cli client for compatibility.
+func applyKimiHeaders(r *http.Request, token string, stream bool) {
+	r.Header.Set("Content-Type", "application/json")
+	r.Header.Set("Authorization", "Bearer "+token)
+	// Match kimi-cli headers exactly
+	r.Header.Set("User-Agent", "KimiCLI/1.10.6")
+	r.Header.Set("X-Msh-Platform", "kimi_cli")
+	r.Header.Set("X-Msh-Version", "1.10.6")
+	r.Header.Set("X-Msh-Device-Name", getKimiHostname())
+	r.Header.Set("X-Msh-Device-Model", getKimiDeviceModel())
+	r.Header.Set("X-Msh-Device-Id", getKimiDeviceID())
+	if stream {
+		r.Header.Set("Accept", "text/event-stream")
+		return
+	}
+	r.Header.Set("Accept", "application/json")
+}
+
+func resolveKimiDeviceIDFromAuth(auth *cliproxyauth.Auth) string {
+	if auth == nil || auth.Metadata == nil {
+		return ""
+	}
+
+	deviceIDRaw, ok := auth.Metadata["device_id"]
+	if !ok {
+		return ""
+	}
+
+	deviceID, ok := deviceIDRaw.(string)
+	if !ok {
+		return ""
+	}
+
+	return strings.TrimSpace(deviceID)
+}
+
+func resolveKimiDeviceIDFromStorage(auth *cliproxyauth.Auth) string {
+	if auth == nil {
+		return ""
+	}
+
+	storage, ok := auth.Storage.(*kimiauth.KimiTokenStorage)
+	if !ok || storage == nil {
+		return ""
+	}
+
+	return strings.TrimSpace(storage.DeviceID)
+}
+
+func resolveKimiDeviceID(auth *cliproxyauth.Auth) string {
+	deviceID := resolveKimiDeviceIDFromAuth(auth)
+	if deviceID != "" {
+		return deviceID
+	}
+	return resolveKimiDeviceIDFromStorage(auth)
+}
+
+func applyKimiHeadersWithAuth(r *http.Request, token string, stream bool, auth *cliproxyauth.Auth) {
+	applyKimiHeaders(r, token, stream)
+
+	if deviceID := resolveKimiDeviceID(auth); deviceID != "" {
+		r.Header.Set("X-Msh-Device-Id", deviceID)
+	}
+}
+
+// getKimiHostname returns the machine hostname.
+func getKimiHostname() string {
+	hostname, err := os.Hostname()
+	if err != nil {
+		return "unknown"
+	}
+	return hostname
+}
+
+// getKimiDeviceModel returns a device model string matching kimi-cli format.
+func getKimiDeviceModel() string {
+	return fmt.Sprintf("%s %s", runtime.GOOS, runtime.GOARCH)
+}
+
+// getKimiDeviceID returns a stable device ID, matching kimi-cli storage location.
+func getKimiDeviceID() string {
+	homeDir, err := os.UserHomeDir()
+	if err != nil {
+		return "cli-proxy-api-device"
+	}
+	// Check kimi-cli's device_id location first (platform-specific)
+	var kimiShareDir string
+	switch runtime.GOOS {
+	case "darwin":
+		kimiShareDir = filepath.Join(homeDir, "Library", "Application Support", "kimi")
+	case "windows":
+		appData := os.Getenv("APPDATA")
+		if appData == "" {
+			appData = filepath.Join(homeDir, "AppData", "Roaming")
+		}
+		kimiShareDir = filepath.Join(appData, "kimi")
+	default: // linux and other unix-like
+		kimiShareDir = filepath.Join(homeDir, ".local", "share", "kimi")
+	}
+	deviceIDPath := filepath.Join(kimiShareDir, "device_id")
+	if data, err := os.ReadFile(deviceIDPath); err == nil {
+		return strings.TrimSpace(string(data))
+	}
+	return "cli-proxy-api-device"
+}
+
+// kimiCreds extracts the access token from auth.
+func kimiCreds(a *cliproxyauth.Auth) (token string) {
+	if a == nil {
+		return ""
+	}
+	// Check metadata first (OAuth flow stores tokens here)
+	if a.Metadata != nil {
+		if v, ok := a.Metadata["access_token"].(string); ok && strings.TrimSpace(v) != "" {
+			return v
+		}
+	}
+	// Fallback to attributes (API key style)
+	if a.Attributes != nil {
+		if v := a.Attributes["access_token"]; v != "" {
+			return v
+		}
+		if v := a.Attributes["api_key"]; v != "" {
+			return v
+		}
+	}
+	return ""
+}
+
+// stripKimiPrefix removes the "kimi-" prefix from model names for the upstream API.
+func stripKimiPrefix(model string) string {
+	model = strings.TrimSpace(model)
+	if strings.HasPrefix(strings.ToLower(model), "kimi-") {
+		return model[5:]
+	}
+	return model
+}

internal/runtime/executor/kimi_executor_test.go

@@ -0,0 +1,205 @@
+package executor
+
+import (
+	"testing"
+
+	"github.com/tidwall/gjson"
+)
+
+func TestNormalizeKimiToolMessageLinks_UsesCallIDFallback(t *testing.T) {
+	body := []byte(`{
+		"messages":[
+			{"role":"assistant","tool_calls":[{"id":"list_directory:1","type":"function","function":{"name":"list_directory","arguments":"{}"}}]},
+			{"role":"tool","call_id":"list_directory:1","content":"[]"}
+		]
+	}`)
+
+	out, err := normalizeKimiToolMessageLinks(body)
+	if err != nil {
+		t.Fatalf("normalizeKimiToolMessageLinks() error = %v", err)
+	}
+
+	got := gjson.GetBytes(out, "messages.1.tool_call_id").String()
+	if got != "list_directory:1" {
+		t.Fatalf("messages.1.tool_call_id = %q, want %q", got, "list_directory:1")
+	}
+}
+
+func TestNormalizeKimiToolMessageLinks_InferSinglePendingID(t *testing.T) {
+	body := []byte(`{
+		"messages":[
+			{"role":"assistant","tool_calls":[{"id":"call_123","type":"function","function":{"name":"read_file","arguments":"{}"}}]},
+			{"role":"tool","content":"file-content"}
+		]
+	}`)
+
+	out, err := normalizeKimiToolMessageLinks(body)
+	if err != nil {
+		t.Fatalf("normalizeKimiToolMessageLinks() error = %v", err)
+	}
+
+	got := gjson.GetBytes(out, "messages.1.tool_call_id").String()
+	if got != "call_123" {
+		t.Fatalf("messages.1.tool_call_id = %q, want %q", got, "call_123")
+	}
+}
+
+func TestNormalizeKimiToolMessageLinks_AmbiguousMissingIDIsNotInferred(t *testing.T) {
+	body := []byte(`{
+		"messages":[
+			{"role":"assistant","tool_calls":[
+				{"id":"call_1","type":"function","function":{"name":"list_directory","arguments":"{}"}},
+				{"id":"call_2","type":"function","function":{"name":"read_file","arguments":"{}"}}
+			]},
+			{"role":"tool","content":"result-without-id"}
+		]
+	}`)
+
+	out, err := normalizeKimiToolMessageLinks(body)
+	if err != nil {
+		t.Fatalf("normalizeKimiToolMessageLinks() error = %v", err)
+	}
+
+	if gjson.GetBytes(out, "messages.1.tool_call_id").Exists() {
+		t.Fatalf("messages.1.tool_call_id should be absent for ambiguous case, got %q", gjson.GetBytes(out, "messages.1.tool_call_id").String())
+	}
+}
+
+func TestNormalizeKimiToolMessageLinks_PreservesExistingToolCallID(t *testing.T) {
+	body := []byte(`{
+		"messages":[
+			{"role":"assistant","tool_calls":[{"id":"call_1","type":"function","function":{"name":"list_directory","arguments":"{}"}}]},
+			{"role":"tool","tool_call_id":"call_1","call_id":"different-id","content":"result"}
+		]
+	}`)
+
+	out, err := normalizeKimiToolMessageLinks(body)
+	if err != nil {
+		t.Fatalf("normalizeKimiToolMessageLinks() error = %v", err)
+	}
+
+	got := gjson.GetBytes(out, "messages.1.tool_call_id").String()
+	if got != "call_1" {
+		t.Fatalf("messages.1.tool_call_id = %q, want %q", got, "call_1")
+	}
+}
+
+func TestNormalizeKimiToolMessageLinks_InheritsPreviousReasoningForAssistantToolCalls(t *testing.T) {
+	body := []byte(`{
+		"messages":[
+			{"role":"assistant","content":"plan","reasoning_content":"previous reasoning"},
+			{"role":"assistant","tool_calls":[{"id":"call_1","type":"function","function":{"name":"list_directory","arguments":"{}"}}]}
+		]
+	}`)
+
+	out, err := normalizeKimiToolMessageLinks(body)
+	if err != nil {
+		t.Fatalf("normalizeKimiToolMessageLinks() error = %v", err)
+	}
+
+	got := gjson.GetBytes(out, "messages.1.reasoning_content").String()
+	if got != "previous reasoning" {
+		t.Fatalf("messages.1.reasoning_content = %q, want %q", got, "previous reasoning")
+	}
+}
+
+func TestNormalizeKimiToolMessageLinks_InsertsFallbackReasoningWhenMissing(t *testing.T) {
+	body := []byte(`{
+		"messages":[
+			{"role":"assistant","tool_calls":[{"id":"call_1","type":"function","function":{"name":"list_directory","arguments":"{}"}}]}
+		]
+	}`)
+
+	out, err := normalizeKimiToolMessageLinks(body)
+	if err != nil {
+		t.Fatalf("normalizeKimiToolMessageLinks() error = %v", err)
+	}
+
+	reasoning := gjson.GetBytes(out, "messages.0.reasoning_content")
+	if !reasoning.Exists() {
+		t.Fatalf("messages.0.reasoning_content should exist")
+	}
+	if reasoning.String() != "[reasoning unavailable]" {
+		t.Fatalf("messages.0.reasoning_content = %q, want %q", reasoning.String(), "[reasoning unavailable]")
+	}
+}
+
+func TestNormalizeKimiToolMessageLinks_UsesContentAsReasoningFallback(t *testing.T) {
+	body := []byte(`{
+		"messages":[
+			{"role":"assistant","content":[{"type":"text","text":"first line"},{"type":"text","text":"second line"}],"tool_calls":[{"id":"call_1","type":"function","function":{"name":"list_directory","arguments":"{}"}}]}
+		]
+	}`)
+
+	out, err := normalizeKimiToolMessageLinks(body)
+	if err != nil {
+		t.Fatalf("normalizeKimiToolMessageLinks() error = %v", err)
+	}
+
+	got := gjson.GetBytes(out, "messages.0.reasoning_content").String()
+	if got != "first line\nsecond line" {
+		t.Fatalf("messages.0.reasoning_content = %q, want %q", got, "first line\nsecond line")
+	}
+}
+
+func TestNormalizeKimiToolMessageLinks_ReplacesEmptyReasoningContent(t *testing.T) {
+	body := []byte(`{
+		"messages":[
+			{"role":"assistant","content":"assistant summary","tool_calls":[{"id":"call_1","type":"function","function":{"name":"list_directory","arguments":"{}"}}],"reasoning_content":""}
+		]
+	}`)
+
+	out, err := normalizeKimiToolMessageLinks(body)
+	if err != nil {
+		t.Fatalf("normalizeKimiToolMessageLinks() error = %v", err)
+	}
+
+	got := gjson.GetBytes(out, "messages.0.reasoning_content").String()
+	if got != "assistant summary" {
+		t.Fatalf("messages.0.reasoning_content = %q, want %q", got, "assistant summary")
+	}
+}
+
+func TestNormalizeKimiToolMessageLinks_PreservesExistingAssistantReasoning(t *testing.T) {
+	body := []byte(`{
+		"messages":[
+			{"role":"assistant","tool_calls":[{"id":"call_1","type":"function","function":{"name":"list_directory","arguments":"{}"}}],"reasoning_content":"keep me"}
+		]
+	}`)
+
+	out, err := normalizeKimiToolMessageLinks(body)
+	if err != nil {
+		t.Fatalf("normalizeKimiToolMessageLinks() error = %v", err)
+	}
+
+	got := gjson.GetBytes(out, "messages.0.reasoning_content").String()
+	if got != "keep me" {
+		t.Fatalf("messages.0.reasoning_content = %q, want %q", got, "keep me")
+	}
+}
+
+func TestNormalizeKimiToolMessageLinks_RepairsIDsAndReasoningTogether(t *testing.T) {
+	body := []byte(`{
+		"messages":[
+			{"role":"assistant","tool_calls":[{"id":"call_1","type":"function","function":{"name":"list_directory","arguments":"{}"}}],"reasoning_content":"r1"},
+			{"role":"tool","call_id":"call_1","content":"[]"},
+			{"role":"assistant","tool_calls":[{"id":"call_2","type":"function","function":{"name":"read_file","arguments":"{}"}}]},
+			{"role":"tool","call_id":"call_2","content":"file"}
+		]
+	}`)
+
+	out, err := normalizeKimiToolMessageLinks(body)
+	if err != nil {
+		t.Fatalf("normalizeKimiToolMessageLinks() error = %v", err)
+	}
+
+	if got := gjson.GetBytes(out, "messages.1.tool_call_id").String(); got != "call_1" {
+		t.Fatalf("messages.1.tool_call_id = %q, want %q", got, "call_1")
+	}
+	if got := gjson.GetBytes(out, "messages.3.tool_call_id").String(); got != "call_2" {
+		t.Fatalf("messages.3.tool_call_id = %q, want %q", got, "call_2")
+	}
+	if got := gjson.GetBytes(out, "messages.2.reasoning_content").String(); got != "r1" {
+		t.Fatalf("messages.2.reasoning_content = %q, want %q", got, "r1")
+	}
+}

internal/runtime/executor/openai_compat_executor.go

@@ -172,11 +172,11 @@ func (e *OpenAICompatExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 	// Translate response back to source format when needed
 	var param any
 	out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, translated, body, &param)
-	resp = cliproxyexecutor.Response{Payload: []byte(out)}
+	resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
 	return resp, nil
 }
 
-func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	reporter := newUsageReporter(ctx, e.Identifier(), baseModel, auth)
@@ -258,7 +258,6 @@ func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 		return nil, err
 	}
 	out := make(chan cliproxyexecutor.StreamChunk)
-	stream = out
 	go func() {
 		defer close(out)
 		defer func() {
@@ -298,7 +297,7 @@ func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 		// Ensure we record the request if no usage chunk was ever seen
 		reporter.ensurePublished(ctx)
 	}()
-	return stream, nil
+	return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
 }
 
 func (e *OpenAICompatExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {

internal/runtime/executor/qwen_executor.go

@@ -22,9 +22,7 @@ import (
 )
 
 const (
-	qwenUserAgent           = "google-api-nodejs-client/9.15.1"
-	qwenXGoogAPIClient      = "gl-node/22.17.0"
-	qwenClientMetadataValue = "ideType=IDE_UNSPECIFIED,platform=PLATFORM_UNSPECIFIED,pluginType=GEMINI"
+	qwenUserAgent = "QwenCode/0.10.3 (darwin; arm64)"
 )
 
 // QwenExecutor is a stateless executor for Qwen Code using OpenAI-compatible chat completions.
@@ -152,11 +150,11 @@ func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req
 	// Note: TranslateNonStream uses req.Model (original with suffix) to preserve
 	// the original model name in the response for client compatibility.
 	out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, opts.OriginalRequest, body, data, &param)
-	resp = cliproxyexecutor.Response{Payload: []byte(out)}
+	resp = cliproxyexecutor.Response{Payload: []byte(out), Headers: httpResp.Header.Clone()}
 	return resp, nil
 }
 
-func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
 	if opts.Alt == "responses/compact" {
 		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
 	}
@@ -238,7 +236,6 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
 		return nil, err
 	}
 	out := make(chan cliproxyexecutor.StreamChunk)
-	stream = out
 	go func() {
 		defer close(out)
 		defer func() {
@@ -270,7 +267,7 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
 			out <- cliproxyexecutor.StreamChunk{Err: errScan}
 		}
 	}()
-	return stream, nil
+	return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
 }
 
 func (e *QwenExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
@@ -344,8 +341,18 @@ func applyQwenHeaders(r *http.Request, token string, stream bool) {
 	r.Header.Set("Content-Type", "application/json")
 	r.Header.Set("Authorization", "Bearer "+token)
 	r.Header.Set("User-Agent", qwenUserAgent)
-	r.Header.Set("X-Goog-Api-Client", qwenXGoogAPIClient)
-	r.Header.Set("Client-Metadata", qwenClientMetadataValue)
+	r.Header.Set("X-Dashscope-Useragent", qwenUserAgent)
+	r.Header.Set("X-Stainless-Runtime-Version", "v22.17.0")
+	r.Header.Set("Sec-Fetch-Mode", "cors")
+	r.Header.Set("X-Stainless-Lang", "js")
+	r.Header.Set("X-Stainless-Arch", "arm64")
+	r.Header.Set("X-Stainless-Package-Version", "5.11.0")
+	r.Header.Set("X-Dashscope-Cachecontrol", "enable")
+	r.Header.Set("X-Stainless-Retry-Count", "0")
+	r.Header.Set("X-Stainless-Os", "MacOS")
+	r.Header.Set("X-Dashscope-Authtype", "qwen-oauth")
+	r.Header.Set("X-Stainless-Runtime", "node")
+
 	if stream {
 		r.Header.Set("Accept", "text/event-stream")
 		return

internal/runtime/executor/thinking_providers.go

@@ -7,5 +7,6 @@ import (
 	_ "github.com/router-for-me/CLIProxyAPI/v6/internal/thinking/provider/gemini"
 	_ "github.com/router-for-me/CLIProxyAPI/v6/internal/thinking/provider/geminicli"
 	_ "github.com/router-for-me/CLIProxyAPI/v6/internal/thinking/provider/iflow"
+	_ "github.com/router-for-me/CLIProxyAPI/v6/internal/thinking/provider/kimi"
 	_ "github.com/router-for-me/CLIProxyAPI/v6/internal/thinking/provider/openai"
 )

internal/runtime/executor/user_id_cache.go

@@ -0,0 +1,89 @@
+package executor
+
+import (
+	"crypto/sha256"
+	"encoding/hex"
+	"sync"
+	"time"
+)
+
+type userIDCacheEntry struct {
+	value  string
+	expire time.Time
+}
+
+var (
+	userIDCache            = make(map[string]userIDCacheEntry)
+	userIDCacheMu          sync.RWMutex
+	userIDCacheCleanupOnce sync.Once
+)
+
+const (
+	userIDTTL                = time.Hour
+	userIDCacheCleanupPeriod = 15 * time.Minute
+)
+
+func startUserIDCacheCleanup() {
+	go func() {
+		ticker := time.NewTicker(userIDCacheCleanupPeriod)
+		defer ticker.Stop()
+		for range ticker.C {
+			purgeExpiredUserIDs()
+		}
+	}()
+}
+
+func purgeExpiredUserIDs() {
+	now := time.Now()
+	userIDCacheMu.Lock()
+	for key, entry := range userIDCache {
+		if !entry.expire.After(now) {
+			delete(userIDCache, key)
+		}
+	}
+	userIDCacheMu.Unlock()
+}
+
+func userIDCacheKey(apiKey string) string {
+	sum := sha256.Sum256([]byte(apiKey))
+	return hex.EncodeToString(sum[:])
+}
+
+func cachedUserID(apiKey string) string {
+	if apiKey == "" {
+		return generateFakeUserID()
+	}
+
+	userIDCacheCleanupOnce.Do(startUserIDCacheCleanup)
+
+	key := userIDCacheKey(apiKey)
+	now := time.Now()
+
+	userIDCacheMu.RLock()
+	entry, ok := userIDCache[key]
+	valid := ok && entry.value != "" && entry.expire.After(now) && isValidUserID(entry.value)
+	userIDCacheMu.RUnlock()
+	if valid {
+		userIDCacheMu.Lock()
+		entry = userIDCache[key]
+		if entry.value != "" && entry.expire.After(now) && isValidUserID(entry.value) {
+			entry.expire = now.Add(userIDTTL)
+			userIDCache[key] = entry
+			userIDCacheMu.Unlock()
+			return entry.value
+		}
+		userIDCacheMu.Unlock()
+	}
+
+	newID := generateFakeUserID()
+
+	userIDCacheMu.Lock()
+	entry, ok = userIDCache[key]
+	if !ok || entry.value == "" || !entry.expire.After(now) || !isValidUserID(entry.value) {
+		entry.value = newID
+	}
+	entry.expire = now.Add(userIDTTL)
+	userIDCache[key] = entry
+	userIDCacheMu.Unlock()
+	return entry.value
+}

internal/runtime/executor/user_id_cache_test.go

@@ -0,0 +1,86 @@
+package executor
+
+import (
+	"testing"
+	"time"
+)
+
+func resetUserIDCache() {
+	userIDCacheMu.Lock()
+	userIDCache = make(map[string]userIDCacheEntry)
+	userIDCacheMu.Unlock()
+}
+
+func TestCachedUserID_ReusesWithinTTL(t *testing.T) {
+	resetUserIDCache()
+
+	first := cachedUserID("api-key-1")
+	second := cachedUserID("api-key-1")
+
+	if first == "" {
+		t.Fatal("expected generated user_id to be non-empty")
+	}
+	if first != second {
+		t.Fatalf("expected cached user_id to be reused, got %q and %q", first, second)
+	}
+}
+
+func TestCachedUserID_ExpiresAfterTTL(t *testing.T) {
+	resetUserIDCache()
+
+	expiredID := cachedUserID("api-key-expired")
+	cacheKey := userIDCacheKey("api-key-expired")
+	userIDCacheMu.Lock()
+	userIDCache[cacheKey] = userIDCacheEntry{
+		value:  expiredID,
+		expire: time.Now().Add(-time.Minute),
+	}
+	userIDCacheMu.Unlock()
+
+	newID := cachedUserID("api-key-expired")
+	if newID == expiredID {
+		t.Fatalf("expected expired user_id to be replaced, got %q", newID)
+	}
+	if newID == "" {
+		t.Fatal("expected regenerated user_id to be non-empty")
+	}
+}
+
+func TestCachedUserID_IsScopedByAPIKey(t *testing.T) {
+	resetUserIDCache()
+
+	first := cachedUserID("api-key-1")
+	second := cachedUserID("api-key-2")
+
+	if first == second {
+		t.Fatalf("expected different API keys to have different user_ids, got %q", first)
+	}
+}
+
+func TestCachedUserID_RenewsTTLOnHit(t *testing.T) {
+	resetUserIDCache()
+
+	key := "api-key-renew"
+	id := cachedUserID(key)
+	cacheKey := userIDCacheKey(key)
+
+	soon := time.Now()
+	userIDCacheMu.Lock()
+	userIDCache[cacheKey] = userIDCacheEntry{
+		value:  id,
+		expire: soon.Add(2 * time.Second),
+	}
+	userIDCacheMu.Unlock()
+
+	if refreshed := cachedUserID(key); refreshed != id {
+		t.Fatalf("expected cached user_id to be reused before expiry, got %q", refreshed)
+	}
+
+	userIDCacheMu.RLock()
+	entry := userIDCache[cacheKey]
+	userIDCacheMu.RUnlock()
+
+	if entry.expire.Sub(soon) < 30*time.Minute {
+		t.Fatalf("expected TTL to renew, got %v remaining", entry.expire.Sub(soon))
+	}
+}

internal/store/gitstore.go

@@ -21,6 +21,9 @@ import (
 	cliproxyauth "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/auth"
 )
 
+// gcInterval defines minimum time between garbage collection runs.
+const gcInterval = 5 * time.Minute
+
 // GitTokenStore persists token records and auth metadata using git as the backing storage.
 type GitTokenStore struct {
 	mu        sync.Mutex
@@ -31,6 +34,7 @@ type GitTokenStore struct {
 	remote    string
 	username  string
 	password  string
+	lastGC    time.Time
 }
 
 // NewGitTokenStore creates a token store that saves credentials to disk through the
@@ -613,6 +617,7 @@ func (s *GitTokenStore) commitAndPushLocked(message string, relPaths ...string)
 	} else if errRewrite := s.rewriteHeadAsSingleCommit(repo, headRef.Name(), commitHash, message, signature); errRewrite != nil {
 		return errRewrite
 	}
+	s.maybeRunGC(repo)
 	if err = repo.Push(&git.PushOptions{Auth: s.gitAuth(), Force: true}); err != nil {
 		if errors.Is(err, git.NoErrAlreadyUpToDate) {
 			return nil
@@ -652,6 +657,23 @@ func (s *GitTokenStore) rewriteHeadAsSingleCommit(repo *git.Repository, branch p
 	return nil
 }
 
+func (s *GitTokenStore) maybeRunGC(repo *git.Repository) {
+	now := time.Now()
+	if now.Sub(s.lastGC) < gcInterval {
+		return
+	}
+	s.lastGC = now
+
+	pruneOpts := git.PruneOptions{
+		OnlyObjectsOlderThan: now,
+		Handler:              repo.DeleteObject,
+	}
+	if err := repo.Prune(pruneOpts); err != nil && !errors.Is(err, git.ErrLooseObjectsNotSupported) {
+		return
+	}
+	_ = repo.RepackObjects(&git.RepackConfig{})
+}
+
 // PersistConfig commits and pushes configuration changes to git.
 func (s *GitTokenStore) PersistConfig(_ context.Context) error {
 	if err := s.EnsureRepository(); err != nil {

internal/thinking/apply.go

@@ -18,6 +18,7 @@ var providerAppliers = map[string]ProviderApplier{
 	"codex":       nil,
 	"iflow":       nil,
 	"antigravity": nil,
+	"kimi":        nil,
 }
 
 // GetProviderApplier returns the ProviderApplier for the given provider name.
@@ -326,6 +327,9 @@ func extractThinkingConfig(body []byte, provider string) ThinkingConfig {
 			return config
 		}
 		return extractOpenAIConfig(body)
+	case "kimi":
+		// Kimi uses OpenAI-compatible reasoning_effort format
+		return extractOpenAIConfig(body)
 	default:
 		return ThinkingConfig{}
 	}

internal/thinking/provider/kimi/apply.go

@@ -0,0 +1,126 @@
+// Package kimi implements thinking configuration for Kimi (Moonshot AI) models.
+//
+// Kimi models use the OpenAI-compatible reasoning_effort format with discrete levels
+// (low/medium/high). The provider strips any existing thinking config and applies
+// the unified ThinkingConfig in OpenAI format.
+package kimi
+
+import (
+	"fmt"
+
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/thinking"
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
+)
+
+// Applier implements thinking.ProviderApplier for Kimi models.
+//
+// Kimi-specific behavior:
+//   - Output format: reasoning_effort (string: low/medium/high)
+//   - Uses OpenAI-compatible format
+//   - Supports budget-to-level conversion
+type Applier struct{}
+
+var _ thinking.ProviderApplier = (*Applier)(nil)
+
+// NewApplier creates a new Kimi thinking applier.
+func NewApplier() *Applier {
+	return &Applier{}
+}
+
+func init() {
+	thinking.RegisterProvider("kimi", NewApplier())
+}
+
+// Apply applies thinking configuration to Kimi request body.
+//
+// Expected output format:
+//
+//	{
+//	  "reasoning_effort": "high"
+//	}
+func (a *Applier) Apply(body []byte, config thinking.ThinkingConfig, modelInfo *registry.ModelInfo) ([]byte, error) {
+	if thinking.IsUserDefinedModel(modelInfo) {
+		return applyCompatibleKimi(body, config)
+	}
+	if modelInfo.Thinking == nil {
+		return body, nil
+	}
+
+	if len(body) == 0 || !gjson.ValidBytes(body) {
+		body = []byte(`{}`)
+	}
+
+	var effort string
+	switch config.Mode {
+	case thinking.ModeLevel:
+		if config.Level == "" {
+			return body, nil
+		}
+		effort = string(config.Level)
+	case thinking.ModeNone:
+		// Kimi uses "none" to disable thinking
+		effort = string(thinking.LevelNone)
+	case thinking.ModeBudget:
+		// Convert budget to level using threshold mapping
+		level, ok := thinking.ConvertBudgetToLevel(config.Budget)
+		if !ok {
+			return body, nil
+		}
+		effort = level
+	case thinking.ModeAuto:
+		// Auto mode maps to "auto" effort
+		effort = string(thinking.LevelAuto)
+	default:
+		return body, nil
+	}
+
+	if effort == "" {
+		return body, nil
+	}
+
+	result, err := sjson.SetBytes(body, "reasoning_effort", effort)
+	if err != nil {
+		return body, fmt.Errorf("kimi thinking: failed to set reasoning_effort: %w", err)
+	}
+	return result, nil
+}
+
+// applyCompatibleKimi applies thinking config for user-defined Kimi models.
+func applyCompatibleKimi(body []byte, config thinking.ThinkingConfig) ([]byte, error) {
+	if len(body) == 0 || !gjson.ValidBytes(body) {
+		body = []byte(`{}`)
+	}
+
+	var effort string
+	switch config.Mode {
+	case thinking.ModeLevel:
+		if config.Level == "" {
+			return body, nil
+		}
+		effort = string(config.Level)
+	case thinking.ModeNone:
+		effort = string(thinking.LevelNone)
+		if config.Level != "" {
+			effort = string(config.Level)
+		}
+	case thinking.ModeAuto:
+		effort = string(thinking.LevelAuto)
+	case thinking.ModeBudget:
+		// Convert budget to level
+		level, ok := thinking.ConvertBudgetToLevel(config.Budget)
+		if !ok {
+			return body, nil
+		}
+		effort = level
+	default:
+		return body, nil
+	}
+
+	result, err := sjson.SetBytes(body, "reasoning_effort", effort)
+	if err != nil {
+		return body, fmt.Errorf("kimi thinking: failed to set reasoning_effort: %w", err)
+	}
+	return result, nil
+}

internal/translator/antigravity/claude/antigravity_claude_request.go

@@ -231,8 +231,12 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 
 							} else if functionResponseResult.IsObject() {
 								functionResponseJSON, _ = sjson.SetRaw(functionResponseJSON, "response.result", functionResponseResult.Raw)
-							} else {
+							} else if functionResponseResult.Raw != "" {
 								functionResponseJSON, _ = sjson.SetRaw(functionResponseJSON, "response.result", functionResponseResult.Raw)
+							} else {
+								// Content field is missing entirely — .Raw is empty which
+								// causes sjson.SetRaw to produce invalid JSON (e.g. "result":}).
+								functionResponseJSON, _ = sjson.Set(functionResponseJSON, "response.result", "")
 							}
 
 							partJSON := `{}`
@@ -344,7 +348,8 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 	// Inject interleaved thinking hint when both tools and thinking are active
 	hasTools := toolDeclCount > 0
 	thinkingResult := gjson.GetBytes(rawJSON, "thinking")
-	hasThinking := thinkingResult.Exists() && thinkingResult.IsObject() && thinkingResult.Get("type").String() == "enabled"
+	thinkingType := thinkingResult.Get("type").String()
+	hasThinking := thinkingResult.Exists() && thinkingResult.IsObject() && (thinkingType == "enabled" || thinkingType == "adaptive")
 	isClaudeThinking := util.IsClaudeThinkingModel(modelName)
 
 	if hasTools && hasThinking && isClaudeThinking {
@@ -377,12 +382,18 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 
 	// Map Anthropic thinking -> Gemini thinkingBudget/include_thoughts when type==enabled
 	if t := gjson.GetBytes(rawJSON, "thinking"); enableThoughtTranslate && t.Exists() && t.IsObject() {
-		if t.Get("type").String() == "enabled" {
+		switch t.Get("type").String() {
+		case "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.includeThoughts", true)
 			}
+		case "adaptive":
+			// Keep adaptive as a high level sentinel; ApplyThinking resolves it
+			// to model-specific max capability.
+			out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.thinkingLevel", "high")
+			out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.includeThoughts", true)
 		}
 	}
 	if v := gjson.GetBytes(rawJSON, "temperature"); v.Exists() && v.Type == gjson.Number {

internal/translator/antigravity/claude/antigravity_claude_request_test.go

@@ -661,6 +661,85 @@ func TestConvertClaudeRequestToAntigravity_ThinkingOnly_NoHint(t *testing.T) {
 	}
 }
 
+func TestConvertClaudeRequestToAntigravity_ToolResultNoContent(t *testing.T) {
+	// Bug repro: tool_result with no content field produces invalid JSON
+	inputJSON := []byte(`{
+		"model": "claude-opus-4-6-thinking",
+		"messages": [
+			{
+				"role": "assistant",
+				"content": [
+					{
+						"type": "tool_use",
+						"id": "MyTool-123-456",
+						"name": "MyTool",
+						"input": {"key": "value"}
+					}
+				]
+			},
+			{
+				"role": "user",
+				"content": [
+					{
+						"type": "tool_result",
+						"tool_use_id": "MyTool-123-456"
+					}
+				]
+			}
+		]
+	}`)
+
+	output := ConvertClaudeRequestToAntigravity("claude-opus-4-6-thinking", inputJSON, true)
+	outputStr := string(output)
+
+	if !gjson.Valid(outputStr) {
+		t.Errorf("Result is not valid JSON:\n%s", outputStr)
+	}
+
+	// Verify the functionResponse has a valid result value
+	fr := gjson.Get(outputStr, "request.contents.1.parts.0.functionResponse.response.result")
+	if !fr.Exists() {
+		t.Error("functionResponse.response.result should exist")
+	}
+}
+
+func TestConvertClaudeRequestToAntigravity_ToolResultNullContent(t *testing.T) {
+	// Bug repro: tool_result with null content produces invalid JSON
+	inputJSON := []byte(`{
+		"model": "claude-opus-4-6-thinking",
+		"messages": [
+			{
+				"role": "assistant",
+				"content": [
+					{
+						"type": "tool_use",
+						"id": "MyTool-123-456",
+						"name": "MyTool",
+						"input": {"key": "value"}
+					}
+				]
+			},
+			{
+				"role": "user",
+				"content": [
+					{
+						"type": "tool_result",
+						"tool_use_id": "MyTool-123-456",
+						"content": null
+					}
+				]
+			}
+		]
+	}`)
+
+	output := ConvertClaudeRequestToAntigravity("claude-opus-4-6-thinking", inputJSON, true)
+	outputStr := string(output)
+
+	if !gjson.Valid(outputStr) {
+		t.Errorf("Result is not valid JSON:\n%s", outputStr)
+	}
+}
+
 func TestConvertClaudeRequestToAntigravity_ToolAndThinking_NoExistingSystem(t *testing.T) {
 	// When tools + thinking but no system instruction, should create one with hint
 	inputJSON := []byte(`{

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

@@ -95,9 +95,9 @@ func ConvertAntigravityResponseToOpenAI(_ context.Context, _ string, originalReq
 		if totalTokenCountResult := usageResult.Get("totalTokenCount"); totalTokenCountResult.Exists() {
 			template, _ = sjson.Set(template, "usage.total_tokens", totalTokenCountResult.Int())
 		}
-		promptTokenCount := usageResult.Get("promptTokenCount").Int() - cachedTokenCount
+		promptTokenCount := usageResult.Get("promptTokenCount").Int()
 		thoughtsTokenCount := usageResult.Get("thoughtsTokenCount").Int()
-		template, _ = sjson.Set(template, "usage.prompt_tokens", promptTokenCount+thoughtsTokenCount)
+		template, _ = sjson.Set(template, "usage.prompt_tokens", promptTokenCount)
 		if thoughtsTokenCount > 0 {
 			template, _ = sjson.Set(template, "usage.completion_tokens_details.reasoning_tokens", thoughtsTokenCount)
 		}

internal/translator/claude/openai/chat-completions/claude_openai_request.go

@@ -199,6 +199,21 @@ func ConvertOpenAIRequestToClaude(modelName string, inputRawJSON []byte, stream
 									msg, _ = sjson.SetRaw(msg, "content.-1", imagePart)
 								}
 							}
+
+						case "file":
+							fileData := part.Get("file.file_data").String()
+							if strings.HasPrefix(fileData, "data:") {
+								semicolonIdx := strings.Index(fileData, ";")
+								commaIdx := strings.Index(fileData, ",")
+								if semicolonIdx != -1 && commaIdx != -1 && commaIdx > semicolonIdx {
+									mediaType := strings.TrimPrefix(fileData[:semicolonIdx], "data:")
+									data := fileData[commaIdx+1:]
+									docPart := `{"type":"document","source":{"type":"base64","media_type":"","data":""}}`
+									docPart, _ = sjson.Set(docPart, "source.media_type", mediaType)
+									docPart, _ = sjson.Set(docPart, "source.data", data)
+									msg, _ = sjson.SetRaw(msg, "content.-1", docPart)
+								}
+							}
 						}
 						return true
 					})

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

@@ -155,6 +155,7 @@ func ConvertOpenAIResponsesRequestToClaude(modelName string, inputRawJSON []byte
 				var textAggregate strings.Builder
 				var partsJSON []string
 				hasImage := false
+				hasFile := false
 				if parts := item.Get("content"); parts.Exists() && parts.IsArray() {
 					parts.ForEach(func(_, part gjson.Result) bool {
 						ptype := part.Get("type").String()
@@ -207,6 +208,30 @@ func ConvertOpenAIResponsesRequestToClaude(modelName string, inputRawJSON []byte
 									hasImage = true
 								}
 							}
+						case "input_file":
+							fileData := part.Get("file_data").String()
+							if fileData != "" {
+								mediaType := "application/octet-stream"
+								data := fileData
+								if strings.HasPrefix(fileData, "data:") {
+									trimmed := strings.TrimPrefix(fileData, "data:")
+									mediaAndData := strings.SplitN(trimmed, ";base64,", 2)
+									if len(mediaAndData) == 2 {
+										if mediaAndData[0] != "" {
+											mediaType = mediaAndData[0]
+										}
+										data = mediaAndData[1]
+									}
+								}
+								contentPart := `{"type":"document","source":{"type":"base64","media_type":"","data":""}}`
+								contentPart, _ = sjson.Set(contentPart, "source.media_type", mediaType)
+								contentPart, _ = sjson.Set(contentPart, "source.data", data)
+								partsJSON = append(partsJSON, contentPart)
+								if role == "" {
+									role = "user"
+								}
+								hasFile = true
+							}
 						}
 						return true
 					})
@@ -228,7 +253,7 @@ func ConvertOpenAIResponsesRequestToClaude(modelName string, inputRawJSON []byte
 				if len(partsJSON) > 0 {
 					msg := `{"role":"","content":[]}`
 					msg, _ = sjson.Set(msg, "role", role)
-					if len(partsJSON) == 1 && !hasImage {
+					if len(partsJSON) == 1 && !hasImage && !hasFile {
 						// Preserve legacy behavior for single text content
 						msg, _ = sjson.Delete(msg, "content")
 						textPart := gjson.Parse(partsJSON[0])

internal/translator/codex/claude/codex_claude_request.go

@@ -222,6 +222,10 @@ func ConvertClaudeRequestToCodex(modelName string, inputRawJSON []byte, _ bool)
 					reasoningEffort = effort
 				}
 			}
+		case "adaptive":
+			// Claude adaptive means "enable with max capacity"; keep it as highest level
+			// and let ApplyThinking normalize per target model capability.
+			reasoningEffort = string(thinking.LevelXHigh)
 		case "disabled":
 			if effort, ok := thinking.ConvertBudgetToLevel(0); ok && effort != "" {
 				reasoningEffort = effort

internal/translator/codex/claude/codex_claude_response.go

@@ -22,8 +22,9 @@ var (
 
 // ConvertCodexResponseToClaudeParams holds parameters for response conversion.
 type ConvertCodexResponseToClaudeParams struct {
-	HasToolCall bool
-	BlockIndex  int
+	HasToolCall              bool
+	BlockIndex               int
+	HasReceivedArgumentsDelta bool
 }
 
 // ConvertCodexResponseToClaude performs sophisticated streaming response format conversion.
@@ -113,10 +114,10 @@ func ConvertCodexResponseToClaude(_ context.Context, _ string, originalRequestRa
 		template = `{"type":"message_delta","delta":{"stop_reason":"tool_use","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
 		p := (*param).(*ConvertCodexResponseToClaudeParams).HasToolCall
 		stopReason := rootResult.Get("response.stop_reason").String()
-		if stopReason != "" {
-			template, _ = sjson.Set(template, "delta.stop_reason", stopReason)
-		} else if p {
+		if p {
 			template, _ = sjson.Set(template, "delta.stop_reason", "tool_use")
+		} else if stopReason == "max_tokens" || stopReason == "stop" {
+			template, _ = sjson.Set(template, "delta.stop_reason", stopReason)
 		} else {
 			template, _ = sjson.Set(template, "delta.stop_reason", "end_turn")
 		}
@@ -137,6 +138,7 @@ func ConvertCodexResponseToClaude(_ context.Context, _ string, originalRequestRa
 		itemType := itemResult.Get("type").String()
 		if itemType == "function_call" {
 			(*param).(*ConvertCodexResponseToClaudeParams).HasToolCall = true
+			(*param).(*ConvertCodexResponseToClaudeParams).HasReceivedArgumentsDelta = false
 			template = `{"type":"content_block_start","index":0,"content_block":{"type":"tool_use","id":"","name":"","input":{}}}`
 			template, _ = sjson.Set(template, "index", (*param).(*ConvertCodexResponseToClaudeParams).BlockIndex)
 			template, _ = sjson.Set(template, "content_block.id", itemResult.Get("call_id").String())
@@ -171,12 +173,29 @@ func ConvertCodexResponseToClaude(_ context.Context, _ string, originalRequestRa
 			output += fmt.Sprintf("data: %s\n\n", template)
 		}
 	} else if typeStr == "response.function_call_arguments.delta" {
+		(*param).(*ConvertCodexResponseToClaudeParams).HasReceivedArgumentsDelta = true
 		template = `{"type":"content_block_delta","index":0,"delta":{"type":"input_json_delta","partial_json":""}}`
 		template, _ = sjson.Set(template, "index", (*param).(*ConvertCodexResponseToClaudeParams).BlockIndex)
 		template, _ = sjson.Set(template, "delta.partial_json", rootResult.Get("delta").String())
 
 		output += "event: content_block_delta\n"
 		output += fmt.Sprintf("data: %s\n\n", template)
+	} else if typeStr == "response.function_call_arguments.done" {
+		// Some models (e.g. gpt-5.3-codex-spark) send function call arguments
+		// in a single "done" event without preceding "delta" events.
+		// Emit the full arguments as a single input_json_delta so the
+		// downstream Claude client receives the complete tool input.
+		// When delta events were already received, skip to avoid duplicating arguments.
+		if !(*param).(*ConvertCodexResponseToClaudeParams).HasReceivedArgumentsDelta {
+			if args := rootResult.Get("arguments").String(); args != "" {
+				template = `{"type":"content_block_delta","index":0,"delta":{"type":"input_json_delta","partial_json":""}}`
+				template, _ = sjson.Set(template, "index", (*param).(*ConvertCodexResponseToClaudeParams).BlockIndex)
+				template, _ = sjson.Set(template, "delta.partial_json", args)
+
+				output += "event: content_block_delta\n"
+				output += fmt.Sprintf("data: %s\n\n", template)
+			}
+		}
 	}
 
 	return []string{output}

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

@@ -180,7 +180,19 @@ func ConvertOpenAIRequestToCodex(modelName string, inputRawJSON []byte, stream b
 								msg, _ = sjson.SetRaw(msg, "content.-1", part)
 							}
 						case "file":
-							// Files are not specified in examples; skip for now
+							if role == "user" {
+								fileData := it.Get("file.file_data").String()
+								filename := it.Get("file.filename").String()
+								if fileData != "" {
+									part := `{}`
+									part, _ = sjson.Set(part, "type", "input_file")
+									part, _ = sjson.Set(part, "file_data", fileData)
+									if filename != "" {
+										part, _ = sjson.Set(part, "filename", filename)
+									}
+									msg, _ = sjson.SetRaw(msg, "content.-1", part)
+								}
+							}
 						}
 					}
 				}

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

@@ -20,10 +20,12 @@ var (
 
 // ConvertCliToOpenAIParams holds parameters for response conversion.
 type ConvertCliToOpenAIParams struct {
-	ResponseID        string
-	CreatedAt         int64
-	Model             string
-	FunctionCallIndex int
+	ResponseID                string
+	CreatedAt                 int64
+	Model                     string
+	FunctionCallIndex         int
+	HasReceivedArgumentsDelta bool
+	HasToolCallAnnounced      bool
 }
 
 // ConvertCodexResponseToOpenAI translates a single chunk of a streaming response from the
@@ -43,10 +45,12 @@ type ConvertCliToOpenAIParams struct {
 func ConvertCodexResponseToOpenAI(_ context.Context, modelName string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, param *any) []string {
 	if *param == nil {
 		*param = &ConvertCliToOpenAIParams{
-			Model:             modelName,
-			CreatedAt:         0,
-			ResponseID:        "",
-			FunctionCallIndex: -1,
+			Model:                     modelName,
+			CreatedAt:                 0,
+			ResponseID:                "",
+			FunctionCallIndex:         -1,
+			HasReceivedArgumentsDelta: false,
+			HasToolCallAnnounced:      false,
 		}
 	}
 
@@ -90,6 +94,9 @@ func ConvertCodexResponseToOpenAI(_ context.Context, modelName string, originalR
 		if inputTokensResult := usageResult.Get("input_tokens"); inputTokensResult.Exists() {
 			template, _ = sjson.Set(template, "usage.prompt_tokens", inputTokensResult.Int())
 		}
+		if cachedTokensResult := usageResult.Get("input_tokens_details.cached_tokens"); cachedTokensResult.Exists() {
+			template, _ = sjson.Set(template, "usage.prompt_tokens_details.cached_tokens", cachedTokensResult.Int())
+		}
 		if reasoningTokensResult := usageResult.Get("output_tokens_details.reasoning_tokens"); reasoningTokensResult.Exists() {
 			template, _ = sjson.Set(template, "usage.completion_tokens_details.reasoning_tokens", reasoningTokensResult.Int())
 		}
@@ -115,35 +122,93 @@ func ConvertCodexResponseToOpenAI(_ context.Context, modelName string, originalR
 		}
 		template, _ = sjson.Set(template, "choices.0.finish_reason", finishReason)
 		template, _ = sjson.Set(template, "choices.0.native_finish_reason", finishReason)
-	} else if dataType == "response.output_item.done" {
-		functionCallItemTemplate := `{"index":0,"id":"","type":"function","function":{"name":"","arguments":""}}`
+	} else if dataType == "response.output_item.added" {
 		itemResult := rootResult.Get("item")
-		if itemResult.Exists() {
-			if itemResult.Get("type").String() != "function_call" {
-				return []string{}
-			}
-
-			// set the index
-			(*param).(*ConvertCliToOpenAIParams).FunctionCallIndex++
-			functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "index", (*param).(*ConvertCliToOpenAIParams).FunctionCallIndex)
-
-			template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls", `[]`)
-			functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "id", itemResult.Get("call_id").String())
-
-			// Restore original tool name if it was shortened
-			name := itemResult.Get("name").String()
-			// Build reverse map on demand from original request tools
-			rev := buildReverseMapFromOriginalOpenAI(originalRequestRawJSON)
-			if orig, ok := rev[name]; ok {
-				name = orig
-			}
-			functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.name", name)
-
-			functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.arguments", itemResult.Get("arguments").String())
-			template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
-			template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls.-1", functionCallItemTemplate)
+		if !itemResult.Exists() || itemResult.Get("type").String() != "function_call" {
+			return []string{}
 		}
 
+		// Increment index for this new function call item.
+		(*param).(*ConvertCliToOpenAIParams).FunctionCallIndex++
+		(*param).(*ConvertCliToOpenAIParams).HasReceivedArgumentsDelta = false
+		(*param).(*ConvertCliToOpenAIParams).HasToolCallAnnounced = true
+
+		functionCallItemTemplate := `{"index":0,"id":"","type":"function","function":{"name":"","arguments":""}}`
+		functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "index", (*param).(*ConvertCliToOpenAIParams).FunctionCallIndex)
+		functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "id", itemResult.Get("call_id").String())
+
+		// Restore original tool name if it was shortened.
+		name := itemResult.Get("name").String()
+		rev := buildReverseMapFromOriginalOpenAI(originalRequestRawJSON)
+		if orig, ok := rev[name]; ok {
+			name = orig
+		}
+		functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.name", name)
+		functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.arguments", "")
+
+		template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
+		template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls", `[]`)
+		template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls.-1", functionCallItemTemplate)
+
+	} else if dataType == "response.function_call_arguments.delta" {
+		(*param).(*ConvertCliToOpenAIParams).HasReceivedArgumentsDelta = true
+
+		deltaValue := rootResult.Get("delta").String()
+		functionCallItemTemplate := `{"index":0,"function":{"arguments":""}}`
+		functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "index", (*param).(*ConvertCliToOpenAIParams).FunctionCallIndex)
+		functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.arguments", deltaValue)
+
+		template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls", `[]`)
+		template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls.-1", functionCallItemTemplate)
+
+	} else if dataType == "response.function_call_arguments.done" {
+		if (*param).(*ConvertCliToOpenAIParams).HasReceivedArgumentsDelta {
+			// Arguments were already streamed via delta events; nothing to emit.
+			return []string{}
+		}
+
+		// Fallback: no delta events were received, emit the full arguments as a single chunk.
+		fullArgs := rootResult.Get("arguments").String()
+		functionCallItemTemplate := `{"index":0,"function":{"arguments":""}}`
+		functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "index", (*param).(*ConvertCliToOpenAIParams).FunctionCallIndex)
+		functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.arguments", fullArgs)
+
+		template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls", `[]`)
+		template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls.-1", functionCallItemTemplate)
+
+	} else if dataType == "response.output_item.done" {
+		itemResult := rootResult.Get("item")
+		if !itemResult.Exists() || itemResult.Get("type").String() != "function_call" {
+			return []string{}
+		}
+
+		if (*param).(*ConvertCliToOpenAIParams).HasToolCallAnnounced {
+			// Tool call was already announced via output_item.added; skip emission.
+			(*param).(*ConvertCliToOpenAIParams).HasToolCallAnnounced = false
+			return []string{}
+		}
+
+		// Fallback path: model skipped output_item.added, so emit complete tool call now.
+		(*param).(*ConvertCliToOpenAIParams).FunctionCallIndex++
+
+		functionCallItemTemplate := `{"index":0,"id":"","type":"function","function":{"name":"","arguments":""}}`
+		functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "index", (*param).(*ConvertCliToOpenAIParams).FunctionCallIndex)
+
+		template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls", `[]`)
+		functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "id", itemResult.Get("call_id").String())
+
+		// Restore original tool name if it was shortened.
+		name := itemResult.Get("name").String()
+		rev := buildReverseMapFromOriginalOpenAI(originalRequestRawJSON)
+		if orig, ok := rev[name]; ok {
+			name = orig
+		}
+		functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.name", name)
+
+		functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.arguments", itemResult.Get("arguments").String())
+		template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
+		template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls.-1", functionCallItemTemplate)
+
 	} else {
 		return []string{}
 	}
@@ -205,6 +270,9 @@ func ConvertCodexResponseToOpenAINonStream(_ context.Context, _ string, original
 		if inputTokensResult := usageResult.Get("input_tokens"); inputTokensResult.Exists() {
 			template, _ = sjson.Set(template, "usage.prompt_tokens", inputTokensResult.Int())
 		}
+		if cachedTokensResult := usageResult.Get("input_tokens_details.cached_tokens"); cachedTokensResult.Exists() {
+			template, _ = sjson.Set(template, "usage.prompt_tokens_details.cached_tokens", cachedTokensResult.Int())
+		}
 		if reasoningTokensResult := usageResult.Get("output_tokens_details.reasoning_tokens"); reasoningTokensResult.Exists() {
 			template, _ = sjson.Set(template, "usage.completion_tokens_details.reasoning_tokens", reasoningTokensResult.Int())
 		}

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

@@ -27,6 +27,9 @@ func ConvertOpenAIResponsesRequestToCodex(modelName string, inputRawJSON []byte,
 	rawJSON, _ = sjson.DeleteBytes(rawJSON, "top_p")
 	rawJSON, _ = sjson.DeleteBytes(rawJSON, "service_tier")
 
+	// Delete the user field as it is not supported by the Codex upstream.
+	rawJSON, _ = sjson.DeleteBytes(rawJSON, "user")
+
 	// Convert role "system" to "developer" in input array to comply with Codex API requirements.
 	rawJSON = convertSystemRoleToDeveloper(rawJSON)
 

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

@@ -263,3 +263,20 @@ func TestConvertSystemRoleToDeveloper_AssistantRole(t *testing.T) {
 		t.Errorf("Expected third role 'assistant', got '%s'", thirdRole.String())
 	}
 }
+
+func TestUserFieldDeletion(t *testing.T) {  
+	inputJSON := []byte(`{  
+		"model": "gpt-5.2",  
+		"user": "test-user",  
+		"input": [{"role": "user", "content": "Hello"}]  
+	}`)  
+	  
+	output := ConvertOpenAIResponsesRequestToCodex("gpt-5.2", inputJSON, false)  
+	outputStr := string(output)  
+	  
+	// Verify user field is deleted  
+	userField := gjson.Get(outputStr, "user")  
+	if userField.Exists() {
+		t.Errorf("user field should be deleted, but it was found with value: %s", userField.Raw)
+	}
+}

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

@@ -173,12 +173,18 @@ func ConvertClaudeRequestToCLI(modelName string, inputRawJSON []byte, _ bool) []
 
 	// Map Anthropic thinking -> Gemini thinkingBudget/include_thoughts when type==enabled
 	if t := gjson.GetBytes(rawJSON, "thinking"); t.Exists() && t.IsObject() {
-		if t.Get("type").String() == "enabled" {
+		switch t.Get("type").String() {
+		case "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.includeThoughts", true)
 			}
+		case "adaptive":
+			// Keep adaptive as a high level sentinel; ApplyThinking resolves it
+			// to model-specific max capability.
+			out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.thinkingLevel", "high")
+			out, _ = sjson.Set(out, "request.generationConfig.thinkingConfig.includeThoughts", true)
 		}
 	}
 	if v := gjson.GetBytes(rawJSON, "temperature"); v.Exists() && v.Type == gjson.Number {

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

@@ -14,6 +14,7 @@ import (
 	"time"
 
 	. "github.com/router-for-me/CLIProxyAPI/v6/internal/translator/gemini/openai/chat-completions"
+	log "github.com/sirupsen/logrus"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
 )
@@ -77,14 +78,20 @@ func ConvertCliResponseToOpenAI(_ context.Context, _ string, originalRequestRawJ
 		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()))
+	finishReason := ""
+	if stopReasonResult := gjson.GetBytes(rawJSON, "response.stop_reason"); stopReasonResult.Exists() {
+		finishReason = stopReasonResult.String()
 	}
+	if finishReason == "" {
+		if finishReasonResult := gjson.GetBytes(rawJSON, "response.candidates.0.finishReason"); finishReasonResult.Exists() {
+			finishReason = finishReasonResult.String()
+		}
+	}
+	finishReason = strings.ToLower(finishReason)
 
 	// Extract and set usage metadata (token counts).
 	if usageResult := gjson.GetBytes(rawJSON, "response.usageMetadata"); usageResult.Exists() {
+		cachedTokenCount := usageResult.Get("cachedContentTokenCount").Int()
 		if candidatesTokenCountResult := usageResult.Get("candidatesTokenCount"); candidatesTokenCountResult.Exists() {
 			template, _ = sjson.Set(template, "usage.completion_tokens", candidatesTokenCountResult.Int())
 		}
@@ -93,10 +100,18 @@ func ConvertCliResponseToOpenAI(_ context.Context, _ string, originalRequestRawJ
 		}
 		promptTokenCount := usageResult.Get("promptTokenCount").Int()
 		thoughtsTokenCount := usageResult.Get("thoughtsTokenCount").Int()
-		template, _ = sjson.Set(template, "usage.prompt_tokens", promptTokenCount+thoughtsTokenCount)
+		template, _ = sjson.Set(template, "usage.prompt_tokens", promptTokenCount)
 		if thoughtsTokenCount > 0 {
 			template, _ = sjson.Set(template, "usage.completion_tokens_details.reasoning_tokens", thoughtsTokenCount)
 		}
+		// Include cached token count if present (indicates prompt caching is working)
+		if cachedTokenCount > 0 {
+			var err error
+			template, err = sjson.Set(template, "usage.prompt_tokens_details.cached_tokens", cachedTokenCount)
+			if err != nil {
+				log.Warnf("gemini-cli openai response: failed to set cached_tokens: %v", err)
+			}
+		}
 	}
 
 	// Process the main content part of the response.
@@ -187,6 +202,12 @@ func ConvertCliResponseToOpenAI(_ context.Context, _ string, originalRequestRawJ
 	if hasFunctionCall {
 		template, _ = sjson.Set(template, "choices.0.finish_reason", "tool_calls")
 		template, _ = sjson.Set(template, "choices.0.native_finish_reason", "tool_calls")
+	} else if finishReason != "" && (*param).(*convertCliResponseToOpenAIChatParams).FunctionIndex == 0 {
+		// Only pass through specific finish reasons
+		if finishReason == "max_tokens" || finishReason == "stop" {
+			template, _ = sjson.Set(template, "choices.0.finish_reason", finishReason)
+			template, _ = sjson.Set(template, "choices.0.native_finish_reason", finishReason)
+		}
 	}
 
 	return []string{template}

internal/translator/gemini/claude/gemini_claude_request.go

@@ -154,12 +154,18 @@ func ConvertClaudeRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 	// Map Anthropic thinking -> Gemini thinkingBudget/include_thoughts when enabled
 	// Translator only does format conversion, ApplyThinking handles model capability validation.
 	if t := gjson.GetBytes(rawJSON, "thinking"); t.Exists() && t.IsObject() {
-		if t.Get("type").String() == "enabled" {
+		switch t.Get("type").String() {
+		case "enabled":
 			if b := t.Get("budget_tokens"); b.Exists() && b.Type == gjson.Number {
 				budget := int(b.Int())
 				out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingBudget", budget)
 				out, _ = sjson.Set(out, "generationConfig.thinkingConfig.includeThoughts", true)
 			}
+		case "adaptive":
+			// Keep adaptive as a high level sentinel; ApplyThinking resolves it
+			// to model-specific max capability.
+			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.thinkingLevel", "high")
+			out, _ = sjson.Set(out, "generationConfig.thinkingConfig.includeThoughts", true)
 		}
 	}
 	if v := gjson.GetBytes(rawJSON, "temperature"); v.Exists() && v.Type == gjson.Number {

internal/translator/gemini/openai/chat-completions/gemini_openai_response.go

@@ -100,9 +100,9 @@ func ConvertGeminiResponseToOpenAI(_ context.Context, _ string, originalRequestR
 		if totalTokenCountResult := usageResult.Get("totalTokenCount"); totalTokenCountResult.Exists() {
 			baseTemplate, _ = sjson.Set(baseTemplate, "usage.total_tokens", totalTokenCountResult.Int())
 		}
-		promptTokenCount := usageResult.Get("promptTokenCount").Int() - cachedTokenCount
+		promptTokenCount := usageResult.Get("promptTokenCount").Int()
 		thoughtsTokenCount := usageResult.Get("thoughtsTokenCount").Int()
-		baseTemplate, _ = sjson.Set(baseTemplate, "usage.prompt_tokens", promptTokenCount+thoughtsTokenCount)
+		baseTemplate, _ = sjson.Set(baseTemplate, "usage.prompt_tokens", promptTokenCount)
 		if thoughtsTokenCount > 0 {
 			baseTemplate, _ = sjson.Set(baseTemplate, "usage.completion_tokens_details.reasoning_tokens", thoughtsTokenCount)
 		}
@@ -129,11 +129,16 @@ func ConvertGeminiResponseToOpenAI(_ context.Context, _ string, originalRequestR
 			candidateIndex := int(candidate.Get("index").Int())
 			template, _ = sjson.Set(template, "choices.0.index", candidateIndex)
 
-			// Extract and set the finish reason.
-			if finishReasonResult := candidate.Get("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()))
+			finishReason := ""
+			if stopReasonResult := gjson.GetBytes(rawJSON, "stop_reason"); stopReasonResult.Exists() {
+				finishReason = stopReasonResult.String()
 			}
+			if finishReason == "" {
+				if finishReasonResult := gjson.GetBytes(rawJSON, "candidates.0.finishReason"); finishReasonResult.Exists() {
+					finishReason = finishReasonResult.String()
+				}
+			}
+			finishReason = strings.ToLower(finishReason)
 
 			partsResult := candidate.Get("content.parts")
 			hasFunctionCall := false
@@ -225,6 +230,12 @@ func ConvertGeminiResponseToOpenAI(_ context.Context, _ string, originalRequestR
 			if hasFunctionCall {
 				template, _ = sjson.Set(template, "choices.0.finish_reason", "tool_calls")
 				template, _ = sjson.Set(template, "choices.0.native_finish_reason", "tool_calls")
+			} else if finishReason != "" {
+				// Only pass through specific finish reasons
+				if finishReason == "max_tokens" || finishReason == "stop" {
+					template, _ = sjson.Set(template, "choices.0.finish_reason", finishReason)
+					template, _ = sjson.Set(template, "choices.0.native_finish_reason", finishReason)
+				}
 			}
 
 			responseStrings = append(responseStrings, template)
@@ -286,7 +297,7 @@ func ConvertGeminiResponseToOpenAINonStream(_ context.Context, _ string, origina
 		promptTokenCount := usageResult.Get("promptTokenCount").Int()
 		thoughtsTokenCount := usageResult.Get("thoughtsTokenCount").Int()
 		cachedTokenCount := usageResult.Get("cachedContentTokenCount").Int()
-		template, _ = sjson.Set(template, "usage.prompt_tokens", promptTokenCount+thoughtsTokenCount)
+		template, _ = sjson.Set(template, "usage.prompt_tokens", promptTokenCount)
 		if thoughtsTokenCount > 0 {
 			template, _ = sjson.Set(template, "usage.completion_tokens_details.reasoning_tokens", thoughtsTokenCount)
 		}

internal/translator/gemini/openai/responses/gemini_openai-responses_request.go

@@ -117,19 +117,29 @@ func ConvertOpenAIResponsesRequestToGemini(modelName string, inputRawJSON []byte
 			switch itemType {
 			case "message":
 				if strings.EqualFold(itemRole, "system") {
-					if contentArray := item.Get("content"); contentArray.Exists() && contentArray.IsArray() {
-						var builder strings.Builder
-						contentArray.ForEach(func(_, contentItem gjson.Result) bool {
-							text := contentItem.Get("text").String()
-							if builder.Len() > 0 && text != "" {
-								builder.WriteByte('\n')
-							}
-							builder.WriteString(text)
-							return true
-						})
-						if !gjson.Get(out, "system_instruction").Exists() {
-							systemInstr := `{"parts":[{"text":""}]}`
-							systemInstr, _ = sjson.Set(systemInstr, "parts.0.text", builder.String())
+					if contentArray := item.Get("content"); contentArray.Exists() {
+						systemInstr := ""
+						if systemInstructionResult := gjson.Get(out, "system_instruction"); systemInstructionResult.Exists() {
+							systemInstr = systemInstructionResult.Raw
+						} else {
+							systemInstr = `{"parts":[]}`
+						}
+
+						if contentArray.IsArray() {
+							contentArray.ForEach(func(_, contentItem gjson.Result) bool {
+								part := `{"text":""}`
+								text := contentItem.Get("text").String()
+								part, _ = sjson.Set(part, "text", text)
+								systemInstr, _ = sjson.SetRaw(systemInstr, "parts.-1", part)
+								return true
+							})
+						} else if contentArray.Type == gjson.String {
+							part := `{"text":""}`
+							part, _ = sjson.Set(part, "text", contentArray.String())
+							systemInstr, _ = sjson.SetRaw(systemInstr, "parts.-1", part)
+						}
+
+						if systemInstr != `{"parts":[]}` {
 							out, _ = sjson.SetRaw(out, "system_instruction", systemInstr)
 						}
 					}
@@ -236,8 +246,22 @@ func ConvertOpenAIResponsesRequestToGemini(modelName string, inputRawJSON []byte
 					})
 
 					flush()
-				}
+				} else if contentArray.Type == gjson.String {
+					effRole := "user"
+					if itemRole != "" {
+						switch strings.ToLower(itemRole) {
+						case "assistant", "model":
+							effRole = "model"
+						default:
+							effRole = strings.ToLower(itemRole)
+						}
+					}
 
+					one := `{"role":"","parts":[{"text":""}]}`
+					one, _ = sjson.Set(one, "role", effRole)
+					one, _ = sjson.Set(one, "parts.0.text", contentArray.String())
+					out, _ = sjson.SetRaw(out, "contents.-1", one)
+				}
 			case "function_call":
 				// Handle function calls - convert to model message with functionCall
 				name := item.Get("name").String()

internal/translator/gemini/openai/responses/gemini_openai-responses_response.go

@@ -531,8 +531,8 @@ func ConvertGeminiResponseToOpenAIResponses(_ context.Context, modelName string,
 
 		// usage mapping
 		if um := root.Get("usageMetadata"); um.Exists() {
-			// input tokens = prompt + thoughts
-			input := um.Get("promptTokenCount").Int() + um.Get("thoughtsTokenCount").Int()
+			// input tokens = prompt only (thoughts go to output)
+			input := um.Get("promptTokenCount").Int()
 			completed, _ = sjson.Set(completed, "response.usage.input_tokens", input)
 			// cached token details: align with OpenAI "cached_tokens" semantics.
 			completed, _ = sjson.Set(completed, "response.usage.input_tokens_details.cached_tokens", um.Get("cachedContentTokenCount").Int())
@@ -737,8 +737,8 @@ func ConvertGeminiResponseToOpenAIResponsesNonStream(_ context.Context, _ string
 
 	// usage mapping
 	if um := root.Get("usageMetadata"); um.Exists() {
-		// input tokens = prompt + thoughts
-		input := um.Get("promptTokenCount").Int() + um.Get("thoughtsTokenCount").Int()
+		// input tokens = prompt only (thoughts go to output)
+		input := um.Get("promptTokenCount").Int()
 		resp, _ = sjson.Set(resp, "usage.input_tokens", input)
 		// cached token details: align with OpenAI "cached_tokens" semantics.
 		resp, _ = sjson.Set(resp, "usage.input_tokens_details.cached_tokens", um.Get("cachedContentTokenCount").Int())

internal/translator/openai/claude/openai_claude_request.go

@@ -75,6 +75,10 @@ func ConvertClaudeRequestToOpenAI(modelName string, inputRawJSON []byte, stream
 						out, _ = sjson.Set(out, "reasoning_effort", effort)
 					}
 				}
+			case "adaptive":
+				// Claude adaptive means "enable with max capacity"; keep it as highest level
+				// and let ApplyThinking normalize per target model capability.
+				out, _ = sjson.Set(out, "reasoning_effort", string(thinking.LevelXHigh))
 			case "disabled":
 				if effort, ok := thinking.ConvertBudgetToLevel(0); ok && effort != "" {
 					out, _ = sjson.Set(out, "reasoning_effort", effort)

internal/translator/openai/openai/responses/openai_openai-responses_request.go

@@ -70,7 +70,7 @@ func ConvertOpenAIResponsesRequestToOpenAIChatCompletions(modelName string, inpu
 				if role == "developer" {
 					role = "user"
 				}
-				message := `{"role":"","content":""}`
+				message := `{"role":"","content":[]}`
 				message, _ = sjson.Set(message, "role", role)
 
 				if content := item.Get("content"); content.Exists() && content.IsArray() {
@@ -84,20 +84,16 @@ func ConvertOpenAIResponsesRequestToOpenAIChatCompletions(modelName string, inpu
 						}
 
 						switch contentType {
-						case "input_text":
+						case "input_text", "output_text":
 							text := contentItem.Get("text").String()
-							if messageContent != "" {
-								messageContent += "\n" + text
-							} else {
-								messageContent = text
-							}
-						case "output_text":
-							text := contentItem.Get("text").String()
-							if messageContent != "" {
-								messageContent += "\n" + text
-							} else {
-								messageContent = text
-							}
+							contentPart := `{"type":"text","text":""}`
+							contentPart, _ = sjson.Set(contentPart, "text", text)
+							message, _ = sjson.SetRaw(message, "content.-1", contentPart)
+						case "input_image":
+							imageURL := contentItem.Get("image_url").String()
+							contentPart := `{"type":"image_url","image_url":{"url":""}}`
+							contentPart, _ = sjson.Set(contentPart, "image_url.url", imageURL)
+							message, _ = sjson.SetRaw(message, "content.-1", contentPart)
 						}
 						return true
 					})

internal/tui/app.go

@@ -0,0 +1,542 @@
+package tui
+
+import (
+	"fmt"
+	"io"
+	"os"
+	"strings"
+
+	"github.com/charmbracelet/bubbles/textinput"
+	tea "github.com/charmbracelet/bubbletea"
+	"github.com/charmbracelet/lipgloss"
+)
+
+// Tab identifiers
+const (
+	tabDashboard = iota
+	tabConfig
+	tabAuthFiles
+	tabAPIKeys
+	tabOAuth
+	tabUsage
+	tabLogs
+)
+
+// App is the root bubbletea model that contains all tab sub-models.
+type App struct {
+	activeTab int
+	tabs      []string
+
+	standalone  bool
+	logsEnabled bool
+
+	authenticated  bool
+	authInput      textinput.Model
+	authError      string
+	authConnecting bool
+
+	dashboard dashboardModel
+	config    configTabModel
+	auth      authTabModel
+	keys      keysTabModel
+	oauth     oauthTabModel
+	usage     usageTabModel
+	logs      logsTabModel
+
+	client *Client
+
+	width  int
+	height int
+	ready  bool
+
+	// Track which tabs have been initialized (fetched data)
+	initialized [7]bool
+}
+
+type authConnectMsg struct {
+	cfg map[string]any
+	err error
+}
+
+// NewApp creates the root TUI application model.
+func NewApp(port int, secretKey string, hook *LogHook) App {
+	standalone := hook != nil
+	authRequired := !standalone
+	ti := textinput.New()
+	ti.CharLimit = 512
+	ti.EchoMode = textinput.EchoPassword
+	ti.EchoCharacter = '*'
+	ti.SetValue(strings.TrimSpace(secretKey))
+	ti.Focus()
+
+	client := NewClient(port, secretKey)
+	app := App{
+		activeTab:     tabDashboard,
+		standalone:    standalone,
+		logsEnabled:   true,
+		authenticated: !authRequired,
+		authInput:     ti,
+		dashboard:     newDashboardModel(client),
+		config:        newConfigTabModel(client),
+		auth:          newAuthTabModel(client),
+		keys:          newKeysTabModel(client),
+		oauth:         newOAuthTabModel(client),
+		usage:         newUsageTabModel(client),
+		logs:          newLogsTabModel(client, hook),
+		client:        client,
+		initialized: [7]bool{
+			tabDashboard: true,
+			tabLogs:      true,
+		},
+	}
+
+	app.refreshTabs()
+	if authRequired {
+		app.initialized = [7]bool{}
+	}
+	app.setAuthInputPrompt()
+	return app
+}
+
+func (a App) Init() tea.Cmd {
+	if !a.authenticated {
+		return textinput.Blink
+	}
+	cmds := []tea.Cmd{a.dashboard.Init()}
+	if a.logsEnabled {
+		cmds = append(cmds, a.logs.Init())
+	}
+	return tea.Batch(cmds...)
+}
+
+func (a App) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
+	switch msg := msg.(type) {
+	case tea.WindowSizeMsg:
+		a.width = msg.Width
+		a.height = msg.Height
+		a.ready = true
+		if a.width > 0 {
+			a.authInput.Width = a.width - 6
+		}
+		contentH := a.height - 4 // tab bar + status bar
+		if contentH < 1 {
+			contentH = 1
+		}
+		contentW := a.width
+		a.dashboard.SetSize(contentW, contentH)
+		a.config.SetSize(contentW, contentH)
+		a.auth.SetSize(contentW, contentH)
+		a.keys.SetSize(contentW, contentH)
+		a.oauth.SetSize(contentW, contentH)
+		a.usage.SetSize(contentW, contentH)
+		a.logs.SetSize(contentW, contentH)
+		return a, nil
+
+	case authConnectMsg:
+		a.authConnecting = false
+		if msg.err != nil {
+			a.authError = fmt.Sprintf(T("auth_gate_connect_fail"), msg.err.Error())
+			return a, nil
+		}
+		a.authError = ""
+		a.authenticated = true
+		a.logsEnabled = a.standalone || isLogsEnabledFromConfig(msg.cfg)
+		a.refreshTabs()
+		a.initialized = [7]bool{}
+		a.initialized[tabDashboard] = true
+		cmds := []tea.Cmd{a.dashboard.Init()}
+		if a.logsEnabled {
+			a.initialized[tabLogs] = true
+			cmds = append(cmds, a.logs.Init())
+		}
+		return a, tea.Batch(cmds...)
+
+	case configUpdateMsg:
+		var cmdLogs tea.Cmd
+		if !a.standalone && msg.err == nil && msg.path == "logging-to-file" {
+			logsEnabledConfig, okConfig := msg.value.(bool)
+			if okConfig {
+				logsEnabledBefore := a.logsEnabled
+				a.logsEnabled = logsEnabledConfig
+				if logsEnabledBefore != a.logsEnabled {
+					a.refreshTabs()
+				}
+				if !a.logsEnabled {
+					a.initialized[tabLogs] = false
+				}
+				if !logsEnabledBefore && a.logsEnabled {
+					a.initialized[tabLogs] = true
+					cmdLogs = a.logs.Init()
+				}
+			}
+		}
+
+		var cmdConfig tea.Cmd
+		a.config, cmdConfig = a.config.Update(msg)
+		if cmdConfig != nil && cmdLogs != nil {
+			return a, tea.Batch(cmdConfig, cmdLogs)
+		}
+		if cmdConfig != nil {
+			return a, cmdConfig
+		}
+		return a, cmdLogs
+
+	case tea.KeyMsg:
+		if !a.authenticated {
+			switch msg.String() {
+			case "ctrl+c", "q":
+				return a, tea.Quit
+			case "L":
+				ToggleLocale()
+				a.refreshTabs()
+				a.setAuthInputPrompt()
+				return a, nil
+			case "enter":
+				if a.authConnecting {
+					return a, nil
+				}
+				password := strings.TrimSpace(a.authInput.Value())
+				if password == "" {
+					a.authError = T("auth_gate_password_required")
+					return a, nil
+				}
+				a.authError = ""
+				a.authConnecting = true
+				return a, a.connectWithPassword(password)
+			default:
+				var cmd tea.Cmd
+				a.authInput, cmd = a.authInput.Update(msg)
+				return a, cmd
+			}
+		}
+
+		switch msg.String() {
+		case "ctrl+c":
+			return a, tea.Quit
+		case "q":
+			// Only quit if not in logs tab (where 'q' might be useful)
+			if !a.logsEnabled || a.activeTab != tabLogs {
+				return a, tea.Quit
+			}
+		case "L":
+			ToggleLocale()
+			a.refreshTabs()
+			return a.broadcastToAllTabs(localeChangedMsg{})
+		case "tab":
+			if len(a.tabs) == 0 {
+				return a, nil
+			}
+			prevTab := a.activeTab
+			a.activeTab = (a.activeTab + 1) % len(a.tabs)
+			return a, a.initTabIfNeeded(prevTab)
+		case "shift+tab":
+			if len(a.tabs) == 0 {
+				return a, nil
+			}
+			prevTab := a.activeTab
+			a.activeTab = (a.activeTab - 1 + len(a.tabs)) % len(a.tabs)
+			return a, a.initTabIfNeeded(prevTab)
+		}
+	}
+
+	if !a.authenticated {
+		var cmd tea.Cmd
+		a.authInput, cmd = a.authInput.Update(msg)
+		return a, cmd
+	}
+
+	// Route msg to active tab
+	var cmd tea.Cmd
+	switch a.activeTab {
+	case tabDashboard:
+		a.dashboard, cmd = a.dashboard.Update(msg)
+	case tabConfig:
+		a.config, cmd = a.config.Update(msg)
+	case tabAuthFiles:
+		a.auth, cmd = a.auth.Update(msg)
+	case tabAPIKeys:
+		a.keys, cmd = a.keys.Update(msg)
+	case tabOAuth:
+		a.oauth, cmd = a.oauth.Update(msg)
+	case tabUsage:
+		a.usage, cmd = a.usage.Update(msg)
+	case tabLogs:
+		a.logs, cmd = a.logs.Update(msg)
+	}
+
+	// Keep logs polling alive even when logs tab is not active.
+	if a.logsEnabled && a.activeTab != tabLogs {
+		switch msg.(type) {
+		case logsPollMsg, logsTickMsg, logLineMsg:
+			var logCmd tea.Cmd
+			a.logs, logCmd = a.logs.Update(msg)
+			if logCmd != nil {
+				cmd = logCmd
+			}
+		}
+	}
+
+	return a, cmd
+}
+
+// localeChangedMsg is broadcast to all tabs when the user toggles locale.
+type localeChangedMsg struct{}
+
+func (a *App) refreshTabs() {
+	names := TabNames()
+	if a.logsEnabled {
+		a.tabs = names
+	} else {
+		filtered := make([]string, 0, len(names)-1)
+		for idx, name := range names {
+			if idx == tabLogs {
+				continue
+			}
+			filtered = append(filtered, name)
+		}
+		a.tabs = filtered
+	}
+
+	if len(a.tabs) == 0 {
+		a.activeTab = tabDashboard
+		return
+	}
+	if a.activeTab >= len(a.tabs) {
+		a.activeTab = len(a.tabs) - 1
+	}
+}
+
+func (a *App) initTabIfNeeded(_ int) tea.Cmd {
+	if a.initialized[a.activeTab] {
+		return nil
+	}
+	a.initialized[a.activeTab] = true
+	switch a.activeTab {
+	case tabDashboard:
+		return a.dashboard.Init()
+	case tabConfig:
+		return a.config.Init()
+	case tabAuthFiles:
+		return a.auth.Init()
+	case tabAPIKeys:
+		return a.keys.Init()
+	case tabOAuth:
+		return a.oauth.Init()
+	case tabUsage:
+		return a.usage.Init()
+	case tabLogs:
+		if !a.logsEnabled {
+			return nil
+		}
+		return a.logs.Init()
+	}
+	return nil
+}
+
+func (a App) View() string {
+	if !a.authenticated {
+		return a.renderAuthView()
+	}
+
+	if !a.ready {
+		return T("initializing_tui")
+	}
+
+	var sb strings.Builder
+
+	// Tab bar
+	sb.WriteString(a.renderTabBar())
+	sb.WriteString("\n")
+
+	// Content
+	switch a.activeTab {
+	case tabDashboard:
+		sb.WriteString(a.dashboard.View())
+	case tabConfig:
+		sb.WriteString(a.config.View())
+	case tabAuthFiles:
+		sb.WriteString(a.auth.View())
+	case tabAPIKeys:
+		sb.WriteString(a.keys.View())
+	case tabOAuth:
+		sb.WriteString(a.oauth.View())
+	case tabUsage:
+		sb.WriteString(a.usage.View())
+	case tabLogs:
+		if a.logsEnabled {
+			sb.WriteString(a.logs.View())
+		}
+	}
+
+	// Status bar
+	sb.WriteString("\n")
+	sb.WriteString(a.renderStatusBar())
+
+	return sb.String()
+}
+
+func (a App) renderAuthView() string {
+	var sb strings.Builder
+
+	sb.WriteString(titleStyle.Render(T("auth_gate_title")))
+	sb.WriteString("\n")
+	sb.WriteString(helpStyle.Render(T("auth_gate_help")))
+	sb.WriteString("\n\n")
+	if a.authConnecting {
+		sb.WriteString(warningStyle.Render(T("auth_gate_connecting")))
+		sb.WriteString("\n\n")
+	}
+	if strings.TrimSpace(a.authError) != "" {
+		sb.WriteString(errorStyle.Render(a.authError))
+		sb.WriteString("\n\n")
+	}
+	sb.WriteString(a.authInput.View())
+	sb.WriteString("\n")
+	sb.WriteString(helpStyle.Render(T("auth_gate_enter")))
+	return sb.String()
+}
+
+func (a App) renderTabBar() string {
+	var tabs []string
+	for i, name := range a.tabs {
+		if i == a.activeTab {
+			tabs = append(tabs, tabActiveStyle.Render(name))
+		} else {
+			tabs = append(tabs, tabInactiveStyle.Render(name))
+		}
+	}
+	tabBar := lipgloss.JoinHorizontal(lipgloss.Top, tabs...)
+	return tabBarStyle.Width(a.width).Render(tabBar)
+}
+
+func (a App) renderStatusBar() string {
+	left := strings.TrimRight(T("status_left"), " ")
+	right := strings.TrimRight(T("status_right"), " ")
+
+	width := a.width
+	if width < 1 {
+		width = 1
+	}
+
+	// statusBarStyle has left/right padding(1), so content area is width-2.
+	contentWidth := width - 2
+	if contentWidth < 0 {
+		contentWidth = 0
+	}
+
+	if lipgloss.Width(left) > contentWidth {
+		left = fitStringWidth(left, contentWidth)
+		right = ""
+	}
+
+	remaining := contentWidth - lipgloss.Width(left)
+	if remaining < 0 {
+		remaining = 0
+	}
+	if lipgloss.Width(right) > remaining {
+		right = fitStringWidth(right, remaining)
+	}
+
+	gap := contentWidth - lipgloss.Width(left) - lipgloss.Width(right)
+	if gap < 0 {
+		gap = 0
+	}
+	return statusBarStyle.Width(width).Render(left + strings.Repeat(" ", gap) + right)
+}
+
+func fitStringWidth(text string, maxWidth int) string {
+	if maxWidth <= 0 {
+		return ""
+	}
+	if lipgloss.Width(text) <= maxWidth {
+		return text
+	}
+
+	out := ""
+	for _, r := range text {
+		next := out + string(r)
+		if lipgloss.Width(next) > maxWidth {
+			break
+		}
+		out = next
+	}
+	return out
+}
+
+func isLogsEnabledFromConfig(cfg map[string]any) bool {
+	if cfg == nil {
+		return true
+	}
+	value, ok := cfg["logging-to-file"]
+	if !ok {
+		return true
+	}
+	enabled, ok := value.(bool)
+	if !ok {
+		return true
+	}
+	return enabled
+}
+
+func (a *App) setAuthInputPrompt() {
+	if a == nil {
+		return
+	}
+	a.authInput.Prompt = fmt.Sprintf("  %s: ", T("auth_gate_password"))
+}
+
+func (a App) connectWithPassword(password string) tea.Cmd {
+	return func() tea.Msg {
+		a.client.SetSecretKey(password)
+		cfg, errGetConfig := a.client.GetConfig()
+		return authConnectMsg{cfg: cfg, err: errGetConfig}
+	}
+}
+
+// Run starts the TUI application.
+// output specifies where bubbletea renders. If nil, defaults to os.Stdout.
+func Run(port int, secretKey string, hook *LogHook, output io.Writer) error {
+	if output == nil {
+		output = os.Stdout
+	}
+	app := NewApp(port, secretKey, hook)
+	p := tea.NewProgram(app, tea.WithAltScreen(), tea.WithOutput(output))
+	_, err := p.Run()
+	return err
+}
+
+func (a App) broadcastToAllTabs(msg tea.Msg) (tea.Model, tea.Cmd) {
+	var cmds []tea.Cmd
+	var cmd tea.Cmd
+
+	a.dashboard, cmd = a.dashboard.Update(msg)
+	if cmd != nil {
+		cmds = append(cmds, cmd)
+	}
+	a.config, cmd = a.config.Update(msg)
+	if cmd != nil {
+		cmds = append(cmds, cmd)
+	}
+	a.auth, cmd = a.auth.Update(msg)
+	if cmd != nil {
+		cmds = append(cmds, cmd)
+	}
+	a.keys, cmd = a.keys.Update(msg)
+	if cmd != nil {
+		cmds = append(cmds, cmd)
+	}
+	a.oauth, cmd = a.oauth.Update(msg)
+	if cmd != nil {
+		cmds = append(cmds, cmd)
+	}
+	a.usage, cmd = a.usage.Update(msg)
+	if cmd != nil {
+		cmds = append(cmds, cmd)
+	}
+	a.logs, cmd = a.logs.Update(msg)
+	if cmd != nil {
+		cmds = append(cmds, cmd)
+	}
+
+	return a, tea.Batch(cmds...)
+}

internal/tui/auth_tab.go

@@ -0,0 +1,456 @@
+package tui
+
+import (
+	"fmt"
+	"strconv"
+	"strings"
+
+	"github.com/charmbracelet/bubbles/textinput"
+	"github.com/charmbracelet/bubbles/viewport"
+	tea "github.com/charmbracelet/bubbletea"
+	"github.com/charmbracelet/lipgloss"
+)
+
+// editableField represents an editable field on an auth file.
+type editableField struct {
+	label string
+	key   string // API field key: "prefix", "proxy_url", "priority"
+}
+
+var authEditableFields = []editableField{
+	{label: "Prefix", key: "prefix"},
+	{label: "Proxy URL", key: "proxy_url"},
+	{label: "Priority", key: "priority"},
+}
+
+// authTabModel displays auth credential files with interactive management.
+type authTabModel struct {
+	client   *Client
+	viewport viewport.Model
+	files    []map[string]any
+	err      error
+	width    int
+	height   int
+	ready    bool
+	cursor   int
+	expanded int // -1 = none expanded, >=0 = expanded index
+	confirm  int // -1 = no confirmation, >=0 = confirm delete for index
+	status   string
+
+	// Editing state
+	editing      bool            // true when editing a field
+	editField    int             // index into authEditableFields
+	editInput    textinput.Model // text input for editing
+	editFileName string          // name of file being edited
+}
+
+type authFilesMsg struct {
+	files []map[string]any
+	err   error
+}
+
+type authActionMsg struct {
+	action string // "deleted", "toggled", "updated"
+	err    error
+}
+
+func newAuthTabModel(client *Client) authTabModel {
+	ti := textinput.New()
+	ti.CharLimit = 256
+	return authTabModel{
+		client:    client,
+		expanded:  -1,
+		confirm:   -1,
+		editInput: ti,
+	}
+}
+
+func (m authTabModel) Init() tea.Cmd {
+	return m.fetchFiles
+}
+
+func (m authTabModel) fetchFiles() tea.Msg {
+	files, err := m.client.GetAuthFiles()
+	return authFilesMsg{files: files, err: err}
+}
+
+func (m authTabModel) Update(msg tea.Msg) (authTabModel, tea.Cmd) {
+	switch msg := msg.(type) {
+	case localeChangedMsg:
+		m.viewport.SetContent(m.renderContent())
+		return m, nil
+	case authFilesMsg:
+		if msg.err != nil {
+			m.err = msg.err
+		} else {
+			m.err = nil
+			m.files = msg.files
+			if m.cursor >= len(m.files) {
+				m.cursor = max(0, len(m.files)-1)
+			}
+			m.status = ""
+		}
+		m.viewport.SetContent(m.renderContent())
+		return m, nil
+
+	case authActionMsg:
+		if msg.err != nil {
+			m.status = errorStyle.Render("✗ " + msg.err.Error())
+		} else {
+			m.status = successStyle.Render("✓ " + msg.action)
+		}
+		m.confirm = -1
+		m.viewport.SetContent(m.renderContent())
+		return m, m.fetchFiles
+
+	case tea.KeyMsg:
+		// ---- Editing mode ----
+		if m.editing {
+			return m.handleEditInput(msg)
+		}
+
+		// ---- Delete confirmation mode ----
+		if m.confirm >= 0 {
+			return m.handleConfirmInput(msg)
+		}
+
+		// ---- Normal mode ----
+		return m.handleNormalInput(msg)
+	}
+
+	var cmd tea.Cmd
+	m.viewport, cmd = m.viewport.Update(msg)
+	return m, cmd
+}
+
+// startEdit activates inline editing for a field on the currently selected auth file.
+func (m *authTabModel) startEdit(fieldIdx int) tea.Cmd {
+	if m.cursor >= len(m.files) {
+		return nil
+	}
+	f := m.files[m.cursor]
+	m.editFileName = getString(f, "name")
+	m.editField = fieldIdx
+	m.editing = true
+
+	// Pre-populate with current value
+	key := authEditableFields[fieldIdx].key
+	currentVal := getAnyString(f, key)
+	m.editInput.SetValue(currentVal)
+	m.editInput.Focus()
+	m.editInput.Prompt = fmt.Sprintf("  %s: ", authEditableFields[fieldIdx].label)
+	m.viewport.SetContent(m.renderContent())
+	return textinput.Blink
+}
+
+func (m *authTabModel) SetSize(w, h int) {
+	m.width = w
+	m.height = h
+	m.editInput.Width = w - 20
+	if !m.ready {
+		m.viewport = viewport.New(w, h)
+		m.viewport.SetContent(m.renderContent())
+		m.ready = true
+	} else {
+		m.viewport.Width = w
+		m.viewport.Height = h
+	}
+}
+
+func (m authTabModel) View() string {
+	if !m.ready {
+		return T("loading")
+	}
+	return m.viewport.View()
+}
+
+func (m authTabModel) renderContent() string {
+	var sb strings.Builder
+
+	sb.WriteString(titleStyle.Render(T("auth_title")))
+	sb.WriteString("\n")
+	sb.WriteString(helpStyle.Render(T("auth_help1")))
+	sb.WriteString("\n")
+	sb.WriteString(helpStyle.Render(T("auth_help2")))
+	sb.WriteString("\n")
+	sb.WriteString(strings.Repeat("─", m.width))
+	sb.WriteString("\n")
+
+	if m.err != nil {
+		sb.WriteString(errorStyle.Render("⚠ Error: " + m.err.Error()))
+		sb.WriteString("\n")
+		return sb.String()
+	}
+
+	if len(m.files) == 0 {
+		sb.WriteString(subtitleStyle.Render(T("no_auth_files")))
+		sb.WriteString("\n")
+		return sb.String()
+	}
+
+	for i, f := range m.files {
+		name := getString(f, "name")
+		channel := getString(f, "channel")
+		email := getString(f, "email")
+		disabled := getBool(f, "disabled")
+
+		statusIcon := successStyle.Render("●")
+		statusText := T("status_active")
+		if disabled {
+			statusIcon = lipgloss.NewStyle().Foreground(colorMuted).Render("○")
+			statusText = T("status_disabled")
+		}
+
+		cursor := "  "
+		rowStyle := lipgloss.NewStyle()
+		if i == m.cursor {
+			cursor = "▸ "
+			rowStyle = lipgloss.NewStyle().Bold(true)
+		}
+
+		displayName := name
+		if len(displayName) > 24 {
+			displayName = displayName[:21] + "..."
+		}
+		displayEmail := email
+		if len(displayEmail) > 28 {
+			displayEmail = displayEmail[:25] + "..."
+		}
+
+		row := fmt.Sprintf("%s%s %-24s %-12s %-28s %s",
+			cursor, statusIcon, displayName, channel, displayEmail, statusText)
+		sb.WriteString(rowStyle.Render(row))
+		sb.WriteString("\n")
+
+		// Delete confirmation
+		if m.confirm == i {
+			sb.WriteString(warningStyle.Render(fmt.Sprintf("    "+T("confirm_delete"), name)))
+			sb.WriteString("\n")
+		}
+
+		// Inline edit input
+		if m.editing && i == m.cursor {
+			sb.WriteString(m.editInput.View())
+			sb.WriteString("\n")
+			sb.WriteString(helpStyle.Render("    " + T("enter_save") + " • " + T("esc_cancel")))
+			sb.WriteString("\n")
+		}
+
+		// Expanded detail view
+		if m.expanded == i {
+			sb.WriteString(m.renderDetail(f))
+		}
+	}
+
+	if m.status != "" {
+		sb.WriteString("\n")
+		sb.WriteString(m.status)
+		sb.WriteString("\n")
+	}
+
+	return sb.String()
+}
+
+func (m authTabModel) renderDetail(f map[string]any) string {
+	var sb strings.Builder
+
+	labelStyle := lipgloss.NewStyle().
+		Foreground(lipgloss.Color("111")).
+		Bold(true)
+	valueStyle := lipgloss.NewStyle().
+		Foreground(lipgloss.Color("252"))
+	editableMarker := lipgloss.NewStyle().
+		Foreground(lipgloss.Color("214")).
+		Render(" ✎")
+
+	sb.WriteString("    ┌─────────────────────────────────────────────\n")
+
+	fields := []struct {
+		label    string
+		key      string
+		editable bool
+	}{
+		{"Name", "name", false},
+		{"Channel", "channel", false},
+		{"Email", "email", false},
+		{"Status", "status", false},
+		{"Status Msg", "status_message", false},
+		{"File Name", "file_name", false},
+		{"Auth Type", "auth_type", false},
+		{"Prefix", "prefix", true},
+		{"Proxy URL", "proxy_url", true},
+		{"Priority", "priority", true},
+		{"Project ID", "project_id", false},
+		{"Disabled", "disabled", false},
+		{"Created", "created_at", false},
+		{"Updated", "updated_at", false},
+	}
+
+	for _, field := range fields {
+		val := getAnyString(f, field.key)
+		if val == "" || val == "<nil>" {
+			if field.editable {
+				val = T("not_set")
+			} else {
+				continue
+			}
+		}
+		editMark := ""
+		if field.editable {
+			editMark = editableMarker
+		}
+		line := fmt.Sprintf("    │ %s %s%s",
+			labelStyle.Render(fmt.Sprintf("%-12s:", field.label)),
+			valueStyle.Render(val),
+			editMark)
+		sb.WriteString(line)
+		sb.WriteString("\n")
+	}
+
+	sb.WriteString("    └─────────────────────────────────────────────\n")
+	return sb.String()
+}
+
+// getAnyString converts any value to its string representation.
+func getAnyString(m map[string]any, key string) string {
+	v, ok := m[key]
+	if !ok || v == nil {
+		return ""
+	}
+	return fmt.Sprintf("%v", v)
+}
+
+func max(a, b int) int {
+	if a > b {
+		return a
+	}
+	return b
+}
+
+func (m authTabModel) handleEditInput(msg tea.KeyMsg) (authTabModel, tea.Cmd) {
+	switch msg.String() {
+	case "enter":
+		value := m.editInput.Value()
+		fieldKey := authEditableFields[m.editField].key
+		fileName := m.editFileName
+		m.editing = false
+		m.editInput.Blur()
+		fields := map[string]any{}
+		if fieldKey == "priority" {
+			p, err := strconv.Atoi(value)
+			if err != nil {
+				return m, func() tea.Msg {
+					return authActionMsg{err: fmt.Errorf("%s: %s", T("invalid_int"), value)}
+				}
+			}
+			fields[fieldKey] = p
+		} else {
+			fields[fieldKey] = value
+		}
+		return m, func() tea.Msg {
+			err := m.client.PatchAuthFileFields(fileName, fields)
+			if err != nil {
+				return authActionMsg{err: err}
+			}
+			return authActionMsg{action: fmt.Sprintf(T("updated_field"), fieldKey, fileName)}
+		}
+	case "esc":
+		m.editing = false
+		m.editInput.Blur()
+		m.viewport.SetContent(m.renderContent())
+		return m, nil
+	default:
+		var cmd tea.Cmd
+		m.editInput, cmd = m.editInput.Update(msg)
+		m.viewport.SetContent(m.renderContent())
+		return m, cmd
+	}
+}
+
+func (m authTabModel) handleConfirmInput(msg tea.KeyMsg) (authTabModel, tea.Cmd) {
+	switch msg.String() {
+	case "y", "Y":
+		idx := m.confirm
+		m.confirm = -1
+		if idx < len(m.files) {
+			name := getString(m.files[idx], "name")
+			return m, func() tea.Msg {
+				err := m.client.DeleteAuthFile(name)
+				if err != nil {
+					return authActionMsg{err: err}
+				}
+				return authActionMsg{action: fmt.Sprintf(T("deleted"), name)}
+			}
+		}
+		m.viewport.SetContent(m.renderContent())
+		return m, nil
+	case "n", "N", "esc":
+		m.confirm = -1
+		m.viewport.SetContent(m.renderContent())
+		return m, nil
+	}
+	return m, nil
+}
+
+func (m authTabModel) handleNormalInput(msg tea.KeyMsg) (authTabModel, tea.Cmd) {
+	switch msg.String() {
+	case "j", "down":
+		if len(m.files) > 0 {
+			m.cursor = (m.cursor + 1) % len(m.files)
+			m.viewport.SetContent(m.renderContent())
+		}
+		return m, nil
+	case "k", "up":
+		if len(m.files) > 0 {
+			m.cursor = (m.cursor - 1 + len(m.files)) % len(m.files)
+			m.viewport.SetContent(m.renderContent())
+		}
+		return m, nil
+	case "enter", " ":
+		if m.expanded == m.cursor {
+			m.expanded = -1
+		} else {
+			m.expanded = m.cursor
+		}
+		m.viewport.SetContent(m.renderContent())
+		return m, nil
+	case "d", "D":
+		if m.cursor < len(m.files) {
+			m.confirm = m.cursor
+			m.viewport.SetContent(m.renderContent())
+		}
+		return m, nil
+	case "e", "E":
+		if m.cursor < len(m.files) {
+			f := m.files[m.cursor]
+			name := getString(f, "name")
+			disabled := getBool(f, "disabled")
+			newDisabled := !disabled
+			return m, func() tea.Msg {
+				err := m.client.ToggleAuthFile(name, newDisabled)
+				if err != nil {
+					return authActionMsg{err: err}
+				}
+				action := T("enabled")
+				if newDisabled {
+					action = T("disabled")
+				}
+				return authActionMsg{action: fmt.Sprintf("%s %s", action, name)}
+			}
+		}
+		return m, nil
+	case "1":
+		return m, m.startEdit(0) // prefix
+	case "2":
+		return m, m.startEdit(1) // proxy_url
+	case "3":
+		return m, m.startEdit(2) // priority
+	case "r":
+		m.status = ""
+		return m, m.fetchFiles
+	default:
+		var cmd tea.Cmd
+		m.viewport, cmd = m.viewport.Update(msg)
+		return m, cmd
+	}
+}

internal/tui/browser.go

@@ -0,0 +1,20 @@
+package tui
+
+import (
+	"os/exec"
+	"runtime"
+)
+
+// openBrowser opens the specified URL in the user's default browser.
+func openBrowser(url string) error {
+	switch runtime.GOOS {
+	case "darwin":
+		return exec.Command("open", url).Start()
+	case "linux":
+		return exec.Command("xdg-open", url).Start()
+	case "windows":
+		return exec.Command("rundll32", "url.dll,FileProtocolHandler", url).Start()
+	default:
+		return exec.Command("xdg-open", url).Start()
+	}
+}

internal/tui/client.go

@@ -0,0 +1,400 @@
+package tui
+
+import (
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+	"strconv"
+	"strings"
+	"time"
+)
+
+// Client wraps HTTP calls to the management API.
+type Client struct {
+	baseURL   string
+	secretKey string
+	http      *http.Client
+}
+
+// NewClient creates a new management API client.
+func NewClient(port int, secretKey string) *Client {
+	return &Client{
+		baseURL:   fmt.Sprintf("http://127.0.0.1:%d", port),
+		secretKey: strings.TrimSpace(secretKey),
+		http: &http.Client{
+			Timeout: 10 * time.Second,
+		},
+	}
+}
+
+// SetSecretKey updates management API bearer token used by this client.
+func (c *Client) SetSecretKey(secretKey string) {
+	c.secretKey = strings.TrimSpace(secretKey)
+}
+
+func (c *Client) doRequest(method, path string, body io.Reader) ([]byte, int, error) {
+	url := c.baseURL + path
+	req, err := http.NewRequest(method, url, body)
+	if err != nil {
+		return nil, 0, err
+	}
+	if c.secretKey != "" {
+		req.Header.Set("Authorization", "Bearer "+c.secretKey)
+	}
+	if body != nil {
+		req.Header.Set("Content-Type", "application/json")
+	}
+	resp, err := c.http.Do(req)
+	if err != nil {
+		return nil, 0, err
+	}
+	defer resp.Body.Close()
+	data, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, resp.StatusCode, err
+	}
+	return data, resp.StatusCode, nil
+}
+
+func (c *Client) get(path string) ([]byte, error) {
+	data, code, err := c.doRequest("GET", path, nil)
+	if err != nil {
+		return nil, err
+	}
+	if code >= 400 {
+		return nil, fmt.Errorf("HTTP %d: %s", code, strings.TrimSpace(string(data)))
+	}
+	return data, nil
+}
+
+func (c *Client) put(path string, body io.Reader) ([]byte, error) {
+	data, code, err := c.doRequest("PUT", path, body)
+	if err != nil {
+		return nil, err
+	}
+	if code >= 400 {
+		return nil, fmt.Errorf("HTTP %d: %s", code, strings.TrimSpace(string(data)))
+	}
+	return data, nil
+}
+
+func (c *Client) patch(path string, body io.Reader) ([]byte, error) {
+	data, code, err := c.doRequest("PATCH", path, body)
+	if err != nil {
+		return nil, err
+	}
+	if code >= 400 {
+		return nil, fmt.Errorf("HTTP %d: %s", code, strings.TrimSpace(string(data)))
+	}
+	return data, nil
+}
+
+// getJSON fetches a path and unmarshals JSON into a generic map.
+func (c *Client) getJSON(path string) (map[string]any, error) {
+	data, err := c.get(path)
+	if err != nil {
+		return nil, err
+	}
+	var result map[string]any
+	if err := json.Unmarshal(data, &result); err != nil {
+		return nil, err
+	}
+	return result, nil
+}
+
+// postJSON sends a JSON body via POST and checks for errors.
+func (c *Client) postJSON(path string, body any) error {
+	jsonBody, err := json.Marshal(body)
+	if err != nil {
+		return err
+	}
+	_, code, err := c.doRequest("POST", path, strings.NewReader(string(jsonBody)))
+	if err != nil {
+		return err
+	}
+	if code >= 400 {
+		return fmt.Errorf("HTTP %d", code)
+	}
+	return nil
+}
+
+// GetConfig fetches the parsed config.
+func (c *Client) GetConfig() (map[string]any, error) {
+	return c.getJSON("/v0/management/config")
+}
+
+// GetConfigYAML fetches the raw config.yaml content.
+func (c *Client) GetConfigYAML() (string, error) {
+	data, err := c.get("/v0/management/config.yaml")
+	if err != nil {
+		return "", err
+	}
+	return string(data), nil
+}
+
+// PutConfigYAML uploads new config.yaml content.
+func (c *Client) PutConfigYAML(yamlContent string) error {
+	_, err := c.put("/v0/management/config.yaml", strings.NewReader(yamlContent))
+	return err
+}
+
+// GetUsage fetches usage statistics.
+func (c *Client) GetUsage() (map[string]any, error) {
+	return c.getJSON("/v0/management/usage")
+}
+
+// GetAuthFiles lists auth credential files.
+// API returns {"files": [...]}.
+func (c *Client) GetAuthFiles() ([]map[string]any, error) {
+	wrapper, err := c.getJSON("/v0/management/auth-files")
+	if err != nil {
+		return nil, err
+	}
+	return extractList(wrapper, "files")
+}
+
+// DeleteAuthFile deletes a single auth file by name.
+func (c *Client) DeleteAuthFile(name string) error {
+	query := url.Values{}
+	query.Set("name", name)
+	path := "/v0/management/auth-files?" + query.Encode()
+	_, code, err := c.doRequest("DELETE", path, nil)
+	if err != nil {
+		return err
+	}
+	if code >= 400 {
+		return fmt.Errorf("delete failed (HTTP %d)", code)
+	}
+	return nil
+}
+
+// ToggleAuthFile enables or disables an auth file.
+func (c *Client) ToggleAuthFile(name string, disabled bool) error {
+	body, _ := json.Marshal(map[string]any{"name": name, "disabled": disabled})
+	_, err := c.patch("/v0/management/auth-files/status", strings.NewReader(string(body)))
+	return err
+}
+
+// PatchAuthFileFields updates editable fields on an auth file.
+func (c *Client) PatchAuthFileFields(name string, fields map[string]any) error {
+	fields["name"] = name
+	body, _ := json.Marshal(fields)
+	_, err := c.patch("/v0/management/auth-files/fields", strings.NewReader(string(body)))
+	return err
+}
+
+// GetLogs fetches log lines from the server.
+func (c *Client) GetLogs(after int64, limit int) ([]string, int64, error) {
+	query := url.Values{}
+	if limit > 0 {
+		query.Set("limit", strconv.Itoa(limit))
+	}
+	if after > 0 {
+		query.Set("after", strconv.FormatInt(after, 10))
+	}
+
+	path := "/v0/management/logs"
+	encodedQuery := query.Encode()
+	if encodedQuery != "" {
+		path += "?" + encodedQuery
+	}
+
+	wrapper, err := c.getJSON(path)
+	if err != nil {
+		return nil, after, err
+	}
+
+	lines := []string{}
+	if rawLines, ok := wrapper["lines"]; ok && rawLines != nil {
+		rawJSON, errMarshal := json.Marshal(rawLines)
+		if errMarshal != nil {
+			return nil, after, errMarshal
+		}
+		if errUnmarshal := json.Unmarshal(rawJSON, &lines); errUnmarshal != nil {
+			return nil, after, errUnmarshal
+		}
+	}
+
+	latest := after
+	if rawLatest, ok := wrapper["latest-timestamp"]; ok {
+		switch value := rawLatest.(type) {
+		case float64:
+			latest = int64(value)
+		case json.Number:
+			if parsed, errParse := value.Int64(); errParse == nil {
+				latest = parsed
+			}
+		case int64:
+			latest = value
+		case int:
+			latest = int64(value)
+		}
+	}
+	if latest < after {
+		latest = after
+	}
+
+	return lines, latest, nil
+}
+
+// GetAPIKeys fetches the list of API keys.
+// API returns {"api-keys": [...]}.
+func (c *Client) GetAPIKeys() ([]string, error) {
+	wrapper, err := c.getJSON("/v0/management/api-keys")
+	if err != nil {
+		return nil, err
+	}
+	arr, ok := wrapper["api-keys"]
+	if !ok {
+		return nil, nil
+	}
+	raw, err := json.Marshal(arr)
+	if err != nil {
+		return nil, err
+	}
+	var result []string
+	if err := json.Unmarshal(raw, &result); err != nil {
+		return nil, err
+	}
+	return result, nil
+}
+
+// AddAPIKey adds a new API key by sending old=nil, new=key which appends.
+func (c *Client) AddAPIKey(key string) error {
+	body := map[string]any{"old": nil, "new": key}
+	jsonBody, _ := json.Marshal(body)
+	_, err := c.patch("/v0/management/api-keys", strings.NewReader(string(jsonBody)))
+	return err
+}
+
+// EditAPIKey replaces an API key at the given index.
+func (c *Client) EditAPIKey(index int, newValue string) error {
+	body := map[string]any{"index": index, "value": newValue}
+	jsonBody, _ := json.Marshal(body)
+	_, err := c.patch("/v0/management/api-keys", strings.NewReader(string(jsonBody)))
+	return err
+}
+
+// DeleteAPIKey deletes an API key by index.
+func (c *Client) DeleteAPIKey(index int) error {
+	_, code, err := c.doRequest("DELETE", fmt.Sprintf("/v0/management/api-keys?index=%d", index), nil)
+	if err != nil {
+		return err
+	}
+	if code >= 400 {
+		return fmt.Errorf("delete failed (HTTP %d)", code)
+	}
+	return nil
+}
+
+// GetGeminiKeys fetches Gemini API keys.
+// API returns {"gemini-api-key": [...]}.
+func (c *Client) GetGeminiKeys() ([]map[string]any, error) {
+	return c.getWrappedKeyList("/v0/management/gemini-api-key", "gemini-api-key")
+}
+
+// GetClaudeKeys fetches Claude API keys.
+func (c *Client) GetClaudeKeys() ([]map[string]any, error) {
+	return c.getWrappedKeyList("/v0/management/claude-api-key", "claude-api-key")
+}
+
+// GetCodexKeys fetches Codex API keys.
+func (c *Client) GetCodexKeys() ([]map[string]any, error) {
+	return c.getWrappedKeyList("/v0/management/codex-api-key", "codex-api-key")
+}
+
+// GetVertexKeys fetches Vertex API keys.
+func (c *Client) GetVertexKeys() ([]map[string]any, error) {
+	return c.getWrappedKeyList("/v0/management/vertex-api-key", "vertex-api-key")
+}
+
+// GetOpenAICompat fetches OpenAI compatibility entries.
+func (c *Client) GetOpenAICompat() ([]map[string]any, error) {
+	return c.getWrappedKeyList("/v0/management/openai-compatibility", "openai-compatibility")
+}
+
+// getWrappedKeyList fetches a wrapped list from the API.
+func (c *Client) getWrappedKeyList(path, key string) ([]map[string]any, error) {
+	wrapper, err := c.getJSON(path)
+	if err != nil {
+		return nil, err
+	}
+	return extractList(wrapper, key)
+}
+
+// extractList pulls an array of maps from a wrapper object by key.
+func extractList(wrapper map[string]any, key string) ([]map[string]any, error) {
+	arr, ok := wrapper[key]
+	if !ok || arr == nil {
+		return nil, nil
+	}
+	raw, err := json.Marshal(arr)
+	if err != nil {
+		return nil, err
+	}
+	var result []map[string]any
+	if err := json.Unmarshal(raw, &result); err != nil {
+		return nil, err
+	}
+	return result, nil
+}
+
+// GetDebug fetches the current debug setting.
+func (c *Client) GetDebug() (bool, error) {
+	wrapper, err := c.getJSON("/v0/management/debug")
+	if err != nil {
+		return false, err
+	}
+	if v, ok := wrapper["debug"]; ok {
+		if b, ok := v.(bool); ok {
+			return b, nil
+		}
+	}
+	return false, nil
+}
+
+// GetAuthStatus polls the OAuth session status.
+// Returns status ("wait", "ok", "error") and optional error message.
+func (c *Client) GetAuthStatus(state string) (string, string, error) {
+	query := url.Values{}
+	query.Set("state", state)
+	path := "/v0/management/get-auth-status?" + query.Encode()
+	wrapper, err := c.getJSON(path)
+	if err != nil {
+		return "", "", err
+	}
+	status := getString(wrapper, "status")
+	errMsg := getString(wrapper, "error")
+	return status, errMsg, nil
+}
+
+// ----- Config field update methods -----
+
+// PutBoolField updates a boolean config field.
+func (c *Client) PutBoolField(path string, value bool) error {
+	body, _ := json.Marshal(map[string]any{"value": value})
+	_, err := c.put("/v0/management/"+path, strings.NewReader(string(body)))
+	return err
+}
+
+// PutIntField updates an integer config field.
+func (c *Client) PutIntField(path string, value int) error {
+	body, _ := json.Marshal(map[string]any{"value": value})
+	_, err := c.put("/v0/management/"+path, strings.NewReader(string(body)))
+	return err
+}
+
+// PutStringField updates a string config field.
+func (c *Client) PutStringField(path string, value string) error {
+	body, _ := json.Marshal(map[string]any{"value": value})
+	_, err := c.put("/v0/management/"+path, strings.NewReader(string(body)))
+	return err
+}
+
+// DeleteField sends a DELETE request for a config field.
+func (c *Client) DeleteField(path string) error {
+	_, _, err := c.doRequest("DELETE", "/v0/management/"+path, nil)
+	return err
+}

internal/tui/config_tab.go

@@ -0,0 +1,413 @@
+package tui
+
+import (
+	"fmt"
+	"strconv"
+	"strings"
+
+	"github.com/charmbracelet/bubbles/textinput"
+	"github.com/charmbracelet/bubbles/viewport"
+	tea "github.com/charmbracelet/bubbletea"
+	"github.com/charmbracelet/lipgloss"
+)
+
+// configField represents a single editable config field.
+type configField struct {
+	label    string
+	apiPath  string // management API path (e.g. "debug", "proxy-url")
+	kind     string // "bool", "int", "string", "readonly"
+	value    string // current display value
+	rawValue any    // raw value from API
+}
+
+// configTabModel displays parsed config with interactive editing.
+type configTabModel struct {
+	client    *Client
+	viewport  viewport.Model
+	fields    []configField
+	cursor    int
+	editing   bool
+	textInput textinput.Model
+	err       error
+	message   string // status message (success/error)
+	width     int
+	height    int
+	ready     bool
+}
+
+type configDataMsg struct {
+	config map[string]any
+	err    error
+}
+
+type configUpdateMsg struct {
+	path  string
+	value any
+	err   error
+}
+
+func newConfigTabModel(client *Client) configTabModel {
+	ti := textinput.New()
+	ti.CharLimit = 256
+	return configTabModel{
+		client:    client,
+		textInput: ti,
+	}
+}
+
+func (m configTabModel) Init() tea.Cmd {
+	return m.fetchConfig
+}
+
+func (m configTabModel) fetchConfig() tea.Msg {
+	cfg, err := m.client.GetConfig()
+	return configDataMsg{config: cfg, err: err}
+}
+
+func (m configTabModel) Update(msg tea.Msg) (configTabModel, tea.Cmd) {
+	switch msg := msg.(type) {
+	case localeChangedMsg:
+		m.viewport.SetContent(m.renderContent())
+		return m, nil
+	case configDataMsg:
+		if msg.err != nil {
+			m.err = msg.err
+			m.fields = nil
+		} else {
+			m.err = nil
+			m.fields = m.parseConfig(msg.config)
+		}
+		m.viewport.SetContent(m.renderContent())
+		return m, nil
+
+	case configUpdateMsg:
+		if msg.err != nil {
+			m.message = errorStyle.Render("✗ " + msg.err.Error())
+		} else {
+			m.message = successStyle.Render(T("updated_ok"))
+		}
+		m.viewport.SetContent(m.renderContent())
+		// Refresh config from server
+		return m, m.fetchConfig
+
+	case tea.KeyMsg:
+		if m.editing {
+			return m.handleEditingKey(msg)
+		}
+		return m.handleNormalKey(msg)
+	}
+
+	var cmd tea.Cmd
+	m.viewport, cmd = m.viewport.Update(msg)
+	return m, cmd
+}
+
+func (m configTabModel) handleNormalKey(msg tea.KeyMsg) (configTabModel, tea.Cmd) {
+	switch msg.String() {
+	case "r":
+		m.message = ""
+		return m, m.fetchConfig
+	case "up", "k":
+		if m.cursor > 0 {
+			m.cursor--
+			m.viewport.SetContent(m.renderContent())
+			// Ensure cursor is visible
+			m.ensureCursorVisible()
+		}
+		return m, nil
+	case "down", "j":
+		if m.cursor < len(m.fields)-1 {
+			m.cursor++
+			m.viewport.SetContent(m.renderContent())
+			m.ensureCursorVisible()
+		}
+		return m, nil
+	case "enter", " ":
+		if m.cursor >= 0 && m.cursor < len(m.fields) {
+			f := m.fields[m.cursor]
+			if f.kind == "readonly" {
+				return m, nil
+			}
+			if f.kind == "bool" {
+				// Toggle directly
+				return m, m.toggleBool(m.cursor)
+			}
+			// Start editing for int/string
+			m.editing = true
+			m.textInput.SetValue(configFieldEditValue(f))
+			m.textInput.Focus()
+			m.viewport.SetContent(m.renderContent())
+			return m, textinput.Blink
+		}
+		return m, nil
+	}
+
+	var cmd tea.Cmd
+	m.viewport, cmd = m.viewport.Update(msg)
+	return m, cmd
+}
+
+func (m configTabModel) handleEditingKey(msg tea.KeyMsg) (configTabModel, tea.Cmd) {
+	switch msg.String() {
+	case "enter":
+		m.editing = false
+		m.textInput.Blur()
+		return m, m.submitEdit(m.cursor, m.textInput.Value())
+	case "esc":
+		m.editing = false
+		m.textInput.Blur()
+		m.viewport.SetContent(m.renderContent())
+		return m, nil
+	default:
+		var cmd tea.Cmd
+		m.textInput, cmd = m.textInput.Update(msg)
+		m.viewport.SetContent(m.renderContent())
+		return m, cmd
+	}
+}
+
+func (m configTabModel) toggleBool(idx int) tea.Cmd {
+	return func() tea.Msg {
+		f := m.fields[idx]
+		current := f.value == "true"
+		newValue := !current
+		errPutBool := m.client.PutBoolField(f.apiPath, newValue)
+		return configUpdateMsg{
+			path:  f.apiPath,
+			value: newValue,
+			err:   errPutBool,
+		}
+	}
+}
+
+func (m configTabModel) submitEdit(idx int, newValue string) tea.Cmd {
+	return func() tea.Msg {
+		f := m.fields[idx]
+		var err error
+		var value any
+		switch f.kind {
+		case "int":
+			valueInt, errAtoi := strconv.Atoi(newValue)
+			if errAtoi != nil {
+				return configUpdateMsg{
+					path: f.apiPath,
+					err:  fmt.Errorf("%s: %s", T("invalid_int"), newValue),
+				}
+			}
+			value = valueInt
+			err = m.client.PutIntField(f.apiPath, valueInt)
+		case "string":
+			value = newValue
+			err = m.client.PutStringField(f.apiPath, newValue)
+		}
+		return configUpdateMsg{
+			path:  f.apiPath,
+			value: value,
+			err:   err,
+		}
+	}
+}
+
+func configFieldEditValue(f configField) string {
+	if rawString, ok := f.rawValue.(string); ok {
+		return rawString
+	}
+	return f.value
+}
+
+func (m *configTabModel) SetSize(w, h int) {
+	m.width = w
+	m.height = h
+	if !m.ready {
+		m.viewport = viewport.New(w, h)
+		m.viewport.SetContent(m.renderContent())
+		m.ready = true
+	} else {
+		m.viewport.Width = w
+		m.viewport.Height = h
+	}
+}
+
+func (m *configTabModel) ensureCursorVisible() {
+	// Each field takes ~1 line, header takes ~4 lines
+	targetLine := m.cursor + 5
+	if targetLine < m.viewport.YOffset {
+		m.viewport.SetYOffset(targetLine)
+	}
+	if targetLine >= m.viewport.YOffset+m.viewport.Height {
+		m.viewport.SetYOffset(targetLine - m.viewport.Height + 1)
+	}
+}
+
+func (m configTabModel) View() string {
+	if !m.ready {
+		return T("loading")
+	}
+	return m.viewport.View()
+}
+
+func (m configTabModel) renderContent() string {
+	var sb strings.Builder
+
+	sb.WriteString(titleStyle.Render(T("config_title")))
+	sb.WriteString("\n")
+
+	if m.message != "" {
+		sb.WriteString("  " + m.message)
+		sb.WriteString("\n")
+	}
+
+	sb.WriteString(helpStyle.Render(T("config_help1")))
+	sb.WriteString("\n")
+	sb.WriteString(helpStyle.Render(T("config_help2")))
+	sb.WriteString("\n\n")
+
+	if m.err != nil {
+		sb.WriteString(errorStyle.Render("  ⚠ Error: " + m.err.Error()))
+		return sb.String()
+	}
+
+	if len(m.fields) == 0 {
+		sb.WriteString(subtitleStyle.Render(T("no_config")))
+		return sb.String()
+	}
+
+	currentSection := ""
+	for i, f := range m.fields {
+		// Section headers
+		section := fieldSection(f.apiPath)
+		if section != currentSection {
+			currentSection = section
+			sb.WriteString("\n")
+			sb.WriteString(lipgloss.NewStyle().Bold(true).Foreground(colorHighlight).Render("  ── " + section + " "))
+			sb.WriteString("\n")
+		}
+
+		isSelected := i == m.cursor
+		prefix := "  "
+		if isSelected {
+			prefix = "▸ "
+		}
+
+		labelStr := lipgloss.NewStyle().
+			Foreground(colorInfo).
+			Bold(isSelected).
+			Width(32).
+			Render(f.label)
+
+		var valueStr string
+		if m.editing && isSelected {
+			valueStr = m.textInput.View()
+		} else {
+			switch f.kind {
+			case "bool":
+				if f.value == "true" {
+					valueStr = successStyle.Render("● ON")
+				} else {
+					valueStr = lipgloss.NewStyle().Foreground(colorMuted).Render("○ OFF")
+				}
+			case "readonly":
+				valueStr = lipgloss.NewStyle().Foreground(colorSubtext).Render(f.value)
+			default:
+				valueStr = valueStyle.Render(f.value)
+			}
+		}
+
+		line := prefix + labelStr + "  " + valueStr
+		if isSelected && !m.editing {
+			line = lipgloss.NewStyle().Background(colorSurface).Render(line)
+		}
+		sb.WriteString(line + "\n")
+	}
+
+	return sb.String()
+}
+
+func (m configTabModel) parseConfig(cfg map[string]any) []configField {
+	var fields []configField
+
+	// Server settings
+	fields = append(fields, configField{"Port", "port", "readonly", fmt.Sprintf("%.0f", getFloat(cfg, "port")), nil})
+	fields = append(fields, configField{"Host", "host", "readonly", getString(cfg, "host"), nil})
+	fields = append(fields, configField{"Debug", "debug", "bool", fmt.Sprintf("%v", getBool(cfg, "debug")), nil})
+	fields = append(fields, configField{"Proxy URL", "proxy-url", "string", getString(cfg, "proxy-url"), nil})
+	fields = append(fields, configField{"Request Retry", "request-retry", "int", fmt.Sprintf("%.0f", getFloat(cfg, "request-retry")), nil})
+	fields = append(fields, configField{"Max Retry Interval (s)", "max-retry-interval", "int", fmt.Sprintf("%.0f", getFloat(cfg, "max-retry-interval")), nil})
+	fields = append(fields, configField{"Force Model Prefix", "force-model-prefix", "string", getString(cfg, "force-model-prefix"), nil})
+
+	// Logging
+	fields = append(fields, configField{"Logging to File", "logging-to-file", "bool", fmt.Sprintf("%v", getBool(cfg, "logging-to-file")), nil})
+	fields = append(fields, configField{"Logs Max Total Size (MB)", "logs-max-total-size-mb", "int", fmt.Sprintf("%.0f", getFloat(cfg, "logs-max-total-size-mb")), nil})
+	fields = append(fields, configField{"Error Logs Max Files", "error-logs-max-files", "int", fmt.Sprintf("%.0f", getFloat(cfg, "error-logs-max-files")), nil})
+	fields = append(fields, configField{"Usage Stats Enabled", "usage-statistics-enabled", "bool", fmt.Sprintf("%v", getBool(cfg, "usage-statistics-enabled")), nil})
+	fields = append(fields, configField{"Request Log", "request-log", "bool", fmt.Sprintf("%v", getBool(cfg, "request-log")), nil})
+
+	// Quota exceeded
+	fields = append(fields, configField{"Switch Project on Quota", "quota-exceeded/switch-project", "bool", fmt.Sprintf("%v", getBoolNested(cfg, "quota-exceeded", "switch-project")), nil})
+	fields = append(fields, configField{"Switch Preview Model", "quota-exceeded/switch-preview-model", "bool", fmt.Sprintf("%v", getBoolNested(cfg, "quota-exceeded", "switch-preview-model")), nil})
+
+	// Routing
+	if routing, ok := cfg["routing"].(map[string]any); ok {
+		fields = append(fields, configField{"Routing Strategy", "routing/strategy", "string", getString(routing, "strategy"), nil})
+	} else {
+		fields = append(fields, configField{"Routing Strategy", "routing/strategy", "string", "", nil})
+	}
+
+	// WebSocket auth
+	fields = append(fields, configField{"WebSocket Auth", "ws-auth", "bool", fmt.Sprintf("%v", getBool(cfg, "ws-auth")), nil})
+
+	// AMP settings
+	if amp, ok := cfg["ampcode"].(map[string]any); ok {
+		upstreamURL := getString(amp, "upstream-url")
+		upstreamAPIKey := getString(amp, "upstream-api-key")
+		fields = append(fields, configField{"AMP Upstream URL", "ampcode/upstream-url", "string", upstreamURL, upstreamURL})
+		fields = append(fields, configField{"AMP Upstream API Key", "ampcode/upstream-api-key", "string", maskIfNotEmpty(upstreamAPIKey), upstreamAPIKey})
+		fields = append(fields, configField{"AMP Restrict Mgmt Localhost", "ampcode/restrict-management-to-localhost", "bool", fmt.Sprintf("%v", getBool(amp, "restrict-management-to-localhost")), nil})
+	}
+
+	return fields
+}
+
+func fieldSection(apiPath string) string {
+	if strings.HasPrefix(apiPath, "ampcode/") {
+		return T("section_ampcode")
+	}
+	if strings.HasPrefix(apiPath, "quota-exceeded/") {
+		return T("section_quota")
+	}
+	if strings.HasPrefix(apiPath, "routing/") {
+		return T("section_routing")
+	}
+	switch apiPath {
+	case "port", "host", "debug", "proxy-url", "request-retry", "max-retry-interval", "force-model-prefix":
+		return T("section_server")
+	case "logging-to-file", "logs-max-total-size-mb", "error-logs-max-files", "usage-statistics-enabled", "request-log":
+		return T("section_logging")
+	case "ws-auth":
+		return T("section_websocket")
+	default:
+		return T("section_other")
+	}
+}
+
+func getBoolNested(m map[string]any, keys ...string) bool {
+	current := m
+	for i, key := range keys {
+		if i == len(keys)-1 {
+			return getBool(current, key)
+		}
+		if nested, ok := current[key].(map[string]any); ok {
+			current = nested
+		} else {
+			return false
+		}
+	}
+	return false
+}
+
+func maskIfNotEmpty(s string) string {
+	if s == "" {
+		return T("not_set")
+	}
+	return maskKey(s)
+}

internal/tui/dashboard.go

@@ -0,0 +1,360 @@
+package tui
+
+import (
+	"encoding/json"
+	"fmt"
+	"strings"
+
+	"github.com/charmbracelet/bubbles/viewport"
+	tea "github.com/charmbracelet/bubbletea"
+	"github.com/charmbracelet/lipgloss"
+)
+
+// dashboardModel displays server info, stats cards, and config overview.
+type dashboardModel struct {
+	client   *Client
+	viewport viewport.Model
+	content  string
+	err      error
+	width    int
+	height   int
+	ready    bool
+
+	// Cached data for re-rendering on locale change
+	lastConfig    map[string]any
+	lastUsage     map[string]any
+	lastAuthFiles []map[string]any
+	lastAPIKeys   []string
+}
+
+type dashboardDataMsg struct {
+	config    map[string]any
+	usage     map[string]any
+	authFiles []map[string]any
+	apiKeys   []string
+	err       error
+}
+
+func newDashboardModel(client *Client) dashboardModel {
+	return dashboardModel{
+		client: client,
+	}
+}
+
+func (m dashboardModel) Init() tea.Cmd {
+	return m.fetchData
+}
+
+func (m dashboardModel) fetchData() tea.Msg {
+	cfg, cfgErr := m.client.GetConfig()
+	usage, usageErr := m.client.GetUsage()
+	authFiles, authErr := m.client.GetAuthFiles()
+	apiKeys, keysErr := m.client.GetAPIKeys()
+
+	var err error
+	for _, e := range []error{cfgErr, usageErr, authErr, keysErr} {
+		if e != nil {
+			err = e
+			break
+		}
+	}
+	return dashboardDataMsg{config: cfg, usage: usage, authFiles: authFiles, apiKeys: apiKeys, err: err}
+}
+
+func (m dashboardModel) Update(msg tea.Msg) (dashboardModel, tea.Cmd) {
+	switch msg := msg.(type) {
+	case localeChangedMsg:
+		// Re-render immediately with cached data using new locale
+		m.content = m.renderDashboard(m.lastConfig, m.lastUsage, m.lastAuthFiles, m.lastAPIKeys)
+		m.viewport.SetContent(m.content)
+		// Also fetch fresh data in background
+		return m, m.fetchData
+
+	case dashboardDataMsg:
+		if msg.err != nil {
+			m.err = msg.err
+			m.content = errorStyle.Render("⚠ Error: " + msg.err.Error())
+		} else {
+			m.err = nil
+			// Cache data for locale switching
+			m.lastConfig = msg.config
+			m.lastUsage = msg.usage
+			m.lastAuthFiles = msg.authFiles
+			m.lastAPIKeys = msg.apiKeys
+
+			m.content = m.renderDashboard(msg.config, msg.usage, msg.authFiles, msg.apiKeys)
+		}
+		m.viewport.SetContent(m.content)
+		return m, nil
+
+	case tea.KeyMsg:
+		if msg.String() == "r" {
+			return m, m.fetchData
+		}
+		var cmd tea.Cmd
+		m.viewport, cmd = m.viewport.Update(msg)
+		return m, cmd
+	}
+
+	var cmd tea.Cmd
+	m.viewport, cmd = m.viewport.Update(msg)
+	return m, cmd
+}
+
+func (m *dashboardModel) SetSize(w, h int) {
+	m.width = w
+	m.height = h
+	if !m.ready {
+		m.viewport = viewport.New(w, h)
+		m.viewport.SetContent(m.content)
+		m.ready = true
+	} else {
+		m.viewport.Width = w
+		m.viewport.Height = h
+	}
+}
+
+func (m dashboardModel) View() string {
+	if !m.ready {
+		return T("loading")
+	}
+	return m.viewport.View()
+}
+
+func (m dashboardModel) renderDashboard(cfg, usage map[string]any, authFiles []map[string]any, apiKeys []string) string {
+	var sb strings.Builder
+
+	sb.WriteString(titleStyle.Render(T("dashboard_title")))
+	sb.WriteString("\n")
+	sb.WriteString(helpStyle.Render(T("dashboard_help")))
+	sb.WriteString("\n\n")
+
+	// ━━━ Connection Status ━━━
+	connStyle := lipgloss.NewStyle().Bold(true).Foreground(colorSuccess)
+	sb.WriteString(connStyle.Render(T("connected")))
+	sb.WriteString(fmt.Sprintf("  %s", m.client.baseURL))
+	sb.WriteString("\n\n")
+
+	// ━━━ Stats Cards ━━━
+	cardWidth := 25
+	if m.width > 0 {
+		cardWidth = (m.width - 6) / 4
+		if cardWidth < 18 {
+			cardWidth = 18
+		}
+	}
+
+	cardStyle := lipgloss.NewStyle().
+		Border(lipgloss.RoundedBorder()).
+		BorderForeground(lipgloss.Color("240")).
+		Padding(0, 1).
+		Width(cardWidth).
+		Height(2)
+
+	// Card 1: API Keys
+	keyCount := len(apiKeys)
+	card1 := cardStyle.Render(fmt.Sprintf(
+		"%s\n%s",
+		lipgloss.NewStyle().Bold(true).Foreground(lipgloss.Color("111")).Render(fmt.Sprintf("🔑 %d", keyCount)),
+		lipgloss.NewStyle().Foreground(colorMuted).Render(T("mgmt_keys")),
+	))
+
+	// Card 2: Auth Files
+	authCount := len(authFiles)
+	activeAuth := 0
+	for _, f := range authFiles {
+		if !getBool(f, "disabled") {
+			activeAuth++
+		}
+	}
+	card2 := cardStyle.Render(fmt.Sprintf(
+		"%s\n%s",
+		lipgloss.NewStyle().Bold(true).Foreground(lipgloss.Color("76")).Render(fmt.Sprintf("📄 %d", authCount)),
+		lipgloss.NewStyle().Foreground(colorMuted).Render(fmt.Sprintf("%s (%d %s)", T("auth_files_label"), activeAuth, T("active_suffix"))),
+	))
+
+	// Card 3: Total Requests
+	totalReqs := int64(0)
+	successReqs := int64(0)
+	failedReqs := int64(0)
+	totalTokens := int64(0)
+	if usage != nil {
+		if usageMap, ok := usage["usage"].(map[string]any); ok {
+			totalReqs = int64(getFloat(usageMap, "total_requests"))
+			successReqs = int64(getFloat(usageMap, "success_count"))
+			failedReqs = int64(getFloat(usageMap, "failure_count"))
+			totalTokens = int64(getFloat(usageMap, "total_tokens"))
+		}
+	}
+	card3 := cardStyle.Render(fmt.Sprintf(
+		"%s\n%s",
+		lipgloss.NewStyle().Bold(true).Foreground(lipgloss.Color("214")).Render(fmt.Sprintf("📈 %d", totalReqs)),
+		lipgloss.NewStyle().Foreground(colorMuted).Render(fmt.Sprintf("%s (✓%d ✗%d)", T("total_requests"), successReqs, failedReqs)),
+	))
+
+	// Card 4: Total Tokens
+	tokenStr := formatLargeNumber(totalTokens)
+	card4 := cardStyle.Render(fmt.Sprintf(
+		"%s\n%s",
+		lipgloss.NewStyle().Bold(true).Foreground(lipgloss.Color("170")).Render(fmt.Sprintf("🔤 %s", tokenStr)),
+		lipgloss.NewStyle().Foreground(colorMuted).Render(T("total_tokens")),
+	))
+
+	sb.WriteString(lipgloss.JoinHorizontal(lipgloss.Top, card1, " ", card2, " ", card3, " ", card4))
+	sb.WriteString("\n\n")
+
+	// ━━━ Current Config ━━━
+	sb.WriteString(lipgloss.NewStyle().Bold(true).Foreground(colorHighlight).Render(T("current_config")))
+	sb.WriteString("\n")
+	sb.WriteString(strings.Repeat("─", minInt(m.width, 60)))
+	sb.WriteString("\n")
+
+	if cfg != nil {
+		debug := getBool(cfg, "debug")
+		retry := getFloat(cfg, "request-retry")
+		proxyURL := getString(cfg, "proxy-url")
+		loggingToFile := getBool(cfg, "logging-to-file")
+		usageEnabled := true
+		if v, ok := cfg["usage-statistics-enabled"]; ok {
+			if b, ok2 := v.(bool); ok2 {
+				usageEnabled = b
+			}
+		}
+
+		configItems := []struct {
+			label string
+			value string
+		}{
+			{T("debug_mode"), boolEmoji(debug)},
+			{T("usage_stats"), boolEmoji(usageEnabled)},
+			{T("log_to_file"), boolEmoji(loggingToFile)},
+			{T("retry_count"), fmt.Sprintf("%.0f", retry)},
+		}
+		if proxyURL != "" {
+			configItems = append(configItems, struct {
+				label string
+				value string
+			}{T("proxy_url"), proxyURL})
+		}
+
+		// Render config items as a compact row
+		for _, item := range configItems {
+			sb.WriteString(fmt.Sprintf("  %s %s\n",
+				labelStyle.Render(item.label+":"),
+				valueStyle.Render(item.value)))
+		}
+
+		// Routing strategy
+		strategy := "round-robin"
+		if routing, ok := cfg["routing"].(map[string]any); ok {
+			if s := getString(routing, "strategy"); s != "" {
+				strategy = s
+			}
+		}
+		sb.WriteString(fmt.Sprintf("  %s %s\n",
+			labelStyle.Render(T("routing_strategy")+":"),
+			valueStyle.Render(strategy)))
+	}
+
+	sb.WriteString("\n")
+
+	// ━━━ Per-Model Usage ━━━
+	if usage != nil {
+		if usageMap, ok := usage["usage"].(map[string]any); ok {
+			if apis, ok := usageMap["apis"].(map[string]any); ok && len(apis) > 0 {
+				sb.WriteString(lipgloss.NewStyle().Bold(true).Foreground(colorHighlight).Render(T("model_stats")))
+				sb.WriteString("\n")
+				sb.WriteString(strings.Repeat("─", minInt(m.width, 60)))
+				sb.WriteString("\n")
+
+				header := fmt.Sprintf("  %-40s %10s %12s", T("model"), T("requests"), T("tokens"))
+				sb.WriteString(tableHeaderStyle.Render(header))
+				sb.WriteString("\n")
+
+				for _, apiSnap := range apis {
+					if apiMap, ok := apiSnap.(map[string]any); ok {
+						if models, ok := apiMap["models"].(map[string]any); ok {
+							for model, v := range models {
+								if stats, ok := v.(map[string]any); ok {
+									reqs := int64(getFloat(stats, "total_requests"))
+									toks := int64(getFloat(stats, "total_tokens"))
+									row := fmt.Sprintf("  %-40s %10d %12s", truncate(model, 40), reqs, formatLargeNumber(toks))
+									sb.WriteString(tableCellStyle.Render(row))
+									sb.WriteString("\n")
+								}
+							}
+						}
+					}
+				}
+			}
+		}
+	}
+
+	return sb.String()
+}
+
+func formatKV(key, value string) string {
+	return fmt.Sprintf("  %s %s\n", labelStyle.Render(key+":"), valueStyle.Render(value))
+}
+
+func getString(m map[string]any, key string) string {
+	if v, ok := m[key]; ok {
+		if s, ok := v.(string); ok {
+			return s
+		}
+	}
+	return ""
+}
+
+func getFloat(m map[string]any, key string) float64 {
+	if v, ok := m[key]; ok {
+		switch n := v.(type) {
+		case float64:
+			return n
+		case json.Number:
+			f, _ := n.Float64()
+			return f
+		}
+	}
+	return 0
+}
+
+func getBool(m map[string]any, key string) bool {
+	if v, ok := m[key]; ok {
+		if b, ok := v.(bool); ok {
+			return b
+		}
+	}
+	return false
+}
+
+func boolEmoji(b bool) string {
+	if b {
+		return T("bool_yes")
+	}
+	return T("bool_no")
+}
+
+func formatLargeNumber(n int64) string {
+	if n >= 1_000_000 {
+		return fmt.Sprintf("%.1fM", float64(n)/1_000_000)
+	}
+	if n >= 1_000 {
+		return fmt.Sprintf("%.1fK", float64(n)/1_000)
+	}
+	return fmt.Sprintf("%d", n)
+}
+
+func truncate(s string, maxLen int) string {
+	if len(s) > maxLen {
+		return s[:maxLen-3] + "..."
+	}
+	return s
+}
+
+func minInt(a, b int) int {
+	if a < b {
+		return a
+	}
+	return b
+}