fix(translator): include token usage in message_delta for Claude responses

fix: skip empty text parts and messages to avoid Gemini API error

.github/workflows/docker-image.yml

@@ -10,13 +10,11 @@ env:
   DOCKERHUB_REPO: eceasy/cli-proxy-api
 
 jobs:
-  docker:
+  docker_amd64:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
         uses: actions/checkout@v4
-      - name: Set up QEMU
-        uses: docker/setup-qemu-action@v3
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
       - name: Login to DockerHub
@@ -29,18 +27,113 @@ jobs:
           echo VERSION=`git describe --tags --always --dirty` >> $GITHUB_ENV
           echo COMMIT=`git rev-parse --short HEAD` >> $GITHUB_ENV
           echo BUILD_DATE=`date -u +%Y-%m-%dT%H:%M:%SZ` >> $GITHUB_ENV
-      - name: Build and push
+      - name: Build and push (amd64)
         uses: docker/build-push-action@v6
         with:
           context: .
-          platforms: |
-            linux/amd64
-            linux/arm64
+          platforms: linux/amd64
           push: true
           build-args: |
             VERSION=${{ env.VERSION }}
             COMMIT=${{ env.COMMIT }}
             BUILD_DATE=${{ env.BUILD_DATE }}
           tags: |
-            ${{ env.DOCKERHUB_REPO }}:latest
-            ${{ env.DOCKERHUB_REPO }}:${{ env.VERSION }}
+            ${{ env.DOCKERHUB_REPO }}:latest-amd64
+            ${{ env.DOCKERHUB_REPO }}:${{ env.VERSION }}-amd64
+
+  docker_arm64:
+    runs-on: ubuntu-24.04-arm
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+      - name: Login to DockerHub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+      - name: Generate Build Metadata
+        run: |
+          echo VERSION=`git describe --tags --always --dirty` >> $GITHUB_ENV
+          echo COMMIT=`git rev-parse --short HEAD` >> $GITHUB_ENV
+          echo BUILD_DATE=`date -u +%Y-%m-%dT%H:%M:%SZ` >> $GITHUB_ENV
+      - name: Build and push (arm64)
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          platforms: linux/arm64
+          push: true
+          build-args: |
+            VERSION=${{ env.VERSION }}
+            COMMIT=${{ env.COMMIT }}
+            BUILD_DATE=${{ env.BUILD_DATE }}
+          tags: |
+            ${{ env.DOCKERHUB_REPO }}:latest-arm64
+            ${{ env.DOCKERHUB_REPO }}:${{ env.VERSION }}-arm64
+
+  docker_manifest:
+    runs-on: ubuntu-latest
+    needs:
+      - docker_amd64
+      - docker_arm64
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+      - name: Login to DockerHub
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+      - name: Generate Build Metadata
+        run: |
+          echo VERSION=`git describe --tags --always --dirty` >> $GITHUB_ENV
+          echo COMMIT=`git rev-parse --short HEAD` >> $GITHUB_ENV
+          echo BUILD_DATE=`date -u +%Y-%m-%dT%H:%M:%SZ` >> $GITHUB_ENV
+      - name: Create and push multi-arch manifests
+        run: |
+          docker buildx imagetools create \
+            --tag "${DOCKERHUB_REPO}:latest" \
+            "${DOCKERHUB_REPO}:latest-amd64" \
+            "${DOCKERHUB_REPO}:latest-arm64"
+          docker buildx imagetools create \
+            --tag "${DOCKERHUB_REPO}:${VERSION}" \
+            "${DOCKERHUB_REPO}:${VERSION}-amd64" \
+            "${DOCKERHUB_REPO}:${VERSION}-arm64"
+      - name: Cleanup temporary tags
+        continue-on-error: true
+        env:
+          DOCKERHUB_USERNAME: ${{ secrets.DOCKERHUB_USERNAME }}
+          DOCKERHUB_TOKEN: ${{ secrets.DOCKERHUB_TOKEN }}
+        run: |
+          set -euo pipefail
+          namespace="${DOCKERHUB_REPO%%/*}"
+          repo_name="${DOCKERHUB_REPO#*/}"
+
+          token="$(
+            curl -fsSL \
+              -H 'Content-Type: application/json' \
+              -d "{\"username\":\"${DOCKERHUB_USERNAME}\",\"password\":\"${DOCKERHUB_TOKEN}\"}" \
+              'https://hub.docker.com/v2/users/login/' \
+              | python3 -c 'import json,sys; print(json.load(sys.stdin)["token"])'
+          )"
+
+          delete_tag() {
+            local tag="$1"
+            local url="https://hub.docker.com/v2/repositories/${namespace}/${repo_name}/tags/${tag}/"
+            local http_code
+            http_code="$(curl -sS -o /dev/null -w "%{http_code}" -X DELETE -H "Authorization: JWT ${token}" "${url}" || true)"
+            if [ "${http_code}" = "204" ] || [ "${http_code}" = "404" ]; then
+              echo "Docker Hub tag removed (or missing): ${DOCKERHUB_REPO}:${tag} (HTTP ${http_code})"
+              return 0
+            fi
+            echo "Docker Hub tag delete failed: ${DOCKERHUB_REPO}:${tag} (HTTP ${http_code})"
+            return 0
+          }
+
+          delete_tag "latest-amd64"
+          delete_tag "latest-arm64"
+          delete_tag "${VERSION}-amd64"
+          delete_tag "${VERSION}-arm64"

README.md

@@ -134,6 +134,10 @@ VSCode extension for quick switching between Claude Code models, featuring integ
 
 Windows desktop app built with Tauri + React for monitoring AI coding assistant quotas via CLIProxyAPI. Track usage across Gemini, Claude, OpenAI Codex, and Antigravity accounts with real-time dashboard, system tray integration, and one-click proxy control - no API keys needed.
 
+### [CPA-XXX Panel](https://github.com/ferretgeek/CPA-X)
+
+A lightweight web admin panel for CLIProxyAPI with health checks, resource monitoring, real-time logs, auto-update, request statistics and pricing display. Supports one-click installation and systemd service.
+
 > [!NOTE]  
 > If you developed a project based on CLIProxyAPI, please open a PR to add it to this list.
 

README_CN.md

@@ -133,6 +133,10 @@ CLI 封装器，用于通过 CLIProxyAPI OAuth 即时切换多个 Claude 账户
 
 Windows 桌面应用，基于 Tauri + React 构建，用于通过 CLIProxyAPI 监控 AI 编程助手配额。支持跨 Gemini、Claude、OpenAI Codex 和 Antigravity 账户的使用量追踪，提供实时仪表盘、系统托盘集成和一键代理控制，无需 API 密钥。
 
+### [CPA-XXX Panel](https://github.com/ferretgeek/CPA-X)
+
+面向 CLIProxyAPI 的 Web 管理面板，提供健康检查、资源监控、日志查看、自动更新、请求统计与定价展示，支持一键安装与 systemd 服务。
+
 > [!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。
 

go.mod

@@ -13,6 +13,7 @@ require (
 	github.com/joho/godotenv v1.5.1
 	github.com/klauspost/compress v1.17.4
 	github.com/minio/minio-go/v7 v7.0.66
+	github.com/refraction-networking/utls v1.8.2
 	github.com/sirupsen/logrus v1.9.3
 	github.com/skratchdot/open-golang v0.0.0-20200116055534-eef842397966
 	github.com/tidwall/gjson v1.18.0

go.sum

@@ -118,6 +118,8 @@ github.com/pjbgf/sha1cd v0.5.0 h1:a+UkboSi1znleCDUNT3M5YxjOnN1fz2FhN48FlwCxs0=
 github.com/pjbgf/sha1cd v0.5.0/go.mod h1:lhpGlyHLpQZoxMv8HcgXvZEhcGs0PG/vsZnEJ7H0iCM=
 github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
 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/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=

internal/api/handlers/management/auth_files.go

@@ -11,7 +11,6 @@ import (
 	"io"
 	"net"
 	"net/http"
-	"net/url"
 	"os"
 	"path/filepath"
 	"sort"
@@ -21,6 +20,7 @@ import (
 	"time"
 
 	"github.com/gin-gonic/gin"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/auth/antigravity"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/auth/claude"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/auth/codex"
 	geminiAuth "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/gemini"
@@ -232,14 +232,6 @@ func stopForwarderInstance(port int, forwarder *callbackForwarder) {
 	log.Infof("callback forwarder on port %d stopped", port)
 }
 
-func sanitizeAntigravityFileName(email string) string {
-	if strings.TrimSpace(email) == "" {
-		return "antigravity.json"
-	}
-	replacer := strings.NewReplacer("@", "_", ".", "_")
-	return fmt.Sprintf("antigravity-%s.json", replacer.Replace(email))
-}
-
 func (h *Handler) managementCallbackURL(path string) (string, error) {
 	if h == nil || h.cfg == nil || h.cfg.Port <= 0 {
 		return "", fmt.Errorf("server port is not configured")
@@ -749,6 +741,72 @@ func (h *Handler) registerAuthFromFile(ctx context.Context, path string, data []
 	return err
 }
 
+// PatchAuthFileStatus toggles the disabled state of an auth file
+func (h *Handler) PatchAuthFileStatus(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"`
+		Disabled *bool  `json:"disabled"`
+	}
+	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
+	}
+	if req.Disabled == nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "disabled 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
+	}
+
+	// Update disabled state
+	targetAuth.Disabled = *req.Disabled
+	if *req.Disabled {
+		targetAuth.Status = coreauth.StatusDisabled
+		targetAuth.StatusMessage = "disabled via management API"
+	} else {
+		targetAuth.Status = coreauth.StatusActive
+		targetAuth.StatusMessage = ""
+	}
+	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", "disabled": *req.Disabled})
+}
+
 func (h *Handler) disableAuth(ctx context.Context, id string) {
 	if h == nil || h.authManager == nil {
 		return
@@ -915,67 +973,14 @@ func (h *Handler) RequestAnthropicToken(c *gin.Context) {
 		rawCode := resultMap["code"]
 		code := strings.Split(rawCode, "#")[0]
 
-		// Exchange code for tokens (replicate logic using updated redirect_uri)
-		// Extract client_id from the modified auth URL
-		clientID := ""
-		if u2, errP := url.Parse(authURL); errP == nil {
-			clientID = u2.Query().Get("client_id")
-		}
-		// Build request
-		bodyMap := map[string]any{
-			"code":          code,
-			"state":         state,
-			"grant_type":    "authorization_code",
-			"client_id":     clientID,
-			"redirect_uri":  "http://localhost:54545/callback",
-			"code_verifier": pkceCodes.CodeVerifier,
-		}
-		bodyJSON, _ := json.Marshal(bodyMap)
-
-		httpClient := util.SetProxy(&h.cfg.SDKConfig, &http.Client{})
-		req, _ := http.NewRequestWithContext(ctx, "POST", "https://console.anthropic.com/v1/oauth/token", strings.NewReader(string(bodyJSON)))
-		req.Header.Set("Content-Type", "application/json")
-		req.Header.Set("Accept", "application/json")
-		resp, errDo := httpClient.Do(req)
-		if errDo != nil {
-			authErr := claude.NewAuthenticationError(claude.ErrCodeExchangeFailed, errDo)
+		// Exchange code for tokens using internal auth service
+		bundle, errExchange := anthropicAuth.ExchangeCodeForTokens(ctx, code, state, pkceCodes)
+		if errExchange != nil {
+			authErr := claude.NewAuthenticationError(claude.ErrCodeExchangeFailed, errExchange)
 			log.Errorf("Failed to exchange authorization code for tokens: %v", authErr)
 			SetOAuthSessionError(state, "Failed to exchange authorization code for tokens")
 			return
 		}
-		defer func() {
-			if errClose := resp.Body.Close(); errClose != nil {
-				log.Errorf("failed to close response body: %v", errClose)
-			}
-		}()
-		respBody, _ := io.ReadAll(resp.Body)
-		if resp.StatusCode != http.StatusOK {
-			log.Errorf("token exchange failed with status %d: %s", resp.StatusCode, string(respBody))
-			SetOAuthSessionError(state, fmt.Sprintf("token exchange failed with status %d", resp.StatusCode))
-			return
-		}
-		var tResp struct {
-			AccessToken  string `json:"access_token"`
-			RefreshToken string `json:"refresh_token"`
-			ExpiresIn    int    `json:"expires_in"`
-			Account      struct {
-				EmailAddress string `json:"email_address"`
-			} `json:"account"`
-		}
-		if errU := json.Unmarshal(respBody, &tResp); errU != nil {
-			log.Errorf("failed to parse token response: %v", errU)
-			SetOAuthSessionError(state, "Failed to parse token response")
-			return
-		}
-		bundle := &claude.ClaudeAuthBundle{
-			TokenData: claude.ClaudeTokenData{
-				AccessToken:  tResp.AccessToken,
-				RefreshToken: tResp.RefreshToken,
-				Email:        tResp.Account.EmailAddress,
-				Expire:       time.Now().Add(time.Duration(tResp.ExpiresIn) * time.Second).Format(time.RFC3339),
-			},
-			LastRefresh: time.Now().Format(time.RFC3339),
-		}
 
 		// Create token storage
 		tokenStorage := anthropicAuth.CreateTokenStorage(bundle)
@@ -1015,17 +1020,13 @@ func (h *Handler) RequestGeminiCLIToken(c *gin.Context) {
 
 	fmt.Println("Initializing Google authentication...")
 
-	// OAuth2 configuration (mirrors internal/auth/gemini)
+	// OAuth2 configuration using exported constants from internal/auth/gemini
 	conf := &oauth2.Config{
-		ClientID:     "681255809395-oo8ft2oprdrnp9e3aqf6av3hmdib135j.apps.googleusercontent.com",
-		ClientSecret: "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",
-		RedirectURL:  "http://localhost:8085/oauth2callback",
-		Scopes: []string{
-			"https://www.googleapis.com/auth/cloud-platform",
-			"https://www.googleapis.com/auth/userinfo.email",
-			"https://www.googleapis.com/auth/userinfo.profile",
-		},
-		Endpoint: google.Endpoint,
+		ClientID:     geminiAuth.ClientID,
+		ClientSecret: geminiAuth.ClientSecret,
+		RedirectURL:  fmt.Sprintf("http://localhost:%d/oauth2callback", geminiAuth.DefaultCallbackPort),
+		Scopes:       geminiAuth.Scopes,
+		Endpoint:     google.Endpoint,
 	}
 
 	// Build authorization URL and return it immediately
@@ -1147,13 +1148,9 @@ func (h *Handler) RequestGeminiCLIToken(c *gin.Context) {
 		}
 
 		ifToken["token_uri"] = "https://oauth2.googleapis.com/token"
-		ifToken["client_id"] = "681255809395-oo8ft2oprdrnp9e3aqf6av3hmdib135j.apps.googleusercontent.com"
-		ifToken["client_secret"] = "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"
-		ifToken["scopes"] = []string{
-			"https://www.googleapis.com/auth/cloud-platform",
-			"https://www.googleapis.com/auth/userinfo.email",
-			"https://www.googleapis.com/auth/userinfo.profile",
-		}
+		ifToken["client_id"] = geminiAuth.ClientID
+		ifToken["client_secret"] = geminiAuth.ClientSecret
+		ifToken["scopes"] = geminiAuth.Scopes
 		ifToken["universe_domain"] = "googleapis.com"
 
 		ts := geminiAuth.GeminiTokenStorage{
@@ -1340,73 +1337,25 @@ func (h *Handler) RequestCodexToken(c *gin.Context) {
 		}
 
 		log.Debug("Authorization code received, exchanging for tokens...")
-		// Extract client_id from authURL
-		clientID := ""
-		if u2, errP := url.Parse(authURL); errP == nil {
-			clientID = u2.Query().Get("client_id")
-		}
-		// Exchange code for tokens with redirect equal to mgmtRedirect
-		form := url.Values{
-			"grant_type":    {"authorization_code"},
-			"client_id":     {clientID},
-			"code":          {code},
-			"redirect_uri":  {"http://localhost:1455/auth/callback"},
-			"code_verifier": {pkceCodes.CodeVerifier},
-		}
-		httpClient := util.SetProxy(&h.cfg.SDKConfig, &http.Client{})
-		req, _ := http.NewRequestWithContext(ctx, "POST", "https://auth.openai.com/oauth/token", strings.NewReader(form.Encode()))
-		req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
-		req.Header.Set("Accept", "application/json")
-		resp, errDo := httpClient.Do(req)
-		if errDo != nil {
-			authErr := codex.NewAuthenticationError(codex.ErrCodeExchangeFailed, errDo)
+		// Exchange code for tokens using internal auth service
+		bundle, errExchange := openaiAuth.ExchangeCodeForTokens(ctx, code, pkceCodes)
+		if errExchange != nil {
+			authErr := codex.NewAuthenticationError(codex.ErrCodeExchangeFailed, errExchange)
 			SetOAuthSessionError(state, "Failed to exchange authorization code for tokens")
 			log.Errorf("Failed to exchange authorization code for tokens: %v", authErr)
 			return
 		}
-		defer func() { _ = resp.Body.Close() }()
-		respBody, _ := io.ReadAll(resp.Body)
-		if resp.StatusCode != http.StatusOK {
-			SetOAuthSessionError(state, fmt.Sprintf("Token exchange failed with status %d", resp.StatusCode))
-			log.Errorf("token exchange failed with status %d: %s", resp.StatusCode, string(respBody))
-			return
-		}
-		var tokenResp struct {
-			AccessToken  string `json:"access_token"`
-			RefreshToken string `json:"refresh_token"`
-			IDToken      string `json:"id_token"`
-			ExpiresIn    int    `json:"expires_in"`
-		}
-		if errU := json.Unmarshal(respBody, &tokenResp); errU != nil {
-			SetOAuthSessionError(state, "Failed to parse token response")
-			log.Errorf("failed to parse token response: %v", errU)
-			return
-		}
-		claims, _ := codex.ParseJWTToken(tokenResp.IDToken)
-		email := ""
-		accountID := ""
+
+		// Extract additional info for filename generation
+		claims, _ := codex.ParseJWTToken(bundle.TokenData.IDToken)
 		planType := ""
-		if claims != nil {
-			email = claims.GetUserEmail()
-			accountID = claims.GetAccountID()
-			planType = strings.TrimSpace(claims.CodexAuthInfo.ChatgptPlanType)
-		}
 		hashAccountID := ""
-		if accountID != "" {
-			digest := sha256.Sum256([]byte(accountID))
-			hashAccountID = hex.EncodeToString(digest[:])[:8]
-		}
-		// Build bundle compatible with existing storage
-		bundle := &codex.CodexAuthBundle{
-			TokenData: codex.CodexTokenData{
-				IDToken:      tokenResp.IDToken,
-				AccessToken:  tokenResp.AccessToken,
-				RefreshToken: tokenResp.RefreshToken,
-				AccountID:    accountID,
-				Email:        email,
-				Expire:       time.Now().Add(time.Duration(tokenResp.ExpiresIn) * time.Second).Format(time.RFC3339),
-			},
-			LastRefresh: time.Now().Format(time.RFC3339),
+		if claims != nil {
+			planType = strings.TrimSpace(claims.CodexAuthInfo.ChatgptPlanType)
+			if accountID := claims.GetAccountID(); accountID != "" {
+				digest := sha256.Sum256([]byte(accountID))
+				hashAccountID = hex.EncodeToString(digest[:])[:8]
+			}
 		}
 
 		// Create token storage and persist
@@ -1441,23 +1390,12 @@ func (h *Handler) RequestCodexToken(c *gin.Context) {
 }
 
 func (h *Handler) RequestAntigravityToken(c *gin.Context) {
-	const (
-		antigravityCallbackPort = 51121
-		antigravityClientID     = "1071006060591-tmhssin2h21lcre235vtolojh4g403ep.apps.googleusercontent.com"
-		antigravityClientSecret = "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"
-	)
-	var antigravityScopes = []string{
-		"https://www.googleapis.com/auth/cloud-platform",
-		"https://www.googleapis.com/auth/userinfo.email",
-		"https://www.googleapis.com/auth/userinfo.profile",
-		"https://www.googleapis.com/auth/cclog",
-		"https://www.googleapis.com/auth/experimentsandconfigs",
-	}
-
 	ctx := context.Background()
 
 	fmt.Println("Initializing Antigravity authentication...")
 
+	authSvc := antigravity.NewAntigravityAuth(h.cfg, nil)
+
 	state, errState := misc.GenerateRandomState()
 	if errState != nil {
 		log.Errorf("Failed to generate state parameter: %v", errState)
@@ -1465,17 +1403,8 @@ func (h *Handler) RequestAntigravityToken(c *gin.Context) {
 		return
 	}
 
-	redirectURI := fmt.Sprintf("http://localhost:%d/oauth-callback", antigravityCallbackPort)
-
-	params := url.Values{}
-	params.Set("access_type", "offline")
-	params.Set("client_id", antigravityClientID)
-	params.Set("prompt", "consent")
-	params.Set("redirect_uri", redirectURI)
-	params.Set("response_type", "code")
-	params.Set("scope", strings.Join(antigravityScopes, " "))
-	params.Set("state", state)
-	authURL := "https://accounts.google.com/o/oauth2/v2/auth?" + params.Encode()
+	redirectURI := fmt.Sprintf("http://localhost:%d/oauth-callback", antigravity.CallbackPort)
+	authURL := authSvc.BuildAuthURL(state, redirectURI)
 
 	RegisterOAuthSession(state, "antigravity")
 
@@ -1489,7 +1418,7 @@ func (h *Handler) RequestAntigravityToken(c *gin.Context) {
 			return
 		}
 		var errStart error
-		if forwarder, errStart = startCallbackForwarder(antigravityCallbackPort, "antigravity", targetURL); errStart != nil {
+		if forwarder, errStart = startCallbackForwarder(antigravity.CallbackPort, "antigravity", targetURL); errStart != nil {
 			log.WithError(errStart).Error("failed to start antigravity callback forwarder")
 			c.JSON(http.StatusInternalServerError, gin.H{"error": "failed to start callback server"})
 			return
@@ -1498,7 +1427,7 @@ func (h *Handler) RequestAntigravityToken(c *gin.Context) {
 
 	go func() {
 		if isWebUI {
-			defer stopCallbackForwarderInstance(antigravityCallbackPort, forwarder)
+			defer stopCallbackForwarderInstance(antigravity.CallbackPort, forwarder)
 		}
 
 		waitFile := filepath.Join(h.cfg.AuthDir, fmt.Sprintf(".oauth-antigravity-%s.oauth", state))
@@ -1538,93 +1467,36 @@ func (h *Handler) RequestAntigravityToken(c *gin.Context) {
 			time.Sleep(500 * time.Millisecond)
 		}
 
-		httpClient := util.SetProxy(&h.cfg.SDKConfig, &http.Client{})
-		form := url.Values{}
-		form.Set("code", authCode)
-		form.Set("client_id", antigravityClientID)
-		form.Set("client_secret", antigravityClientSecret)
-		form.Set("redirect_uri", redirectURI)
-		form.Set("grant_type", "authorization_code")
-
-		req, errNewRequest := http.NewRequestWithContext(ctx, http.MethodPost, "https://oauth2.googleapis.com/token", strings.NewReader(form.Encode()))
-		if errNewRequest != nil {
-			log.Errorf("Failed to build token request: %v", errNewRequest)
-			SetOAuthSessionError(state, "Failed to build token request")
-			return
-		}
-		req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
-
-		resp, errDo := httpClient.Do(req)
-		if errDo != nil {
-			log.Errorf("Failed to execute token request: %v", errDo)
+		tokenResp, errToken := authSvc.ExchangeCodeForTokens(ctx, authCode, redirectURI)
+		if errToken != nil {
+			log.Errorf("Failed to exchange token: %v", errToken)
 			SetOAuthSessionError(state, "Failed to exchange token")
 			return
 		}
-		defer func() {
-			if errClose := resp.Body.Close(); errClose != nil {
-				log.Errorf("antigravity token exchange close error: %v", errClose)
-			}
-		}()
 
-		if resp.StatusCode < http.StatusOK || resp.StatusCode >= http.StatusMultipleChoices {
-			bodyBytes, _ := io.ReadAll(resp.Body)
-			log.Errorf("Antigravity token exchange failed with status %d: %s", resp.StatusCode, string(bodyBytes))
-			SetOAuthSessionError(state, fmt.Sprintf("Token exchange failed: %d", resp.StatusCode))
+		accessToken := strings.TrimSpace(tokenResp.AccessToken)
+		if accessToken == "" {
+			log.Error("antigravity: token exchange returned empty access token")
+			SetOAuthSessionError(state, "Failed to exchange token")
 			return
 		}
 
-		var tokenResp struct {
-			AccessToken  string `json:"access_token"`
-			RefreshToken string `json:"refresh_token"`
-			ExpiresIn    int64  `json:"expires_in"`
-			TokenType    string `json:"token_type"`
-		}
-		if errDecode := json.NewDecoder(resp.Body).Decode(&tokenResp); errDecode != nil {
-			log.Errorf("Failed to parse token response: %v", errDecode)
-			SetOAuthSessionError(state, "Failed to parse token response")
+		email, errInfo := authSvc.FetchUserInfo(ctx, accessToken)
+		if errInfo != nil {
+			log.Errorf("Failed to fetch user info: %v", errInfo)
+			SetOAuthSessionError(state, "Failed to fetch user info")
 			return
 		}
-
-		email := ""
-		if strings.TrimSpace(tokenResp.AccessToken) != "" {
-			infoReq, errInfoReq := http.NewRequestWithContext(ctx, http.MethodGet, "https://www.googleapis.com/oauth2/v1/userinfo?alt=json", nil)
-			if errInfoReq != nil {
-				log.Errorf("Failed to build user info request: %v", errInfoReq)
-				SetOAuthSessionError(state, "Failed to build user info request")
-				return
-			}
-			infoReq.Header.Set("Authorization", "Bearer "+tokenResp.AccessToken)
-
-			infoResp, errInfo := httpClient.Do(infoReq)
-			if errInfo != nil {
-				log.Errorf("Failed to execute user info request: %v", errInfo)
-				SetOAuthSessionError(state, "Failed to execute user info request")
-				return
-			}
-			defer func() {
-				if errClose := infoResp.Body.Close(); errClose != nil {
-					log.Errorf("antigravity user info close error: %v", errClose)
-				}
-			}()
-
-			if infoResp.StatusCode >= http.StatusOK && infoResp.StatusCode < http.StatusMultipleChoices {
-				var infoPayload struct {
-					Email string `json:"email"`
-				}
-				if errDecodeInfo := json.NewDecoder(infoResp.Body).Decode(&infoPayload); errDecodeInfo == nil {
-					email = strings.TrimSpace(infoPayload.Email)
-				}
-			} else {
-				bodyBytes, _ := io.ReadAll(infoResp.Body)
-				log.Errorf("User info request failed with status %d: %s", infoResp.StatusCode, string(bodyBytes))
-				SetOAuthSessionError(state, fmt.Sprintf("User info request failed: %d", infoResp.StatusCode))
-				return
-			}
+		email = strings.TrimSpace(email)
+		if email == "" {
+			log.Error("antigravity: user info returned empty email")
+			SetOAuthSessionError(state, "Failed to fetch user info")
+			return
 		}
 
 		projectID := ""
-		if strings.TrimSpace(tokenResp.AccessToken) != "" {
-			fetchedProjectID, errProject := sdkAuth.FetchAntigravityProjectID(ctx, tokenResp.AccessToken, httpClient)
+		if accessToken != "" {
+			fetchedProjectID, errProject := authSvc.FetchProjectID(ctx, accessToken)
 			if errProject != nil {
 				log.Warnf("antigravity: failed to fetch project ID: %v", errProject)
 			} else {
@@ -1649,7 +1521,7 @@ func (h *Handler) RequestAntigravityToken(c *gin.Context) {
 			metadata["project_id"] = projectID
 		}
 
-		fileName := sanitizeAntigravityFileName(email)
+		fileName := antigravity.CredentialFileName(email)
 		label := strings.TrimSpace(email)
 		if label == "" {
 			label = "antigravity"

internal/api/handlers/management/model_definitions.go

@@ -0,0 +1,33 @@
+package management
+
+import (
+	"net/http"
+	"strings"
+
+	"github.com/gin-gonic/gin"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
+)
+
+// GetStaticModelDefinitions returns static model metadata for a given channel.
+// Channel is provided via path param (:channel) or query param (?channel=...).
+func (h *Handler) GetStaticModelDefinitions(c *gin.Context) {
+	channel := strings.TrimSpace(c.Param("channel"))
+	if channel == "" {
+		channel = strings.TrimSpace(c.Query("channel"))
+	}
+	if channel == "" {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "channel is required"})
+		return
+	}
+
+	models := registry.GetStaticModelDefinitionsByChannel(channel)
+	if models == nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "unknown channel", "channel": channel})
+		return
+	}
+
+	c.JSON(http.StatusOK, gin.H{
+		"channel": strings.ToLower(strings.TrimSpace(channel)),
+		"models":  models,
+	})
+}

internal/api/middleware/request_logging.go

@@ -8,6 +8,7 @@ import (
 	"io"
 	"net/http"
 	"strings"
+	"time"
 
 	"github.com/gin-gonic/gin"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/logging"
@@ -103,6 +104,7 @@ func captureRequestInfo(c *gin.Context) (*RequestInfo, error) {
 		Headers:   headers,
 		Body:      body,
 		RequestID: logging.GetGinRequestID(c),
+		Timestamp: time.Now(),
 	}, nil
 }
 

internal/api/middleware/response_writer.go

@@ -7,6 +7,7 @@ import (
 	"bytes"
 	"net/http"
 	"strings"
+	"time"
 
 	"github.com/gin-gonic/gin"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/interfaces"
@@ -20,22 +21,24 @@ type RequestInfo struct {
 	Headers   map[string][]string // Headers contains the request headers.
 	Body      []byte              // Body is the raw request body.
 	RequestID string              // RequestID is the unique identifier for the request.
+	Timestamp time.Time           // Timestamp is when the request was received.
 }
 
 // ResponseWriterWrapper wraps the standard gin.ResponseWriter to intercept and log response data.
 // It is designed to handle both standard and streaming responses, ensuring that logging operations do not block the client response.
 type ResponseWriterWrapper struct {
 	gin.ResponseWriter
-	body           *bytes.Buffer              // body is a buffer to store the response body for non-streaming responses.
-	isStreaming    bool                       // isStreaming indicates whether the response is a streaming type (e.g., text/event-stream).
-	streamWriter   logging.StreamingLogWriter // streamWriter is a writer for handling streaming log entries.
-	chunkChannel   chan []byte                // chunkChannel is a channel for asynchronously passing response chunks to the logger.
-	streamDone     chan struct{}              // streamDone signals when the streaming goroutine completes.
-	logger         logging.RequestLogger      // logger is the instance of the request logger service.
-	requestInfo    *RequestInfo               // requestInfo holds the details of the original request.
-	statusCode     int                        // statusCode stores the HTTP status code of the response.
-	headers        map[string][]string        // headers stores the response headers.
-	logOnErrorOnly bool                       // logOnErrorOnly enables logging only when an error response is detected.
+	body                *bytes.Buffer              // body is a buffer to store the response body for non-streaming responses.
+	isStreaming         bool                       // isStreaming indicates whether the response is a streaming type (e.g., text/event-stream).
+	streamWriter        logging.StreamingLogWriter // streamWriter is a writer for handling streaming log entries.
+	chunkChannel        chan []byte                // chunkChannel is a channel for asynchronously passing response chunks to the logger.
+	streamDone          chan struct{}              // streamDone signals when the streaming goroutine completes.
+	logger              logging.RequestLogger      // logger is the instance of the request logger service.
+	requestInfo         *RequestInfo               // requestInfo holds the details of the original request.
+	statusCode          int                        // statusCode stores the HTTP status code of the response.
+	headers             map[string][]string        // headers stores the response headers.
+	logOnErrorOnly      bool                       // logOnErrorOnly enables logging only when an error response is detected.
+	firstChunkTimestamp time.Time                  // firstChunkTimestamp captures TTFB for streaming responses.
 }
 
 // NewResponseWriterWrapper creates and initializes a new ResponseWriterWrapper.
@@ -73,6 +76,10 @@ func (w *ResponseWriterWrapper) Write(data []byte) (int, error) {
 
 	// THEN: Handle logging based on response type
 	if w.isStreaming && w.chunkChannel != nil {
+		// Capture TTFB on first chunk (synchronous, before async channel send)
+		if w.firstChunkTimestamp.IsZero() {
+			w.firstChunkTimestamp = time.Now()
+		}
 		// For streaming responses: Send to async logging channel (non-blocking)
 		select {
 		case w.chunkChannel <- append([]byte(nil), data...): // Non-blocking send with copy
@@ -117,6 +124,10 @@ func (w *ResponseWriterWrapper) WriteString(data string) (int, error) {
 
 	// THEN: Capture for logging
 	if w.isStreaming && w.chunkChannel != nil {
+		// Capture TTFB on first chunk (synchronous, before async channel send)
+		if w.firstChunkTimestamp.IsZero() {
+			w.firstChunkTimestamp = time.Now()
+		}
 		select {
 		case w.chunkChannel <- []byte(data):
 		default:
@@ -280,6 +291,8 @@ func (w *ResponseWriterWrapper) Finalize(c *gin.Context) error {
 			w.streamDone = nil
 		}
 
+		w.streamWriter.SetFirstChunkTimestamp(w.firstChunkTimestamp)
+
 		// Write API Request and Response to the streaming log before closing
 		apiRequest := w.extractAPIRequest(c)
 		if len(apiRequest) > 0 {
@@ -297,7 +310,7 @@ 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), slicesAPIResponseError, forceLog)
+	return w.logRequest(finalStatusCode, w.cloneHeaders(), w.body.Bytes(), w.extractAPIRequest(c), w.extractAPIResponse(c), w.extractAPIResponseTimestamp(c), slicesAPIResponseError, forceLog)
 }
 
 func (w *ResponseWriterWrapper) cloneHeaders() map[string][]string {
@@ -337,7 +350,18 @@ func (w *ResponseWriterWrapper) extractAPIResponse(c *gin.Context) []byte {
 	return data
 }
 
-func (w *ResponseWriterWrapper) logRequest(statusCode int, headers map[string][]string, body []byte, apiRequestBody, apiResponseBody []byte, apiResponseErrors []*interfaces.ErrorMessage, forceLog bool) error {
+func (w *ResponseWriterWrapper) extractAPIResponseTimestamp(c *gin.Context) time.Time {
+	ts, isExist := c.Get("API_RESPONSE_TIMESTAMP")
+	if !isExist {
+		return time.Time{}
+	}
+	if t, ok := ts.(time.Time); ok {
+		return t
+	}
+	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 {
 	if w.requestInfo == nil {
 		return nil
 	}
@@ -348,7 +372,7 @@ func (w *ResponseWriterWrapper) logRequest(statusCode int, headers map[string][]
 	}
 
 	if loggerWithOptions, ok := w.logger.(interface {
-		LogRequestWithOptions(string, string, map[string][]string, []byte, int, map[string][]string, []byte, []byte, []byte, []*interfaces.ErrorMessage, bool, string) error
+		LogRequestWithOptions(string, string, map[string][]string, []byte, int, map[string][]string, []byte, []byte, []byte, []*interfaces.ErrorMessage, bool, string, time.Time, time.Time) error
 	}); ok {
 		return loggerWithOptions.LogRequestWithOptions(
 			w.requestInfo.URL,
@@ -363,6 +387,8 @@ func (w *ResponseWriterWrapper) logRequest(statusCode int, headers map[string][]
 			apiResponseErrors,
 			forceLog,
 			w.requestInfo.RequestID,
+			w.requestInfo.Timestamp,
+			apiResponseTimestamp,
 		)
 	}
 
@@ -378,5 +404,7 @@ func (w *ResponseWriterWrapper) logRequest(statusCode int, headers map[string][]
 		apiResponseBody,
 		apiResponseErrors,
 		w.requestInfo.RequestID,
+		w.requestInfo.Timestamp,
+		apiResponseTimestamp,
 	)
 }

internal/api/server.go

@@ -12,6 +12,7 @@ import (
 	"net/http"
 	"os"
 	"path/filepath"
+	"reflect"
 	"strings"
 	"sync"
 	"sync/atomic"
@@ -607,9 +608,11 @@ func (s *Server) registerManagementRoutes() {
 
 		mgmt.GET("/auth-files", s.mgmt.ListAuthFiles)
 		mgmt.GET("/auth-files/models", s.mgmt.GetAuthFileModels)
+		mgmt.GET("/model-definitions/:channel", s.mgmt.GetStaticModelDefinitions)
 		mgmt.GET("/auth-files/download", s.mgmt.DownloadAuthFile)
 		mgmt.POST("/auth-files", s.mgmt.UploadAuthFile)
 		mgmt.DELETE("/auth-files", s.mgmt.DeleteAuthFile)
+		mgmt.PATCH("/auth-files/status", s.mgmt.PatchAuthFileStatus)
 		mgmt.POST("/vertex/import", s.mgmt.ImportVertexCredential)
 
 		mgmt.GET("/anthropic-auth-url", s.mgmt.RequestAnthropicToken)
@@ -988,14 +991,17 @@ func (s *Server) UpdateClients(cfg *config.Config) {
 		s.mgmt.SetAuthManager(s.handlers.AuthManager)
 	}
 
-	// Notify Amp module of config changes (for model mapping hot-reload)
-	if s.ampModule != nil {
-		log.Debugf("triggering amp module config update")
-		if err := s.ampModule.OnConfigUpdated(cfg); err != nil {
-			log.Errorf("failed to update Amp module config: %v", err)
+	// Notify Amp module only when Amp config has changed.
+	ampConfigChanged := oldCfg == nil || !reflect.DeepEqual(oldCfg.AmpCode, cfg.AmpCode)
+	if ampConfigChanged {
+		if s.ampModule != nil {
+			log.Debugf("triggering amp module config update")
+			if err := s.ampModule.OnConfigUpdated(cfg); err != nil {
+				log.Errorf("failed to update Amp module config: %v", err)
+			}
+		} else {
+			log.Warnf("amp module is nil, skipping config update")
 		}
-	} else {
-		log.Warnf("amp module is nil, skipping config update")
 	}
 
 	// Count client sources from configuration and auth store.

internal/auth/antigravity/auth.go

@@ -0,0 +1,344 @@
+// Package antigravity provides OAuth2 authentication functionality for the Antigravity provider.
+package antigravity
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+	"strings"
+	"time"
+
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
+	log "github.com/sirupsen/logrus"
+)
+
+// TokenResponse represents OAuth token response from Google
+type TokenResponse struct {
+	AccessToken  string `json:"access_token"`
+	RefreshToken string `json:"refresh_token"`
+	ExpiresIn    int64  `json:"expires_in"`
+	TokenType    string `json:"token_type"`
+}
+
+// userInfo represents Google user profile
+type userInfo struct {
+	Email string `json:"email"`
+}
+
+// AntigravityAuth handles Antigravity OAuth authentication
+type AntigravityAuth struct {
+	httpClient *http.Client
+}
+
+// NewAntigravityAuth creates a new Antigravity auth service.
+func NewAntigravityAuth(cfg *config.Config, httpClient *http.Client) *AntigravityAuth {
+	if httpClient != nil {
+		return &AntigravityAuth{httpClient: httpClient}
+	}
+	if cfg == nil {
+		cfg = &config.Config{}
+	}
+	return &AntigravityAuth{
+		httpClient: util.SetProxy(&cfg.SDKConfig, &http.Client{}),
+	}
+}
+
+// BuildAuthURL generates the OAuth authorization URL.
+func (o *AntigravityAuth) BuildAuthURL(state, redirectURI string) string {
+	if strings.TrimSpace(redirectURI) == "" {
+		redirectURI = fmt.Sprintf("http://localhost:%d/oauth-callback", CallbackPort)
+	}
+	params := url.Values{}
+	params.Set("access_type", "offline")
+	params.Set("client_id", ClientID)
+	params.Set("prompt", "consent")
+	params.Set("redirect_uri", redirectURI)
+	params.Set("response_type", "code")
+	params.Set("scope", strings.Join(Scopes, " "))
+	params.Set("state", state)
+	return AuthEndpoint + "?" + params.Encode()
+}
+
+// ExchangeCodeForTokens exchanges authorization code for access and refresh tokens
+func (o *AntigravityAuth) ExchangeCodeForTokens(ctx context.Context, code, redirectURI string) (*TokenResponse, error) {
+	data := url.Values{}
+	data.Set("code", code)
+	data.Set("client_id", ClientID)
+	data.Set("client_secret", ClientSecret)
+	data.Set("redirect_uri", redirectURI)
+	data.Set("grant_type", "authorization_code")
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, TokenEndpoint, strings.NewReader(data.Encode()))
+	if err != nil {
+		return nil, fmt.Errorf("antigravity token exchange: create request: %w", err)
+	}
+	req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
+
+	resp, errDo := o.httpClient.Do(req)
+	if errDo != nil {
+		return nil, fmt.Errorf("antigravity token exchange: execute request: %w", errDo)
+	}
+	defer func() {
+		if errClose := resp.Body.Close(); errClose != nil {
+			log.Errorf("antigravity token exchange: close body error: %v", errClose)
+		}
+	}()
+
+	if resp.StatusCode < http.StatusOK || resp.StatusCode >= http.StatusMultipleChoices {
+		bodyBytes, errRead := io.ReadAll(io.LimitReader(resp.Body, 8<<10))
+		if errRead != nil {
+			return nil, fmt.Errorf("antigravity token exchange: read response: %w", errRead)
+		}
+		body := strings.TrimSpace(string(bodyBytes))
+		if body == "" {
+			return nil, fmt.Errorf("antigravity token exchange: request failed: status %d", resp.StatusCode)
+		}
+		return nil, fmt.Errorf("antigravity token exchange: request failed: status %d: %s", resp.StatusCode, body)
+	}
+
+	var token TokenResponse
+	if errDecode := json.NewDecoder(resp.Body).Decode(&token); errDecode != nil {
+		return nil, fmt.Errorf("antigravity token exchange: decode response: %w", errDecode)
+	}
+	return &token, nil
+}
+
+// FetchUserInfo retrieves user email from Google
+func (o *AntigravityAuth) FetchUserInfo(ctx context.Context, accessToken string) (string, error) {
+	accessToken = strings.TrimSpace(accessToken)
+	if accessToken == "" {
+		return "", fmt.Errorf("antigravity userinfo: missing access token")
+	}
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, UserInfoEndpoint, nil)
+	if err != nil {
+		return "", fmt.Errorf("antigravity userinfo: create request: %w", err)
+	}
+	req.Header.Set("Authorization", "Bearer "+accessToken)
+
+	resp, errDo := o.httpClient.Do(req)
+	if errDo != nil {
+		return "", fmt.Errorf("antigravity userinfo: execute request: %w", errDo)
+	}
+	defer func() {
+		if errClose := resp.Body.Close(); errClose != nil {
+			log.Errorf("antigravity userinfo: close body error: %v", errClose)
+		}
+	}()
+
+	if resp.StatusCode < http.StatusOK || resp.StatusCode >= http.StatusMultipleChoices {
+		bodyBytes, errRead := io.ReadAll(io.LimitReader(resp.Body, 8<<10))
+		if errRead != nil {
+			return "", fmt.Errorf("antigravity userinfo: read response: %w", errRead)
+		}
+		body := strings.TrimSpace(string(bodyBytes))
+		if body == "" {
+			return "", fmt.Errorf("antigravity userinfo: request failed: status %d", resp.StatusCode)
+		}
+		return "", fmt.Errorf("antigravity userinfo: request failed: status %d: %s", resp.StatusCode, body)
+	}
+	var info userInfo
+	if errDecode := json.NewDecoder(resp.Body).Decode(&info); errDecode != nil {
+		return "", fmt.Errorf("antigravity userinfo: decode response: %w", errDecode)
+	}
+	email := strings.TrimSpace(info.Email)
+	if email == "" {
+		return "", fmt.Errorf("antigravity userinfo: response missing email")
+	}
+	return email, nil
+}
+
+// FetchProjectID retrieves the project ID for the authenticated user via loadCodeAssist
+func (o *AntigravityAuth) FetchProjectID(ctx context.Context, accessToken string) (string, error) {
+	loadReqBody := map[string]any{
+		"metadata": map[string]string{
+			"ideType":    "ANTIGRAVITY",
+			"platform":   "PLATFORM_UNSPECIFIED",
+			"pluginType": "GEMINI",
+		},
+	}
+
+	rawBody, errMarshal := json.Marshal(loadReqBody)
+	if errMarshal != nil {
+		return "", fmt.Errorf("marshal request body: %w", errMarshal)
+	}
+
+	endpointURL := fmt.Sprintf("%s/%s:loadCodeAssist", APIEndpoint, APIVersion)
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, endpointURL, strings.NewReader(string(rawBody)))
+	if err != nil {
+		return "", fmt.Errorf("create request: %w", err)
+	}
+	req.Header.Set("Authorization", "Bearer "+accessToken)
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("User-Agent", APIUserAgent)
+	req.Header.Set("X-Goog-Api-Client", APIClient)
+	req.Header.Set("Client-Metadata", ClientMetadata)
+
+	resp, errDo := o.httpClient.Do(req)
+	if errDo != nil {
+		return "", fmt.Errorf("execute request: %w", errDo)
+	}
+	defer func() {
+		if errClose := resp.Body.Close(); errClose != nil {
+			log.Errorf("antigravity loadCodeAssist: close body error: %v", errClose)
+		}
+	}()
+
+	bodyBytes, errRead := io.ReadAll(resp.Body)
+	if errRead != nil {
+		return "", fmt.Errorf("read response: %w", errRead)
+	}
+
+	if resp.StatusCode < http.StatusOK || resp.StatusCode >= http.StatusMultipleChoices {
+		return "", fmt.Errorf("request failed with status %d: %s", resp.StatusCode, strings.TrimSpace(string(bodyBytes)))
+	}
+
+	var loadResp map[string]any
+	if errDecode := json.Unmarshal(bodyBytes, &loadResp); errDecode != nil {
+		return "", fmt.Errorf("decode response: %w", errDecode)
+	}
+
+	// Extract projectID from response
+	projectID := ""
+	if id, ok := loadResp["cloudaicompanionProject"].(string); ok {
+		projectID = strings.TrimSpace(id)
+	}
+	if projectID == "" {
+		if projectMap, ok := loadResp["cloudaicompanionProject"].(map[string]any); ok {
+			if id, okID := projectMap["id"].(string); okID {
+				projectID = strings.TrimSpace(id)
+			}
+		}
+	}
+
+	if projectID == "" {
+		tierID := "legacy-tier"
+		if tiers, okTiers := loadResp["allowedTiers"].([]any); okTiers {
+			for _, rawTier := range tiers {
+				tier, okTier := rawTier.(map[string]any)
+				if !okTier {
+					continue
+				}
+				if isDefault, okDefault := tier["isDefault"].(bool); okDefault && isDefault {
+					if id, okID := tier["id"].(string); okID && strings.TrimSpace(id) != "" {
+						tierID = strings.TrimSpace(id)
+						break
+					}
+				}
+			}
+		}
+
+		projectID, err = o.OnboardUser(ctx, accessToken, tierID)
+		if err != nil {
+			return "", err
+		}
+		return projectID, nil
+	}
+
+	return projectID, nil
+}
+
+// OnboardUser attempts to fetch the project ID via onboardUser by polling for completion
+func (o *AntigravityAuth) OnboardUser(ctx context.Context, accessToken, tierID string) (string, error) {
+	log.Infof("Antigravity: onboarding user with tier: %s", tierID)
+	requestBody := map[string]any{
+		"tierId": tierID,
+		"metadata": map[string]string{
+			"ideType":    "ANTIGRAVITY",
+			"platform":   "PLATFORM_UNSPECIFIED",
+			"pluginType": "GEMINI",
+		},
+	}
+
+	rawBody, errMarshal := json.Marshal(requestBody)
+	if errMarshal != nil {
+		return "", fmt.Errorf("marshal request body: %w", errMarshal)
+	}
+
+	maxAttempts := 5
+	for attempt := 1; attempt <= maxAttempts; attempt++ {
+		log.Debugf("Polling attempt %d/%d", attempt, maxAttempts)
+
+		reqCtx := ctx
+		var cancel context.CancelFunc
+		if reqCtx == nil {
+			reqCtx = context.Background()
+		}
+		reqCtx, cancel = context.WithTimeout(reqCtx, 30*time.Second)
+
+		endpointURL := fmt.Sprintf("%s/%s:onboardUser", APIEndpoint, APIVersion)
+		req, errRequest := http.NewRequestWithContext(reqCtx, http.MethodPost, endpointURL, strings.NewReader(string(rawBody)))
+		if errRequest != nil {
+			cancel()
+			return "", fmt.Errorf("create request: %w", errRequest)
+		}
+		req.Header.Set("Authorization", "Bearer "+accessToken)
+		req.Header.Set("Content-Type", "application/json")
+		req.Header.Set("User-Agent", APIUserAgent)
+		req.Header.Set("X-Goog-Api-Client", APIClient)
+		req.Header.Set("Client-Metadata", ClientMetadata)
+
+		resp, errDo := o.httpClient.Do(req)
+		if errDo != nil {
+			cancel()
+			return "", fmt.Errorf("execute request: %w", errDo)
+		}
+
+		bodyBytes, errRead := io.ReadAll(resp.Body)
+		if errClose := resp.Body.Close(); errClose != nil {
+			log.Errorf("close body error: %v", errClose)
+		}
+		cancel()
+
+		if errRead != nil {
+			return "", fmt.Errorf("read response: %w", errRead)
+		}
+
+		if resp.StatusCode == http.StatusOK {
+			var data map[string]any
+			if errDecode := json.Unmarshal(bodyBytes, &data); errDecode != nil {
+				return "", fmt.Errorf("decode response: %w", errDecode)
+			}
+
+			if done, okDone := data["done"].(bool); okDone && done {
+				projectID := ""
+				if responseData, okResp := data["response"].(map[string]any); okResp {
+					switch projectValue := responseData["cloudaicompanionProject"].(type) {
+					case map[string]any:
+						if id, okID := projectValue["id"].(string); okID {
+							projectID = strings.TrimSpace(id)
+						}
+					case string:
+						projectID = strings.TrimSpace(projectValue)
+					}
+				}
+
+				if projectID != "" {
+					log.Infof("Successfully fetched project_id: %s", projectID)
+					return projectID, nil
+				}
+
+				return "", fmt.Errorf("no project_id in response")
+			}
+
+			time.Sleep(2 * time.Second)
+			continue
+		}
+
+		responsePreview := strings.TrimSpace(string(bodyBytes))
+		if len(responsePreview) > 500 {
+			responsePreview = responsePreview[:500]
+		}
+
+		responseErr := responsePreview
+		if len(responseErr) > 200 {
+			responseErr = responseErr[:200]
+		}
+		return "", fmt.Errorf("http %d: %s", resp.StatusCode, responseErr)
+	}
+
+	return "", nil
+}

internal/auth/antigravity/constants.go

@@ -0,0 +1,34 @@
+// Package antigravity provides OAuth2 authentication functionality for the Antigravity provider.
+package antigravity
+
+// OAuth client credentials and configuration
+const (
+	ClientID     = "1071006060591-tmhssin2h21lcre235vtolojh4g403ep.apps.googleusercontent.com"
+	ClientSecret = "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"
+	CallbackPort = 51121
+)
+
+// Scopes defines the OAuth scopes required for Antigravity authentication
+var Scopes = []string{
+	"https://www.googleapis.com/auth/cloud-platform",
+	"https://www.googleapis.com/auth/userinfo.email",
+	"https://www.googleapis.com/auth/userinfo.profile",
+	"https://www.googleapis.com/auth/cclog",
+	"https://www.googleapis.com/auth/experimentsandconfigs",
+}
+
+// OAuth2 endpoints for Google authentication
+const (
+	TokenEndpoint    = "https://oauth2.googleapis.com/token"
+	AuthEndpoint     = "https://accounts.google.com/o/oauth2/v2/auth"
+	UserInfoEndpoint = "https://www.googleapis.com/oauth2/v1/userinfo?alt=json"
+)
+
+// Antigravity API configuration
+const (
+	APIEndpoint    = "https://cloudcode-pa.googleapis.com"
+	APIVersion     = "v1internal"
+	APIUserAgent   = "google-api-nodejs-client/9.15.1"
+	APIClient      = "google-cloud-sdk vscode_cloudshelleditor/0.1"
+	ClientMetadata = `{"ideType":"IDE_UNSPECIFIED","platform":"PLATFORM_UNSPECIFIED","pluginType":"GEMINI"}`
+)

internal/auth/antigravity/filename.go

@@ -0,0 +1,16 @@
+package antigravity
+
+import (
+	"fmt"
+	"strings"
+)
+
+// CredentialFileName returns the filename used to persist Antigravity credentials.
+// It uses the email as a suffix to disambiguate accounts.
+func CredentialFileName(email string) string {
+	email = strings.TrimSpace(email)
+	if email == "" {
+		return "antigravity.json"
+	}
+	return fmt.Sprintf("antigravity-%s.json", email)
+}

internal/auth/claude/anthropic_auth.go

@@ -14,15 +14,15 @@ import (
 	"time"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
-	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 	log "github.com/sirupsen/logrus"
 )
 
+// OAuth configuration constants for Claude/Anthropic
 const (
-	anthropicAuthURL  = "https://claude.ai/oauth/authorize"
-	anthropicTokenURL = "https://console.anthropic.com/v1/oauth/token"
-	anthropicClientID = "9d1c250a-e61b-44d9-88ed-5944d1962f5e"
-	redirectURI       = "http://localhost:54545/callback"
+	AuthURL     = "https://claude.ai/oauth/authorize"
+	TokenURL    = "https://console.anthropic.com/v1/oauth/token"
+	ClientID    = "9d1c250a-e61b-44d9-88ed-5944d1962f5e"
+	RedirectURI = "http://localhost:54545/callback"
 )
 
 // tokenResponse represents the response structure from Anthropic's OAuth token endpoint.
@@ -50,7 +50,8 @@ type ClaudeAuth struct {
 }
 
 // NewClaudeAuth creates a new Anthropic authentication service.
-// It initializes the HTTP client with proxy settings from the configuration.
+// It initializes the HTTP client with a custom TLS transport that uses Firefox
+// fingerprint to bypass Cloudflare's TLS fingerprinting on Anthropic domains.
 //
 // Parameters:
 //   - cfg: The application configuration containing proxy settings
@@ -58,8 +59,10 @@ type ClaudeAuth struct {
 // Returns:
 //   - *ClaudeAuth: A new Claude authentication service instance
 func NewClaudeAuth(cfg *config.Config) *ClaudeAuth {
+	// Use custom HTTP client with Firefox TLS fingerprint to bypass
+	// Cloudflare's bot detection on Anthropic domains
 	return &ClaudeAuth{
-		httpClient: util.SetProxy(&cfg.SDKConfig, &http.Client{}),
+		httpClient: NewAnthropicHttpClient(&cfg.SDKConfig),
 	}
 }
 
@@ -82,16 +85,16 @@ func (o *ClaudeAuth) GenerateAuthURL(state string, pkceCodes *PKCECodes) (string
 
 	params := url.Values{
 		"code":                  {"true"},
-		"client_id":             {anthropicClientID},
+		"client_id":             {ClientID},
 		"response_type":         {"code"},
-		"redirect_uri":          {redirectURI},
+		"redirect_uri":          {RedirectURI},
 		"scope":                 {"org:create_api_key user:profile user:inference"},
 		"code_challenge":        {pkceCodes.CodeChallenge},
 		"code_challenge_method": {"S256"},
 		"state":                 {state},
 	}
 
-	authURL := fmt.Sprintf("%s?%s", anthropicAuthURL, params.Encode())
+	authURL := fmt.Sprintf("%s?%s", AuthURL, params.Encode())
 	return authURL, state, nil
 }
 
@@ -137,8 +140,8 @@ func (o *ClaudeAuth) ExchangeCodeForTokens(ctx context.Context, code, state stri
 		"code":          newCode,
 		"state":         state,
 		"grant_type":    "authorization_code",
-		"client_id":     anthropicClientID,
-		"redirect_uri":  redirectURI,
+		"client_id":     ClientID,
+		"redirect_uri":  RedirectURI,
 		"code_verifier": pkceCodes.CodeVerifier,
 	}
 
@@ -154,7 +157,7 @@ func (o *ClaudeAuth) ExchangeCodeForTokens(ctx context.Context, code, state stri
 
 	// log.Debugf("Token exchange request: %s", string(jsonBody))
 
-	req, err := http.NewRequestWithContext(ctx, "POST", anthropicTokenURL, strings.NewReader(string(jsonBody)))
+	req, err := http.NewRequestWithContext(ctx, "POST", TokenURL, strings.NewReader(string(jsonBody)))
 	if err != nil {
 		return nil, fmt.Errorf("failed to create token request: %w", err)
 	}
@@ -221,7 +224,7 @@ func (o *ClaudeAuth) RefreshTokens(ctx context.Context, refreshToken string) (*C
 	}
 
 	reqBody := map[string]interface{}{
-		"client_id":     anthropicClientID,
+		"client_id":     ClientID,
 		"grant_type":    "refresh_token",
 		"refresh_token": refreshToken,
 	}
@@ -231,7 +234,7 @@ func (o *ClaudeAuth) RefreshTokens(ctx context.Context, refreshToken string) (*C
 		return nil, fmt.Errorf("failed to marshal request body: %w", err)
 	}
 
-	req, err := http.NewRequestWithContext(ctx, "POST", anthropicTokenURL, strings.NewReader(string(jsonBody)))
+	req, err := http.NewRequestWithContext(ctx, "POST", TokenURL, strings.NewReader(string(jsonBody)))
 	if err != nil {
 		return nil, fmt.Errorf("failed to create refresh request: %w", err)
 	}

internal/auth/claude/utls_transport.go

@@ -0,0 +1,165 @@
+// Package claude provides authentication functionality for Anthropic's Claude API.
+// This file implements a custom HTTP transport using utls to bypass TLS fingerprinting.
+package claude
+
+import (
+	"net/http"
+	"net/url"
+	"strings"
+	"sync"
+
+	tls "github.com/refraction-networking/utls"
+	"github.com/router-for-me/CLIProxyAPI/v6/sdk/config"
+	log "github.com/sirupsen/logrus"
+	"golang.org/x/net/http2"
+	"golang.org/x/net/proxy"
+)
+
+// utlsRoundTripper implements http.RoundTripper using utls with Firefox fingerprint
+// to bypass Cloudflare's TLS fingerprinting on Anthropic domains.
+type utlsRoundTripper struct {
+	// mu protects the connections map and pending map
+	mu sync.Mutex
+	// connections caches HTTP/2 client connections per host
+	connections map[string]*http2.ClientConn
+	// pending tracks hosts that are currently being connected to (prevents race condition)
+	pending map[string]*sync.Cond
+	// dialer is used to create network connections, supporting proxies
+	dialer proxy.Dialer
+}
+
+// newUtlsRoundTripper creates a new utls-based round tripper with optional proxy support
+func newUtlsRoundTripper(cfg *config.SDKConfig) *utlsRoundTripper {
+	var dialer proxy.Dialer = proxy.Direct
+	if cfg != nil && cfg.ProxyURL != "" {
+		proxyURL, err := url.Parse(cfg.ProxyURL)
+		if err != nil {
+			log.Errorf("failed to parse proxy URL %q: %v", cfg.ProxyURL, err)
+		} else {
+			pDialer, err := proxy.FromURL(proxyURL, proxy.Direct)
+			if err != nil {
+				log.Errorf("failed to create proxy dialer for %q: %v", cfg.ProxyURL, err)
+			} else {
+				dialer = pDialer
+			}
+		}
+	}
+
+	return &utlsRoundTripper{
+		connections: make(map[string]*http2.ClientConn),
+		pending:     make(map[string]*sync.Cond),
+		dialer:      dialer,
+	}
+}
+
+// getOrCreateConnection gets an existing connection or creates a new one.
+// It uses a per-host locking mechanism to prevent multiple goroutines from
+// creating connections to the same host simultaneously.
+func (t *utlsRoundTripper) getOrCreateConnection(host, addr string) (*http2.ClientConn, error) {
+	t.mu.Lock()
+
+	// Check if connection exists and is usable
+	if h2Conn, ok := t.connections[host]; ok && h2Conn.CanTakeNewRequest() {
+		t.mu.Unlock()
+		return h2Conn, nil
+	}
+
+	// Check if another goroutine is already creating a connection
+	if cond, ok := t.pending[host]; ok {
+		// Wait for the other goroutine to finish
+		cond.Wait()
+		// Check if connection is now available
+		if h2Conn, ok := t.connections[host]; ok && h2Conn.CanTakeNewRequest() {
+			t.mu.Unlock()
+			return h2Conn, nil
+		}
+		// Connection still not available, we'll create one
+	}
+
+	// Mark this host as pending
+	cond := sync.NewCond(&t.mu)
+	t.pending[host] = cond
+	t.mu.Unlock()
+
+	// Create connection outside the lock
+	h2Conn, err := t.createConnection(host, addr)
+
+	t.mu.Lock()
+	defer t.mu.Unlock()
+
+	// Remove pending marker and wake up waiting goroutines
+	delete(t.pending, host)
+	cond.Broadcast()
+
+	if err != nil {
+		return nil, err
+	}
+
+	// Store the new connection
+	t.connections[host] = h2Conn
+	return h2Conn, nil
+}
+
+// createConnection creates a new HTTP/2 connection with Firefox TLS fingerprint
+func (t *utlsRoundTripper) createConnection(host, addr string) (*http2.ClientConn, error) {
+	conn, err := t.dialer.Dial("tcp", addr)
+	if err != nil {
+		return nil, err
+	}
+
+	tlsConfig := &tls.Config{ServerName: host}
+	tlsConn := tls.UClient(conn, tlsConfig, tls.HelloFirefox_Auto)
+
+	if err := tlsConn.Handshake(); err != nil {
+		conn.Close()
+		return nil, err
+	}
+
+	tr := &http2.Transport{}
+	h2Conn, err := tr.NewClientConn(tlsConn)
+	if err != nil {
+		tlsConn.Close()
+		return nil, err
+	}
+
+	return h2Conn, nil
+}
+
+// RoundTrip implements http.RoundTripper
+func (t *utlsRoundTripper) RoundTrip(req *http.Request) (*http.Response, error) {
+	host := req.URL.Host
+	addr := host
+	if !strings.Contains(addr, ":") {
+		addr += ":443"
+	}
+
+	// Get hostname without port for TLS ServerName
+	hostname := req.URL.Hostname()
+
+	h2Conn, err := t.getOrCreateConnection(hostname, addr)
+	if err != nil {
+		return nil, err
+	}
+
+	resp, err := h2Conn.RoundTrip(req)
+	if err != nil {
+		// Connection failed, remove it from cache
+		t.mu.Lock()
+		if cached, ok := t.connections[hostname]; ok && cached == h2Conn {
+			delete(t.connections, hostname)
+		}
+		t.mu.Unlock()
+		return nil, err
+	}
+
+	return resp, nil
+}
+
+// NewAnthropicHttpClient creates an HTTP client that bypasses TLS fingerprinting
+// for Anthropic domains by using utls with Firefox fingerprint.
+// It accepts optional SDK configuration for proxy settings.
+func NewAnthropicHttpClient(cfg *config.SDKConfig) *http.Client {
+	return &http.Client{
+		Transport: newUtlsRoundTripper(cfg),
+	}
+}

internal/auth/codex/filename.go

@@ -4,9 +4,6 @@ import (
 	"fmt"
 	"strings"
 	"unicode"
-
-	"golang.org/x/text/cases"
-	"golang.org/x/text/language"
 )
 
 // CredentialFileName returns the filename used to persist Codex OAuth credentials.
@@ -43,15 +40,7 @@ func normalizePlanTypeForFilename(planType string) string {
 	}
 
 	for i, part := range parts {
-		parts[i] = titleToken(part)
+		parts[i] = strings.ToLower(strings.TrimSpace(part))
 	}
 	return strings.Join(parts, "-")
 }
-
-func titleToken(token string) string {
-	token = strings.TrimSpace(token)
-	if token == "" {
-		return ""
-	}
-	return cases.Title(language.English).String(token)
-}

internal/auth/codex/openai_auth.go

@@ -19,11 +19,12 @@ import (
 	log "github.com/sirupsen/logrus"
 )
 
+// OAuth configuration constants for OpenAI Codex
 const (
-	openaiAuthURL  = "https://auth.openai.com/oauth/authorize"
-	openaiTokenURL = "https://auth.openai.com/oauth/token"
-	openaiClientID = "app_EMoamEEZ73f0CkXaXp7hrann"
-	redirectURI    = "http://localhost:1455/auth/callback"
+	AuthURL     = "https://auth.openai.com/oauth/authorize"
+	TokenURL    = "https://auth.openai.com/oauth/token"
+	ClientID    = "app_EMoamEEZ73f0CkXaXp7hrann"
+	RedirectURI = "http://localhost:1455/auth/callback"
 )
 
 // CodexAuth handles the OpenAI OAuth2 authentication flow.
@@ -50,9 +51,9 @@ func (o *CodexAuth) GenerateAuthURL(state string, pkceCodes *PKCECodes) (string,
 	}
 
 	params := url.Values{
-		"client_id":                  {openaiClientID},
+		"client_id":                  {ClientID},
 		"response_type":              {"code"},
-		"redirect_uri":               {redirectURI},
+		"redirect_uri":               {RedirectURI},
 		"scope":                      {"openid email profile offline_access"},
 		"state":                      {state},
 		"code_challenge":             {pkceCodes.CodeChallenge},
@@ -62,7 +63,7 @@ func (o *CodexAuth) GenerateAuthURL(state string, pkceCodes *PKCECodes) (string,
 		"codex_cli_simplified_flow":  {"true"},
 	}
 
-	authURL := fmt.Sprintf("%s?%s", openaiAuthURL, params.Encode())
+	authURL := fmt.Sprintf("%s?%s", AuthURL, params.Encode())
 	return authURL, nil
 }
 
@@ -77,13 +78,13 @@ func (o *CodexAuth) ExchangeCodeForTokens(ctx context.Context, code string, pkce
 	// Prepare token exchange request
 	data := url.Values{
 		"grant_type":    {"authorization_code"},
-		"client_id":     {openaiClientID},
+		"client_id":     {ClientID},
 		"code":          {code},
-		"redirect_uri":  {redirectURI},
+		"redirect_uri":  {RedirectURI},
 		"code_verifier": {pkceCodes.CodeVerifier},
 	}
 
-	req, err := http.NewRequestWithContext(ctx, "POST", openaiTokenURL, strings.NewReader(data.Encode()))
+	req, err := http.NewRequestWithContext(ctx, "POST", TokenURL, strings.NewReader(data.Encode()))
 	if err != nil {
 		return nil, fmt.Errorf("failed to create token request: %w", err)
 	}
@@ -163,13 +164,13 @@ func (o *CodexAuth) RefreshTokens(ctx context.Context, refreshToken string) (*Co
 	}
 
 	data := url.Values{
-		"client_id":     {openaiClientID},
+		"client_id":     {ClientID},
 		"grant_type":    {"refresh_token"},
 		"refresh_token": {refreshToken},
 		"scope":         {"openid profile email"},
 	}
 
-	req, err := http.NewRequestWithContext(ctx, "POST", openaiTokenURL, strings.NewReader(data.Encode()))
+	req, err := http.NewRequestWithContext(ctx, "POST", TokenURL, strings.NewReader(data.Encode()))
 	if err != nil {
 		return nil, fmt.Errorf("failed to create refresh request: %w", err)
 	}

internal/auth/gemini/gemini_auth.go

@@ -28,19 +28,19 @@ import (
 	"golang.org/x/oauth2/google"
 )
 
+// OAuth configuration constants for Gemini
 const (
-	geminiOauthClientID       = "681255809395-oo8ft2oprdrnp9e3aqf6av3hmdib135j.apps.googleusercontent.com"
-	geminiOauthClientSecret   = "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"
-	geminiDefaultCallbackPort = 8085
+	ClientID            = "681255809395-oo8ft2oprdrnp9e3aqf6av3hmdib135j.apps.googleusercontent.com"
+	ClientSecret        = "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"
+	DefaultCallbackPort = 8085
 )
 
-var (
-	geminiOauthScopes = []string{
-		"https://www.googleapis.com/auth/cloud-platform",
-		"https://www.googleapis.com/auth/userinfo.email",
-		"https://www.googleapis.com/auth/userinfo.profile",
-	}
-)
+// OAuth scopes for Gemini authentication
+var Scopes = []string{
+	"https://www.googleapis.com/auth/cloud-platform",
+	"https://www.googleapis.com/auth/userinfo.email",
+	"https://www.googleapis.com/auth/userinfo.profile",
+}
 
 // GeminiAuth provides methods for handling the Gemini OAuth2 authentication flow.
 // It encapsulates the logic for obtaining, storing, and refreshing authentication tokens
@@ -74,7 +74,7 @@ func NewGeminiAuth() *GeminiAuth {
 //   - *http.Client: An HTTP client configured with authentication
 //   - error: An error if the client configuration fails, nil otherwise
 func (g *GeminiAuth) GetAuthenticatedClient(ctx context.Context, ts *GeminiTokenStorage, cfg *config.Config, opts *WebLoginOptions) (*http.Client, error) {
-	callbackPort := geminiDefaultCallbackPort
+	callbackPort := DefaultCallbackPort
 	if opts != nil && opts.CallbackPort > 0 {
 		callbackPort = opts.CallbackPort
 	}
@@ -112,10 +112,10 @@ func (g *GeminiAuth) GetAuthenticatedClient(ctx context.Context, ts *GeminiToken
 
 	// Configure the OAuth2 client.
 	conf := &oauth2.Config{
-		ClientID:     geminiOauthClientID,
-		ClientSecret: geminiOauthClientSecret,
+		ClientID:     ClientID,
+		ClientSecret: ClientSecret,
 		RedirectURL:  callbackURL, // This will be used by the local server.
-		Scopes:       geminiOauthScopes,
+		Scopes:       Scopes,
 		Endpoint:     google.Endpoint,
 	}
 
@@ -198,9 +198,9 @@ func (g *GeminiAuth) createTokenStorage(ctx context.Context, config *oauth2.Conf
 	}
 
 	ifToken["token_uri"] = "https://oauth2.googleapis.com/token"
-	ifToken["client_id"] = geminiOauthClientID
-	ifToken["client_secret"] = geminiOauthClientSecret
-	ifToken["scopes"] = geminiOauthScopes
+	ifToken["client_id"] = ClientID
+	ifToken["client_secret"] = ClientSecret
+	ifToken["scopes"] = Scopes
 	ifToken["universe_domain"] = "googleapis.com"
 
 	ts := GeminiTokenStorage{
@@ -226,7 +226,7 @@ func (g *GeminiAuth) createTokenStorage(ctx context.Context, config *oauth2.Conf
 //   - *oauth2.Token: The OAuth2 token obtained from the authorization flow
 //   - error: An error if the token acquisition fails, nil otherwise
 func (g *GeminiAuth) getTokenFromWeb(ctx context.Context, config *oauth2.Config, opts *WebLoginOptions) (*oauth2.Token, error) {
-	callbackPort := geminiDefaultCallbackPort
+	callbackPort := DefaultCallbackPort
 	if opts != nil && opts.CallbackPort > 0 {
 		callbackPort = opts.CallbackPort
 	}

internal/cache/signature_cache.go

@@ -3,7 +3,6 @@ package cache
 import (
 	"crypto/sha256"
 	"encoding/hex"
-	"fmt"
 	"strings"
 	"sync"
 	"time"
@@ -25,18 +24,18 @@ const (
 	// MinValidSignatureLen is the minimum length for a signature to be considered valid
 	MinValidSignatureLen = 50
 
-	// SessionCleanupInterval controls how often stale sessions are purged
-	SessionCleanupInterval = 10 * time.Minute
+	// CacheCleanupInterval controls how often stale entries are purged
+	CacheCleanupInterval = 10 * time.Minute
 )
 
-// signatureCache stores signatures by sessionId -> textHash -> SignatureEntry
+// signatureCache stores signatures by model group -> textHash -> SignatureEntry
 var signatureCache sync.Map
 
-// sessionCleanupOnce ensures the background cleanup goroutine starts only once
-var sessionCleanupOnce sync.Once
+// cacheCleanupOnce ensures the background cleanup goroutine starts only once
+var cacheCleanupOnce sync.Once
 
-// sessionCache is the inner map type
-type sessionCache struct {
+// groupCache is the inner map type
+type groupCache struct {
 	mu      sync.RWMutex
 	entries map[string]SignatureEntry
 }
@@ -47,36 +46,36 @@ func hashText(text string) string {
 	return hex.EncodeToString(h[:])[:SignatureTextHashLen]
 }
 
-// getOrCreateSession gets or creates a session cache
-func getOrCreateSession(sessionID string) *sessionCache {
+// getOrCreateGroupCache gets or creates a cache bucket for a model group
+func getOrCreateGroupCache(groupKey string) *groupCache {
 	// Start background cleanup on first access
-	sessionCleanupOnce.Do(startSessionCleanup)
+	cacheCleanupOnce.Do(startCacheCleanup)
 
-	if val, ok := signatureCache.Load(sessionID); ok {
-		return val.(*sessionCache)
+	if val, ok := signatureCache.Load(groupKey); ok {
+		return val.(*groupCache)
 	}
-	sc := &sessionCache{entries: make(map[string]SignatureEntry)}
-	actual, _ := signatureCache.LoadOrStore(sessionID, sc)
-	return actual.(*sessionCache)
+	sc := &groupCache{entries: make(map[string]SignatureEntry)}
+	actual, _ := signatureCache.LoadOrStore(groupKey, sc)
+	return actual.(*groupCache)
 }
 
-// startSessionCleanup launches a background goroutine that periodically
-// removes sessions where all entries have expired.
-func startSessionCleanup() {
+// startCacheCleanup launches a background goroutine that periodically
+// removes caches where all entries have expired.
+func startCacheCleanup() {
 	go func() {
-		ticker := time.NewTicker(SessionCleanupInterval)
+		ticker := time.NewTicker(CacheCleanupInterval)
 		defer ticker.Stop()
 		for range ticker.C {
-			purgeExpiredSessions()
+			purgeExpiredCaches()
 		}
 	}()
 }
 
-// purgeExpiredSessions removes sessions with no valid (non-expired) entries.
-func purgeExpiredSessions() {
+// purgeExpiredCaches removes caches with no valid (non-expired) entries.
+func purgeExpiredCaches() {
 	now := time.Now()
 	signatureCache.Range(func(key, value any) bool {
-		sc := value.(*sessionCache)
+		sc := value.(*groupCache)
 		sc.mu.Lock()
 		// Remove expired entries
 		for k, entry := range sc.entries {
@@ -86,7 +85,7 @@ func purgeExpiredSessions() {
 		}
 		isEmpty := len(sc.entries) == 0
 		sc.mu.Unlock()
-		// Remove session if empty
+		// Remove cache bucket if empty
 		if isEmpty {
 			signatureCache.Delete(key)
 		}
@@ -94,7 +93,7 @@ func purgeExpiredSessions() {
 	})
 }
 
-// CacheSignature stores a thinking signature for a given session and text.
+// CacheSignature stores a thinking signature for a given model group and text.
 // Used for Claude models that require signed thinking blocks in multi-turn conversations.
 func CacheSignature(modelName, text, signature string) {
 	if text == "" || signature == "" {
@@ -104,9 +103,9 @@ func CacheSignature(modelName, text, signature string) {
 		return
 	}
 
-	text = fmt.Sprintf("%s#%s", GetModelGroup(modelName), text)
+	groupKey := GetModelGroup(modelName)
 	textHash := hashText(text)
-	sc := getOrCreateSession(textHash)
+	sc := getOrCreateGroupCache(groupKey)
 	sc.mu.Lock()
 	defer sc.mu.Unlock()
 
@@ -116,26 +115,25 @@ func CacheSignature(modelName, text, signature string) {
 	}
 }
 
-// GetCachedSignature retrieves a cached signature for a given session and text.
+// GetCachedSignature retrieves a cached signature for a given model group and text.
 // Returns empty string if not found or expired.
 func GetCachedSignature(modelName, text string) string {
-	family := GetModelGroup(modelName)
+	groupKey := GetModelGroup(modelName)
 
 	if text == "" {
-		if family == "gemini" {
+		if groupKey == "gemini" {
 			return "skip_thought_signature_validator"
 		}
 		return ""
 	}
-	text = fmt.Sprintf("%s#%s", GetModelGroup(modelName), text)
-	val, ok := signatureCache.Load(hashText(text))
+	val, ok := signatureCache.Load(groupKey)
 	if !ok {
-		if family == "gemini" {
+		if groupKey == "gemini" {
 			return "skip_thought_signature_validator"
 		}
 		return ""
 	}
-	sc := val.(*sessionCache)
+	sc := val.(*groupCache)
 
 	textHash := hashText(text)
 
@@ -145,7 +143,7 @@ func GetCachedSignature(modelName, text string) string {
 	entry, exists := sc.entries[textHash]
 	if !exists {
 		sc.mu.Unlock()
-		if family == "gemini" {
+		if groupKey == "gemini" {
 			return "skip_thought_signature_validator"
 		}
 		return ""
@@ -153,7 +151,7 @@ func GetCachedSignature(modelName, text string) string {
 	if now.Sub(entry.Timestamp) > SignatureCacheTTL {
 		delete(sc.entries, textHash)
 		sc.mu.Unlock()
-		if family == "gemini" {
+		if groupKey == "gemini" {
 			return "skip_thought_signature_validator"
 		}
 		return ""
@@ -167,22 +165,17 @@ func GetCachedSignature(modelName, text string) string {
 	return entry.Signature
 }
 
-// ClearSignatureCache clears signature cache for a specific session or all sessions.
-func ClearSignatureCache(sessionID string) {
-	if sessionID != "" {
-		signatureCache.Range(func(key, _ any) bool {
-			kStr, ok := key.(string)
-			if ok && strings.HasSuffix(kStr, "#"+sessionID) {
-				signatureCache.Delete(key)
-			}
-			return true
-		})
-	} else {
+// ClearSignatureCache clears signature cache for a specific model group or all groups.
+func ClearSignatureCache(modelName string) {
+	if modelName == "" {
 		signatureCache.Range(func(key, _ any) bool {
 			signatureCache.Delete(key)
 			return true
 		})
+		return
 	}
+	groupKey := GetModelGroup(modelName)
+	signatureCache.Delete(groupKey)
 }
 
 // HasValidSignature checks if a signature is valid (non-empty and long enough)

internal/cache/signature_cache_test.go

@@ -5,6 +5,8 @@ import (
 	"time"
 )
 
+const testModelName = "claude-sonnet-4-5"
+
 func TestCacheSignature_BasicStorageAndRetrieval(t *testing.T) {
 	ClearSignatureCache("")
 
@@ -12,30 +14,31 @@ func TestCacheSignature_BasicStorageAndRetrieval(t *testing.T) {
 	signature := "abc123validSignature1234567890123456789012345678901234567890"
 
 	// Store signature
-	CacheSignature("test-model", text, signature)
+	CacheSignature(testModelName, text, signature)
 
 	// Retrieve signature
-	retrieved := GetCachedSignature("test-model", text)
+	retrieved := GetCachedSignature(testModelName, text)
 	if retrieved != signature {
 		t.Errorf("Expected signature '%s', got '%s'", signature, retrieved)
 	}
 }
 
-func TestCacheSignature_DifferentSessions(t *testing.T) {
+func TestCacheSignature_DifferentModelGroups(t *testing.T) {
 	ClearSignatureCache("")
 
-	text := "Same text in different sessions"
+	text := "Same text across models"
 	sig1 := "signature1_1234567890123456789012345678901234567890123456"
 	sig2 := "signature2_1234567890123456789012345678901234567890123456"
 
-	CacheSignature("test-model", text, sig1)
-	CacheSignature("test-model", text, sig2)
+	geminiModel := "gemini-3-pro-preview"
+	CacheSignature(testModelName, text, sig1)
+	CacheSignature(geminiModel, text, sig2)
 
-	if GetCachedSignature("test-model", text) != sig1 {
-		t.Error("Session-a signature mismatch")
+	if GetCachedSignature(testModelName, text) != sig1 {
+		t.Error("Claude signature mismatch")
 	}
-	if GetCachedSignature("test-model", text) != sig2 {
-		t.Error("Session-b signature mismatch")
+	if GetCachedSignature(geminiModel, text) != sig2 {
+		t.Error("Gemini signature mismatch")
 	}
 }
 
@@ -43,13 +46,13 @@ func TestCacheSignature_NotFound(t *testing.T) {
 	ClearSignatureCache("")
 
 	// Non-existent session
-	if got := GetCachedSignature("test-model", "some text"); got != "" {
+	if got := GetCachedSignature(testModelName, "some text"); got != "" {
 		t.Errorf("Expected empty string for nonexistent session, got '%s'", got)
 	}
 
 	// Existing session but different text
-	CacheSignature("test-model", "text-a", "sigA12345678901234567890123456789012345678901234567890")
-	if got := GetCachedSignature("test-model", "text-b"); got != "" {
+	CacheSignature(testModelName, "text-a", "sigA12345678901234567890123456789012345678901234567890")
+	if got := GetCachedSignature(testModelName, "text-b"); got != "" {
 		t.Errorf("Expected empty string for different text, got '%s'", got)
 	}
 }
@@ -58,12 +61,11 @@ func TestCacheSignature_EmptyInputs(t *testing.T) {
 	ClearSignatureCache("")
 
 	// All empty/invalid inputs should be no-ops
-	CacheSignature("test-model", "text", "sig12345678901234567890123456789012345678901234567890")
-	CacheSignature("test-model", "", "sig12345678901234567890123456789012345678901234567890")
-	CacheSignature("test-model", "text", "")
-	CacheSignature("test-model", "text", "short") // Too short
+	CacheSignature(testModelName, "", "sig12345678901234567890123456789012345678901234567890")
+	CacheSignature(testModelName, "text", "")
+	CacheSignature(testModelName, "text", "short") // Too short
 
-	if got := GetCachedSignature("test-model", "text"); got != "" {
+	if got := GetCachedSignature(testModelName, "text"); got != "" {
 		t.Errorf("Expected empty after invalid cache attempts, got '%s'", got)
 	}
 }
@@ -74,27 +76,24 @@ func TestCacheSignature_ShortSignatureRejected(t *testing.T) {
 	text := "Some text"
 	shortSig := "abc123" // Less than 50 chars
 
-	CacheSignature("test-model", text, shortSig)
+	CacheSignature(testModelName, text, shortSig)
 
-	if got := GetCachedSignature("test-model", text); got != "" {
+	if got := GetCachedSignature(testModelName, text); got != "" {
 		t.Errorf("Short signature should be rejected, got '%s'", got)
 	}
 }
 
-func TestClearSignatureCache_SpecificSession(t *testing.T) {
+func TestClearSignatureCache_ModelGroup(t *testing.T) {
 	ClearSignatureCache("")
 
 	sig := "validSig1234567890123456789012345678901234567890123456"
-	CacheSignature("test-model", "text", sig)
-	CacheSignature("test-model", "text", sig)
+	CacheSignature(testModelName, "text", sig)
+	CacheSignature(testModelName, "text-2", sig)
 
 	ClearSignatureCache("session-1")
 
-	if got := GetCachedSignature("test-model", "text"); got != "" {
-		t.Error("session-1 should be cleared")
-	}
-	if got := GetCachedSignature("test-model", "text"); got != sig {
-		t.Error("session-2 should still exist")
+	if got := GetCachedSignature(testModelName, "text"); got != sig {
+		t.Error("signature should remain when clearing unknown session")
 	}
 }
 
@@ -102,35 +101,37 @@ func TestClearSignatureCache_AllSessions(t *testing.T) {
 	ClearSignatureCache("")
 
 	sig := "validSig1234567890123456789012345678901234567890123456"
-	CacheSignature("test-model", "text", sig)
-	CacheSignature("test-model", "text", sig)
+	CacheSignature(testModelName, "text", sig)
+	CacheSignature(testModelName, "text-2", sig)
 
 	ClearSignatureCache("")
 
-	if got := GetCachedSignature("test-model", "text"); got != "" {
-		t.Error("session-1 should be cleared")
+	if got := GetCachedSignature(testModelName, "text"); got != "" {
+		t.Error("text should be cleared")
 	}
-	if got := GetCachedSignature("test-model", "text"); got != "" {
-		t.Error("session-2 should be cleared")
+	if got := GetCachedSignature(testModelName, "text-2"); got != "" {
+		t.Error("text-2 should be cleared")
 	}
 }
 
 func TestHasValidSignature(t *testing.T) {
 	tests := []struct {
 		name      string
+		modelName string
 		signature string
 		expected  bool
 	}{
-		{"valid long signature", "abc123validSignature1234567890123456789012345678901234567890", true},
-		{"exactly 50 chars", "12345678901234567890123456789012345678901234567890", true},
-		{"49 chars - invalid", "1234567890123456789012345678901234567890123456789", false},
-		{"empty string", "", false},
-		{"short signature", "abc", false},
+		{"valid long signature", testModelName, "abc123validSignature1234567890123456789012345678901234567890", true},
+		{"exactly 50 chars", testModelName, "12345678901234567890123456789012345678901234567890", true},
+		{"49 chars - invalid", testModelName, "1234567890123456789012345678901234567890123456789", false},
+		{"empty string", testModelName, "", false},
+		{"short signature", testModelName, "abc", false},
+		{"gemini sentinel", "gemini-3-pro-preview", "skip_thought_signature_validator", true},
 	}
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			result := HasValidSignature("claude-sonnet-4-5-thinking", tt.signature)
+			result := HasValidSignature(tt.modelName, tt.signature)
 			if result != tt.expected {
 				t.Errorf("HasValidSignature(%q) = %v, expected %v", tt.signature, result, tt.expected)
 			}
@@ -147,13 +148,13 @@ func TestCacheSignature_TextHashCollisionResistance(t *testing.T) {
 	sig1 := "signature1_1234567890123456789012345678901234567890123456"
 	sig2 := "signature2_1234567890123456789012345678901234567890123456"
 
-	CacheSignature("test-model", text1, sig1)
-	CacheSignature("test-model", text2, sig2)
+	CacheSignature(testModelName, text1, sig1)
+	CacheSignature(testModelName, text2, sig2)
 
-	if GetCachedSignature("test-model", text1) != sig1 {
+	if GetCachedSignature(testModelName, text1) != sig1 {
 		t.Error("text1 signature mismatch")
 	}
-	if GetCachedSignature("test-model", text2) != sig2 {
+	if GetCachedSignature(testModelName, text2) != sig2 {
 		t.Error("text2 signature mismatch")
 	}
 }
@@ -164,9 +165,9 @@ func TestCacheSignature_UnicodeText(t *testing.T) {
 	text := "한글 텍스트와 이모지 🎉 그리고 特殊文字"
 	sig := "unicodeSig123456789012345678901234567890123456789012345"
 
-	CacheSignature("test-model", text, sig)
+	CacheSignature(testModelName, text, sig)
 
-	if got := GetCachedSignature("test-model", text); got != sig {
+	if got := GetCachedSignature(testModelName, text); got != sig {
 		t.Errorf("Unicode text signature retrieval failed, got '%s'", got)
 	}
 }
@@ -178,10 +179,10 @@ func TestCacheSignature_Overwrite(t *testing.T) {
 	sig1 := "firstSignature12345678901234567890123456789012345678901"
 	sig2 := "secondSignature1234567890123456789012345678901234567890"
 
-	CacheSignature("test-model", text, sig1)
-	CacheSignature("test-model", text, sig2) // Overwrite
+	CacheSignature(testModelName, text, sig1)
+	CacheSignature(testModelName, text, sig2) // Overwrite
 
-	if got := GetCachedSignature("test-model", text); got != sig2 {
+	if got := GetCachedSignature(testModelName, text); got != sig2 {
 		t.Errorf("Expected overwritten signature '%s', got '%s'", sig2, got)
 	}
 }
@@ -196,10 +197,10 @@ func TestCacheSignature_ExpirationLogic(t *testing.T) {
 	text := "text"
 	sig := "validSig1234567890123456789012345678901234567890123456"
 
-	CacheSignature("test-model", text, sig)
+	CacheSignature(testModelName, text, sig)
 
 	// Fresh entry should be retrievable
-	if got := GetCachedSignature("test-model", text); got != sig {
+	if got := GetCachedSignature(testModelName, text); got != sig {
 		t.Errorf("Fresh entry should be retrievable, got '%s'", got)
 	}
 

internal/config/config.go

@@ -923,6 +923,7 @@ func SaveConfigPreserveComments(configFile string, cfg *Config) error {
 	removeLegacyGenerativeLanguageKeys(original.Content[0])
 
 	pruneMappingToGeneratedKeys(original.Content[0], generated.Content[0], "oauth-excluded-models")
+	pruneMappingToGeneratedKeys(original.Content[0], generated.Content[0], "oauth-model-alias")
 
 	// Merge generated into original in-place, preserving comments/order of existing nodes.
 	mergeMappingPreserve(original.Content[0], generated.Content[0])
@@ -1413,6 +1414,16 @@ func pruneMappingToGeneratedKeys(dstRoot, srcRoot *yaml.Node, key string) {
 	}
 	srcIdx := findMapKeyIndex(srcRoot, key)
 	if srcIdx < 0 {
+		// Keep an explicit empty mapping for oauth-model-alias when it was previously present.
+		//
+		// Rationale: LoadConfig runs MigrateOAuthModelAlias before unmarshalling. If the
+		// oauth-model-alias key is missing, migration will add the default antigravity aliases.
+		// When users delete the last channel from oauth-model-alias via the management API,
+		// we want that deletion to persist across hot reloads and restarts.
+		if key == "oauth-model-alias" {
+			dstRoot.Content[dstIdx+1] = &yaml.Node{Kind: yaml.MappingNode, Tag: "!!map"}
+			return
+		}
 		removeMapKey(dstRoot, key)
 		return
 	}

internal/logging/request_logger.go

@@ -44,10 +44,12 @@ type RequestLogger interface {
 	//   - apiRequest: The API request data
 	//   - apiResponse: The API response data
 	//   - requestID: Optional request ID for log file naming
+	//   - requestTimestamp: When the request was received
+	//   - apiResponseTimestamp: When the API response was received
 	//
 	// Returns:
 	//   - error: An error if logging fails, nil otherwise
-	LogRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, requestID string) error
+	LogRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, requestID string, requestTimestamp, apiResponseTimestamp time.Time) error
 
 	// LogStreamingRequest initiates logging for a streaming request and returns a writer for chunks.
 	//
@@ -109,6 +111,12 @@ type StreamingLogWriter interface {
 	//   - error: An error if writing fails, nil otherwise
 	WriteAPIResponse(apiResponse []byte) error
 
+	// SetFirstChunkTimestamp sets the TTFB timestamp captured when first chunk was received.
+	//
+	// Parameters:
+	//   - timestamp: The time when first response chunk was received
+	SetFirstChunkTimestamp(timestamp time.Time)
+
 	// Close finalizes the log file and cleans up resources.
 	//
 	// Returns:
@@ -180,20 +188,22 @@ func (l *FileRequestLogger) SetEnabled(enabled bool) {
 //   - apiRequest: The API request data
 //   - apiResponse: The API response data
 //   - requestID: Optional request ID for log file naming
+//   - requestTimestamp: When the request was received
+//   - apiResponseTimestamp: When the API response was received
 //
 // Returns:
 //   - error: An error if logging fails, nil otherwise
-func (l *FileRequestLogger) LogRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, requestID string) error {
-	return l.logRequest(url, method, requestHeaders, body, statusCode, responseHeaders, response, apiRequest, apiResponse, apiResponseErrors, false, requestID)
+func (l *FileRequestLogger) LogRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, requestID string, requestTimestamp, apiResponseTimestamp time.Time) error {
+	return l.logRequest(url, method, requestHeaders, body, statusCode, responseHeaders, response, apiRequest, apiResponse, apiResponseErrors, false, requestID, requestTimestamp, apiResponseTimestamp)
 }
 
 // LogRequestWithOptions logs a request with optional forced logging behavior.
 // The force flag allows writing error logs even when regular request logging is disabled.
-func (l *FileRequestLogger) LogRequestWithOptions(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, force bool, requestID string) error {
-	return l.logRequest(url, method, requestHeaders, body, statusCode, responseHeaders, response, apiRequest, apiResponse, apiResponseErrors, force, requestID)
+func (l *FileRequestLogger) LogRequestWithOptions(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, force bool, requestID string, requestTimestamp, apiResponseTimestamp time.Time) error {
+	return l.logRequest(url, method, requestHeaders, body, statusCode, responseHeaders, response, apiRequest, apiResponse, apiResponseErrors, force, requestID, requestTimestamp, apiResponseTimestamp)
 }
 
-func (l *FileRequestLogger) logRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, force bool, requestID string) error {
+func (l *FileRequestLogger) logRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage, force bool, requestID string, requestTimestamp, apiResponseTimestamp time.Time) error {
 	if !l.enabled && !force {
 		return nil
 	}
@@ -247,6 +257,8 @@ func (l *FileRequestLogger) logRequest(url, method string, requestHeaders map[st
 		responseHeaders,
 		responseToWrite,
 		decompressErr,
+		requestTimestamp,
+		apiResponseTimestamp,
 	)
 	if errClose := logFile.Close(); errClose != nil {
 		log.WithError(errClose).Warn("failed to close request log file")
@@ -499,17 +511,22 @@ func (l *FileRequestLogger) writeNonStreamingLog(
 	responseHeaders map[string][]string,
 	response []byte,
 	decompressErr error,
+	requestTimestamp time.Time,
+	apiResponseTimestamp time.Time,
 ) error {
-	if errWrite := writeRequestInfoWithBody(w, url, method, requestHeaders, requestBody, requestBodyPath, time.Now()); errWrite != nil {
+	if requestTimestamp.IsZero() {
+		requestTimestamp = time.Now()
+	}
+	if errWrite := writeRequestInfoWithBody(w, url, method, requestHeaders, requestBody, requestBodyPath, requestTimestamp); errWrite != nil {
 		return errWrite
 	}
-	if errWrite := writeAPISection(w, "=== API REQUEST ===\n", "=== API REQUEST", apiRequest); errWrite != nil {
+	if errWrite := writeAPISection(w, "=== API REQUEST ===\n", "=== API REQUEST", apiRequest, time.Time{}); errWrite != nil {
 		return errWrite
 	}
 	if errWrite := writeAPIErrorResponses(w, apiResponseErrors); errWrite != nil {
 		return errWrite
 	}
-	if errWrite := writeAPISection(w, "=== API RESPONSE ===\n", "=== API RESPONSE", apiResponse); errWrite != nil {
+	if errWrite := writeAPISection(w, "=== API RESPONSE ===\n", "=== API RESPONSE", apiResponse, apiResponseTimestamp); errWrite != nil {
 		return errWrite
 	}
 	return writeResponseSection(w, statusCode, true, responseHeaders, bytes.NewReader(response), decompressErr, true)
@@ -583,7 +600,7 @@ func writeRequestInfoWithBody(
 	return nil
 }
 
-func writeAPISection(w io.Writer, sectionHeader string, sectionPrefix string, payload []byte) error {
+func writeAPISection(w io.Writer, sectionHeader string, sectionPrefix string, payload []byte, timestamp time.Time) error {
 	if len(payload) == 0 {
 		return nil
 	}
@@ -601,6 +618,11 @@ func writeAPISection(w io.Writer, sectionHeader string, sectionPrefix string, pa
 		if _, errWrite := io.WriteString(w, sectionHeader); errWrite != nil {
 			return errWrite
 		}
+		if !timestamp.IsZero() {
+			if _, errWrite := io.WriteString(w, fmt.Sprintf("Timestamp: %s\n", timestamp.Format(time.RFC3339Nano))); errWrite != nil {
+				return errWrite
+			}
+		}
 		if _, errWrite := w.Write(payload); errWrite != nil {
 			return errWrite
 		}
@@ -974,6 +996,9 @@ type FileStreamingLogWriter struct {
 
 	// apiResponse stores the upstream API response data.
 	apiResponse []byte
+
+	// apiResponseTimestamp captures when the API response was received.
+	apiResponseTimestamp time.Time
 }
 
 // WriteChunkAsync writes a response chunk asynchronously (non-blocking).
@@ -1053,6 +1078,12 @@ func (w *FileStreamingLogWriter) WriteAPIResponse(apiResponse []byte) error {
 	return nil
 }
 
+func (w *FileStreamingLogWriter) SetFirstChunkTimestamp(timestamp time.Time) {
+	if !timestamp.IsZero() {
+		w.apiResponseTimestamp = timestamp
+	}
+}
+
 // Close finalizes the log file and cleans up resources.
 // It writes all buffered data to the file in the correct order:
 // API REQUEST -> API RESPONSE -> RESPONSE (status, headers, body chunks)
@@ -1140,10 +1171,10 @@ func (w *FileStreamingLogWriter) writeFinalLog(logFile *os.File) error {
 	if errWrite := writeRequestInfoWithBody(logFile, w.url, w.method, w.requestHeaders, nil, w.requestBodyPath, w.timestamp); errWrite != nil {
 		return errWrite
 	}
-	if errWrite := writeAPISection(logFile, "=== API REQUEST ===\n", "=== API REQUEST", w.apiRequest); errWrite != nil {
+	if errWrite := writeAPISection(logFile, "=== API REQUEST ===\n", "=== API REQUEST", w.apiRequest, time.Time{}); errWrite != nil {
 		return errWrite
 	}
-	if errWrite := writeAPISection(logFile, "=== API RESPONSE ===\n", "=== API RESPONSE", w.apiResponse); errWrite != nil {
+	if errWrite := writeAPISection(logFile, "=== API RESPONSE ===\n", "=== API RESPONSE", w.apiResponse, w.apiResponseTimestamp); errWrite != nil {
 		return errWrite
 	}
 
@@ -1220,6 +1251,8 @@ func (w *NoOpStreamingLogWriter) WriteAPIResponse(_ []byte) error {
 	return nil
 }
 
+func (w *NoOpStreamingLogWriter) SetFirstChunkTimestamp(_ time.Time) {}
+
 // Close is a no-op implementation that does nothing and always returns nil.
 //
 // Returns:

internal/registry/model_definitions.go

@@ -1,848 +1,69 @@
-// Package registry provides model definitions for various AI service providers.
-// This file contains static model definitions that can be used by clients
-// when registering their supported models.
+// Package registry provides model definitions and lookup helpers for various AI providers.
+// Static model metadata is stored in model_definitions_static_data.go.
 package registry
 
-// GetClaudeModels returns the standard Claude model definitions
-func GetClaudeModels() []*ModelInfo {
-	return []*ModelInfo{
+import (
+	"sort"
+	"strings"
+)
 
-		{
-			ID:                  "claude-haiku-4-5-20251001",
-			Object:              "model",
-			Created:             1759276800, // 2025-10-01
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4.5 Haiku",
-			ContextLength:       200000,
-			MaxCompletionTokens: 64000,
-			// Thinking: not supported for Haiku models
-		},
-		{
-			ID:                  "claude-sonnet-4-5-20250929",
-			Object:              "model",
-			Created:             1759104000, // 2025-09-29
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4.5 Sonnet",
-			ContextLength:       200000,
-			MaxCompletionTokens: 64000,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
-		},
-		{
-			ID:                  "claude-opus-4-5-20251101",
-			Object:              "model",
-			Created:             1761955200, // 2025-11-01
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4.5 Opus",
-			Description:         "Premium model combining maximum intelligence with practical performance",
-			ContextLength:       200000,
-			MaxCompletionTokens: 64000,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
-		},
-		{
-			ID:                  "claude-opus-4-1-20250805",
-			Object:              "model",
-			Created:             1722945600, // 2025-08-05
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4.1 Opus",
-			ContextLength:       200000,
-			MaxCompletionTokens: 32000,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
-		},
-		{
-			ID:                  "claude-opus-4-20250514",
-			Object:              "model",
-			Created:             1715644800, // 2025-05-14
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4 Opus",
-			ContextLength:       200000,
-			MaxCompletionTokens: 32000,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
-		},
-		{
-			ID:                  "claude-sonnet-4-20250514",
-			Object:              "model",
-			Created:             1715644800, // 2025-05-14
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 4 Sonnet",
-			ContextLength:       200000,
-			MaxCompletionTokens: 64000,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
-		},
-		{
-			ID:                  "claude-3-7-sonnet-20250219",
-			Object:              "model",
-			Created:             1708300800, // 2025-02-19
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 3.7 Sonnet",
-			ContextLength:       128000,
-			MaxCompletionTokens: 8192,
-			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
-		},
-		{
-			ID:                  "claude-3-5-haiku-20241022",
-			Object:              "model",
-			Created:             1729555200, // 2024-10-22
-			OwnedBy:             "anthropic",
-			Type:                "claude",
-			DisplayName:         "Claude 3.5 Haiku",
-			ContextLength:       128000,
-			MaxCompletionTokens: 8192,
-			// Thinking: not supported for Haiku models
-		},
-	}
-}
-
-// GetGeminiModels returns the standard Gemini model definitions
-func GetGeminiModels() []*ModelInfo {
-	return []*ModelInfo{
-		{
-			ID:                         "gemini-2.5-pro",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-pro",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Pro",
-			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash",
-			Version:                    "001",
-			DisplayName:                "Gemini 2.5 Flash",
-			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash-lite",
-			Object:                     "model",
-			Created:                    1753142400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash-lite",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Flash Lite",
-			Description:                "Our smallest and most cost effective model, built for at scale usage.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-3-pro-preview",
-			Object:                     "model",
-			Created:                    1737158400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-pro-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Pro Preview",
-			Description:                "Gemini 3 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",
-			Created:                    1765929600,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-flash-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Flash Preview",
-			Description:                "Gemini 3 Flash Preview",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			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-pro-image-preview",
-			Object:                     "model",
-			Created:                    1737158400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-pro-image-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Pro Image Preview",
-			Description:                "Gemini 3 Pro Image 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"}},
-		},
-	}
-}
-
-func GetGeminiVertexModels() []*ModelInfo {
-	return []*ModelInfo{
-		{
-			ID:                         "gemini-2.5-pro",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-pro",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Pro",
-			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash",
-			Version:                    "001",
-			DisplayName:                "Gemini 2.5 Flash",
-			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash-lite",
-			Object:                     "model",
-			Created:                    1753142400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash-lite",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Flash Lite",
-			Description:                "Our smallest and most cost effective model, built for at scale usage.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-3-pro-preview",
-			Object:                     "model",
-			Created:                    1737158400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-pro-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Pro Preview",
-			Description:                "Gemini 3 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",
-			Created:                    1765929600,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-flash-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Flash Preview",
-			Description:                "Our most intelligent model built for speed, combining frontier intelligence with superior search and grounding.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			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-pro-image-preview",
-			Object:                     "model",
-			Created:                    1737158400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-pro-image-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Pro Image Preview",
-			Description:                "Gemini 3 Pro Image 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"}},
-		},
-		// Imagen image generation models - use :predict action
-		{
-			ID:                         "imagen-4.0-generate-001",
-			Object:                     "model",
-			Created:                    1750000000,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/imagen-4.0-generate-001",
-			Version:                    "4.0",
-			DisplayName:                "Imagen 4.0 Generate",
-			Description:                "Imagen 4.0 image generation model",
-			SupportedGenerationMethods: []string{"predict"},
-		},
-		{
-			ID:                         "imagen-4.0-ultra-generate-001",
-			Object:                     "model",
-			Created:                    1750000000,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/imagen-4.0-ultra-generate-001",
-			Version:                    "4.0",
-			DisplayName:                "Imagen 4.0 Ultra Generate",
-			Description:                "Imagen 4.0 Ultra high-quality image generation model",
-			SupportedGenerationMethods: []string{"predict"},
-		},
-		{
-			ID:                         "imagen-3.0-generate-002",
-			Object:                     "model",
-			Created:                    1740000000,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/imagen-3.0-generate-002",
-			Version:                    "3.0",
-			DisplayName:                "Imagen 3.0 Generate",
-			Description:                "Imagen 3.0 image generation model",
-			SupportedGenerationMethods: []string{"predict"},
-		},
-		{
-			ID:                         "imagen-3.0-fast-generate-001",
-			Object:                     "model",
-			Created:                    1740000000,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/imagen-3.0-fast-generate-001",
-			Version:                    "3.0",
-			DisplayName:                "Imagen 3.0 Fast Generate",
-			Description:                "Imagen 3.0 fast image generation model",
-			SupportedGenerationMethods: []string{"predict"},
-		},
-		{
-			ID:                         "imagen-4.0-fast-generate-001",
-			Object:                     "model",
-			Created:                    1750000000,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/imagen-4.0-fast-generate-001",
-			Version:                    "4.0",
-			DisplayName:                "Imagen 4.0 Fast Generate",
-			Description:                "Imagen 4.0 fast image generation model",
-			SupportedGenerationMethods: []string{"predict"},
-		},
-	}
-}
-
-// GetGeminiCLIModels returns the standard Gemini model definitions
-func GetGeminiCLIModels() []*ModelInfo {
-	return []*ModelInfo{
-		{
-			ID:                         "gemini-2.5-pro",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-pro",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Pro",
-			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash",
-			Version:                    "001",
-			DisplayName:                "Gemini 2.5 Flash",
-			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash-lite",
-			Object:                     "model",
-			Created:                    1753142400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash-lite",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Flash Lite",
-			Description:                "Our smallest and most cost effective model, built for at scale usage.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-3-pro-preview",
-			Object:                     "model",
-			Created:                    1737158400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-pro-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Pro Preview",
-			Description:                "Our most intelligent model with SOTA reasoning and multimodal understanding, and powerful agentic and vibe coding capabilities",
-			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",
-			Created:                    1765929600,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-flash-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Flash Preview",
-			Description:                "Our most intelligent model built for speed, combining frontier intelligence with superior search and grounding.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"minimal", "low", "medium", "high"}},
-		},
-	}
-}
-
-// GetAIStudioModels returns the Gemini model definitions for AI Studio integrations
-func GetAIStudioModels() []*ModelInfo {
-	return []*ModelInfo{
-		{
-			ID:                         "gemini-2.5-pro",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-pro",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Pro",
-			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash",
-			Version:                    "001",
-			DisplayName:                "Gemini 2.5 Flash",
-			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash-lite",
-			Object:                     "model",
-			Created:                    1753142400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash-lite",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Flash Lite",
-			Description:                "Our smallest and most cost effective model, built for at scale usage.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-3-pro-preview",
-			Object:                     "model",
-			Created:                    1737158400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-pro-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Pro Preview",
-			Description:                "Gemini 3 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",
-			Created:                    1765929600,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-3-flash-preview",
-			Version:                    "3.0",
-			DisplayName:                "Gemini 3 Flash Preview",
-			Description:                "Our most intelligent model built for speed, combining frontier intelligence with superior search and grounding.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-pro-latest",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-pro-latest",
-			Version:                    "2.5",
-			DisplayName:                "Gemini Pro Latest",
-			Description:                "Latest release of Gemini Pro",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-flash-latest",
-			Object:                     "model",
-			Created:                    1750118400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-flash-latest",
-			Version:                    "2.5",
-			DisplayName:                "Gemini Flash Latest",
-			Description:                "Latest release of Gemini Flash",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-flash-lite-latest",
-			Object:                     "model",
-			Created:                    1753142400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-flash-lite-latest",
-			Version:                    "2.5",
-			DisplayName:                "Gemini Flash-Lite Latest",
-			Description:                "Latest release of Gemini Flash-Lite",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           65536,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			Thinking:                   &ThinkingSupport{Min: 512, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
-		},
-		{
-			ID:                         "gemini-2.5-flash-image-preview",
-			Object:                     "model",
-			Created:                    1756166400,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash-image-preview",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Flash Image Preview",
-			Description:                "State-of-the-art image generation and editing model.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           8192,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			// image models don't support thinkingConfig; leave Thinking nil
-		},
-		{
-			ID:                         "gemini-2.5-flash-image",
-			Object:                     "model",
-			Created:                    1759363200,
-			OwnedBy:                    "google",
-			Type:                       "gemini",
-			Name:                       "models/gemini-2.5-flash-image",
-			Version:                    "2.5",
-			DisplayName:                "Gemini 2.5 Flash Image",
-			Description:                "State-of-the-art image generation and editing model.",
-			InputTokenLimit:            1048576,
-			OutputTokenLimit:           8192,
-			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
-			// image models don't support thinkingConfig; leave Thinking nil
-		},
-	}
-}
-
-// GetOpenAIModels returns the standard OpenAI model definitions
-func GetOpenAIModels() []*ModelInfo {
-	return []*ModelInfo{
-		{
-			ID:                  "gpt-5",
-			Object:              "model",
-			Created:             1754524800,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5-2025-08-07",
-			DisplayName:         "GPT 5",
-			Description:         "Stable version of GPT 5, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-			Thinking:            &ThinkingSupport{Levels: []string{"minimal", "low", "medium", "high"}},
-		},
-		{
-			ID:                  "gpt-5-codex",
-			Object:              "model",
-			Created:             1757894400,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5-2025-09-15",
-			DisplayName:         "GPT 5 Codex",
-			Description:         "Stable version of GPT 5 Codex, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high"}},
-		},
-		{
-			ID:                  "gpt-5-codex-mini",
-			Object:              "model",
-			Created:             1762473600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5-2025-11-07",
-			DisplayName:         "GPT 5 Codex Mini",
-			Description:         "Stable version of GPT 5 Codex Mini: cheaper, faster, but less capable version of GPT 5 Codex.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high"}},
-		},
-		{
-			ID:                  "gpt-5.1",
-			Object:              "model",
-			Created:             1762905600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-2025-11-12",
-			DisplayName:         "GPT 5",
-			Description:         "Stable version of GPT 5, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-			Thinking:            &ThinkingSupport{Levels: []string{"none", "low", "medium", "high"}},
-		},
-		{
-			ID:                  "gpt-5.1-codex",
-			Object:              "model",
-			Created:             1762905600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-2025-11-12",
-			DisplayName:         "GPT 5.1 Codex",
-			Description:         "Stable version of GPT 5.1 Codex, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high"}},
-		},
-		{
-			ID:                  "gpt-5.1-codex-mini",
-			Object:              "model",
-			Created:             1762905600,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-2025-11-12",
-			DisplayName:         "GPT 5.1 Codex Mini",
-			Description:         "Stable version of GPT 5.1 Codex Mini: cheaper, faster, but less capable version of GPT 5.1 Codex.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high"}},
-		},
-		{
-			ID:                  "gpt-5.1-codex-max",
-			Object:              "model",
-			Created:             1763424000,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.1-max",
-			DisplayName:         "GPT 5.1 Codex Max",
-			Description:         "Stable version of GPT 5.1 Codex Max",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high", "xhigh"}},
-		},
-		{
-			ID:                  "gpt-5.2",
-			Object:              "model",
-			Created:             1765440000,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.2",
-			DisplayName:         "GPT 5.2",
-			Description:         "Stable version of GPT 5.2",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-			Thinking:            &ThinkingSupport{Levels: []string{"none", "low", "medium", "high", "xhigh"}},
-		},
-		{
-			ID:                  "gpt-5.2-codex",
-			Object:              "model",
-			Created:             1765440000,
-			OwnedBy:             "openai",
-			Type:                "openai",
-			Version:             "gpt-5.2",
-			DisplayName:         "GPT 5.2 Codex",
-			Description:         "Stable version of GPT 5.2 Codex, The best model for coding and agentic tasks across domains.",
-			ContextLength:       400000,
-			MaxCompletionTokens: 128000,
-			SupportedParameters: []string{"tools"},
-			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high", "xhigh"}},
-		},
-	}
-}
-
-// GetQwenModels returns the standard Qwen model definitions
-func GetQwenModels() []*ModelInfo {
-	return []*ModelInfo{
-		{
-			ID:                  "qwen3-coder-plus",
-			Object:              "model",
-			Created:             1753228800,
-			OwnedBy:             "qwen",
-			Type:                "qwen",
-			Version:             "3.0",
-			DisplayName:         "Qwen3 Coder Plus",
-			Description:         "Advanced code generation and understanding model",
-			ContextLength:       32768,
-			MaxCompletionTokens: 8192,
-			SupportedParameters: []string{"temperature", "top_p", "max_tokens", "stream", "stop"},
-		},
-		{
-			ID:                  "qwen3-coder-flash",
-			Object:              "model",
-			Created:             1753228800,
-			OwnedBy:             "qwen",
-			Type:                "qwen",
-			Version:             "3.0",
-			DisplayName:         "Qwen3 Coder Flash",
-			Description:         "Fast code generation model",
-			ContextLength:       8192,
-			MaxCompletionTokens: 2048,
-			SupportedParameters: []string{"temperature", "top_p", "max_tokens", "stream", "stop"},
-		},
-		{
-			ID:                  "vision-model",
-			Object:              "model",
-			Created:             1758672000,
-			OwnedBy:             "qwen",
-			Type:                "qwen",
-			Version:             "3.0",
-			DisplayName:         "Qwen3 Vision Model",
-			Description:         "Vision model model",
-			ContextLength:       32768,
-			MaxCompletionTokens: 2048,
-			SupportedParameters: []string{"temperature", "top_p", "max_tokens", "stream", "stop"},
-		},
-	}
-}
-
-// iFlowThinkingSupport is a shared ThinkingSupport configuration for iFlow models
-// that support thinking mode via chat_template_kwargs.enable_thinking (boolean toggle).
-// Uses level-based configuration so standard normalization flows apply before conversion.
-var iFlowThinkingSupport = &ThinkingSupport{
-	Levels: []string{"none", "auto", "minimal", "low", "medium", "high", "xhigh"},
-}
-
-// GetIFlowModels returns supported models for iFlow OAuth accounts.
-func GetIFlowModels() []*ModelInfo {
-	entries := []struct {
-		ID          string
-		DisplayName string
-		Description string
-		Created     int64
-		Thinking    *ThinkingSupport
-	}{
-		{ID: "tstars2.0", DisplayName: "TStars-2.0", Description: "iFlow TStars-2.0 multimodal assistant", Created: 1746489600},
-		{ID: "qwen3-coder-plus", DisplayName: "Qwen3-Coder-Plus", Description: "Qwen3 Coder Plus code generation", Created: 1753228800},
-		{ID: "qwen3-max", DisplayName: "Qwen3-Max", Description: "Qwen3 flagship model", Created: 1758672000},
-		{ID: "qwen3-vl-plus", DisplayName: "Qwen3-VL-Plus", Description: "Qwen3 multimodal vision-language", Created: 1758672000},
-		{ID: "qwen3-max-preview", DisplayName: "Qwen3-Max-Preview", Description: "Qwen3 Max preview build", Created: 1757030400},
-		{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: "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},
-		{ID: "deepseek-v3.2-reasoner", DisplayName: "DeepSeek-V3.2", Description: "DeepSeek V3.2 Reasoner", Created: 1764576000},
-		{ID: "deepseek-v3.2", DisplayName: "DeepSeek-V3.2-Exp", Description: "DeepSeek V3.2 experimental", Created: 1759104000},
-		{ID: "deepseek-v3.1", DisplayName: "DeepSeek-V3.1-Terminus", Description: "DeepSeek V3.1 Terminus", Created: 1756339200},
-		{ID: "deepseek-r1", DisplayName: "DeepSeek-R1", Description: "DeepSeek reasoning model R1", Created: 1737331200},
-		{ID: "deepseek-v3", DisplayName: "DeepSeek-V3-671B", Description: "DeepSeek V3 671B", Created: 1734307200},
-		{ID: "qwen3-32b", DisplayName: "Qwen3-32B", Description: "Qwen3 32B", Created: 1747094400},
-		{ID: "qwen3-235b-a22b-thinking-2507", DisplayName: "Qwen3-235B-A22B-Thinking", Description: "Qwen3 235B A22B Thinking (2507)", Created: 1753401600},
-		{ID: "qwen3-235b-a22b-instruct", DisplayName: "Qwen3-235B-A22B-Instruct", Description: "Qwen3 235B A22B Instruct", Created: 1753401600},
-		{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: "iflow-rome-30ba3b", DisplayName: "iFlow-ROME", Description: "iFlow Rome 30BA3B model", Created: 1736899200},
-	}
-	models := make([]*ModelInfo, 0, len(entries))
-	for _, entry := range entries {
-		models = append(models, &ModelInfo{
-			ID:          entry.ID,
-			Object:      "model",
-			Created:     entry.Created,
-			OwnedBy:     "iflow",
-			Type:        "iflow",
-			DisplayName: entry.DisplayName,
-			Description: entry.Description,
-			Thinking:    entry.Thinking,
+// GetStaticModelDefinitionsByChannel returns static model definitions for a given channel/provider.
+// It returns nil when the channel is unknown.
+//
+// Supported channels:
+//   - claude
+//   - gemini
+//   - vertex
+//   - gemini-cli
+//   - aistudio
+//   - codex
+//   - qwen
+//   - iflow
+//   - antigravity (returns static overrides only)
+func GetStaticModelDefinitionsByChannel(channel string) []*ModelInfo {
+	key := strings.ToLower(strings.TrimSpace(channel))
+	switch key {
+	case "claude":
+		return GetClaudeModels()
+	case "gemini":
+		return GetGeminiModels()
+	case "vertex":
+		return GetGeminiVertexModels()
+	case "gemini-cli":
+		return GetGeminiCLIModels()
+	case "aistudio":
+		return GetAIStudioModels()
+	case "codex":
+		return GetOpenAIModels()
+	case "qwen":
+		return GetQwenModels()
+	case "iflow":
+		return GetIFlowModels()
+	case "antigravity":
+		cfg := GetAntigravityModelConfig()
+		if len(cfg) == 0 {
+			return nil
+		}
+		models := make([]*ModelInfo, 0, len(cfg))
+		for modelID, entry := range cfg {
+			if modelID == "" || entry == nil {
+				continue
+			}
+			models = append(models, &ModelInfo{
+				ID:                  modelID,
+				Object:              "model",
+				OwnedBy:             "antigravity",
+				Type:                "antigravity",
+				Thinking:            entry.Thinking,
+				MaxCompletionTokens: entry.MaxCompletionTokens,
+			})
+		}
+		sort.Slice(models, func(i, j int) bool {
+			return strings.ToLower(models[i].ID) < strings.ToLower(models[j].ID)
 		})
-	}
-	return models
-}
-
-// AntigravityModelConfig captures static antigravity model overrides, including
-// Thinking budget limits and provider max completion tokens.
-type AntigravityModelConfig struct {
-	Thinking            *ThinkingSupport
-	MaxCompletionTokens int
-}
-
-// GetAntigravityModelConfig returns static configuration for antigravity models.
-// Keys use upstream model names returned by the Antigravity models endpoint.
-func GetAntigravityModelConfig() map[string]*AntigravityModelConfig {
-	return map[string]*AntigravityModelConfig{
-		"gemini-2.5-flash":           {Thinking: &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true}},
-		"gemini-2.5-flash-lite":      {Thinking: &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true}},
-		"rev19-uic3-1p":              {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, 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-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-sonnet-4-5":          {MaxCompletionTokens: 64000},
-		"gpt-oss-120b-medium":        {},
-		"tab_flash_lite_preview":     {},
+		return models
+	default:
+		return nil
 	}
 }
 

internal/registry/model_definitions_static_data.go

@@ -0,0 +1,846 @@
+// Package registry provides model definitions for various AI service providers.
+// This file stores the static model metadata catalog.
+package registry
+
+// GetClaudeModels returns the standard Claude model definitions
+func GetClaudeModels() []*ModelInfo {
+	return []*ModelInfo{
+
+		{
+			ID:                  "claude-haiku-4-5-20251001",
+			Object:              "model",
+			Created:             1759276800, // 2025-10-01
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.5 Haiku",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			// Thinking: not supported for Haiku models
+		},
+		{
+			ID:                  "claude-sonnet-4-5-20250929",
+			Object:              "model",
+			Created:             1759104000, // 2025-09-29
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.5 Sonnet",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
+		},
+		{
+			ID:                  "claude-opus-4-5-20251101",
+			Object:              "model",
+			Created:             1761955200, // 2025-11-01
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.5 Opus",
+			Description:         "Premium model combining maximum intelligence with practical performance",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: true, DynamicAllowed: false},
+		},
+		{
+			ID:                  "claude-opus-4-1-20250805",
+			Object:              "model",
+			Created:             1722945600, // 2025-08-05
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4.1 Opus",
+			ContextLength:       200000,
+			MaxCompletionTokens: 32000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
+		},
+		{
+			ID:                  "claude-opus-4-20250514",
+			Object:              "model",
+			Created:             1715644800, // 2025-05-14
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4 Opus",
+			ContextLength:       200000,
+			MaxCompletionTokens: 32000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
+		},
+		{
+			ID:                  "claude-sonnet-4-20250514",
+			Object:              "model",
+			Created:             1715644800, // 2025-05-14
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 4 Sonnet",
+			ContextLength:       200000,
+			MaxCompletionTokens: 64000,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
+		},
+		{
+			ID:                  "claude-3-7-sonnet-20250219",
+			Object:              "model",
+			Created:             1708300800, // 2025-02-19
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 3.7 Sonnet",
+			ContextLength:       128000,
+			MaxCompletionTokens: 8192,
+			Thinking:            &ThinkingSupport{Min: 1024, Max: 128000, ZeroAllowed: false, DynamicAllowed: false},
+		},
+		{
+			ID:                  "claude-3-5-haiku-20241022",
+			Object:              "model",
+			Created:             1729555200, // 2024-10-22
+			OwnedBy:             "anthropic",
+			Type:                "claude",
+			DisplayName:         "Claude 3.5 Haiku",
+			ContextLength:       128000,
+			MaxCompletionTokens: 8192,
+			// Thinking: not supported for Haiku models
+		},
+	}
+}
+
+// GetGeminiModels returns the standard Gemini model definitions
+func GetGeminiModels() []*ModelInfo {
+	return []*ModelInfo{
+		{
+			ID:                         "gemini-2.5-pro",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-pro",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Pro",
+			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash",
+			Version:                    "001",
+			DisplayName:                "Gemini 2.5 Flash",
+			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash-lite",
+			Object:                     "model",
+			Created:                    1753142400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash-lite",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Flash Lite",
+			Description:                "Our smallest and most cost effective model, built for at scale usage.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-3-pro-preview",
+			Object:                     "model",
+			Created:                    1737158400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-pro-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Pro Preview",
+			Description:                "Gemini 3 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",
+			Created:                    1765929600,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-flash-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Flash Preview",
+			Description:                "Gemini 3 Flash Preview",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			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-pro-image-preview",
+			Object:                     "model",
+			Created:                    1737158400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-pro-image-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Pro Image Preview",
+			Description:                "Gemini 3 Pro Image 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"}},
+		},
+	}
+}
+
+func GetGeminiVertexModels() []*ModelInfo {
+	return []*ModelInfo{
+		{
+			ID:                         "gemini-2.5-pro",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-pro",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Pro",
+			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash",
+			Version:                    "001",
+			DisplayName:                "Gemini 2.5 Flash",
+			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash-lite",
+			Object:                     "model",
+			Created:                    1753142400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash-lite",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Flash Lite",
+			Description:                "Our smallest and most cost effective model, built for at scale usage.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-3-pro-preview",
+			Object:                     "model",
+			Created:                    1737158400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-pro-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Pro Preview",
+			Description:                "Gemini 3 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",
+			Created:                    1765929600,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-flash-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Flash Preview",
+			Description:                "Our most intelligent model built for speed, combining frontier intelligence with superior search and grounding.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			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-pro-image-preview",
+			Object:                     "model",
+			Created:                    1737158400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-pro-image-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Pro Image Preview",
+			Description:                "Gemini 3 Pro Image 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"}},
+		},
+		// Imagen image generation models - use :predict action
+		{
+			ID:                         "imagen-4.0-generate-001",
+			Object:                     "model",
+			Created:                    1750000000,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/imagen-4.0-generate-001",
+			Version:                    "4.0",
+			DisplayName:                "Imagen 4.0 Generate",
+			Description:                "Imagen 4.0 image generation model",
+			SupportedGenerationMethods: []string{"predict"},
+		},
+		{
+			ID:                         "imagen-4.0-ultra-generate-001",
+			Object:                     "model",
+			Created:                    1750000000,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/imagen-4.0-ultra-generate-001",
+			Version:                    "4.0",
+			DisplayName:                "Imagen 4.0 Ultra Generate",
+			Description:                "Imagen 4.0 Ultra high-quality image generation model",
+			SupportedGenerationMethods: []string{"predict"},
+		},
+		{
+			ID:                         "imagen-3.0-generate-002",
+			Object:                     "model",
+			Created:                    1740000000,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/imagen-3.0-generate-002",
+			Version:                    "3.0",
+			DisplayName:                "Imagen 3.0 Generate",
+			Description:                "Imagen 3.0 image generation model",
+			SupportedGenerationMethods: []string{"predict"},
+		},
+		{
+			ID:                         "imagen-3.0-fast-generate-001",
+			Object:                     "model",
+			Created:                    1740000000,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/imagen-3.0-fast-generate-001",
+			Version:                    "3.0",
+			DisplayName:                "Imagen 3.0 Fast Generate",
+			Description:                "Imagen 3.0 fast image generation model",
+			SupportedGenerationMethods: []string{"predict"},
+		},
+		{
+			ID:                         "imagen-4.0-fast-generate-001",
+			Object:                     "model",
+			Created:                    1750000000,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/imagen-4.0-fast-generate-001",
+			Version:                    "4.0",
+			DisplayName:                "Imagen 4.0 Fast Generate",
+			Description:                "Imagen 4.0 fast image generation model",
+			SupportedGenerationMethods: []string{"predict"},
+		},
+	}
+}
+
+// GetGeminiCLIModels returns the standard Gemini model definitions
+func GetGeminiCLIModels() []*ModelInfo {
+	return []*ModelInfo{
+		{
+			ID:                         "gemini-2.5-pro",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-pro",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Pro",
+			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash",
+			Version:                    "001",
+			DisplayName:                "Gemini 2.5 Flash",
+			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash-lite",
+			Object:                     "model",
+			Created:                    1753142400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash-lite",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Flash Lite",
+			Description:                "Our smallest and most cost effective model, built for at scale usage.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-3-pro-preview",
+			Object:                     "model",
+			Created:                    1737158400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-pro-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Pro Preview",
+			Description:                "Our most intelligent model with SOTA reasoning and multimodal understanding, and powerful agentic and vibe coding capabilities",
+			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",
+			Created:                    1765929600,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-flash-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Flash Preview",
+			Description:                "Our most intelligent model built for speed, combining frontier intelligence with superior search and grounding.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true, Levels: []string{"minimal", "low", "medium", "high"}},
+		},
+	}
+}
+
+// GetAIStudioModels returns the Gemini model definitions for AI Studio integrations
+func GetAIStudioModels() []*ModelInfo {
+	return []*ModelInfo{
+		{
+			ID:                         "gemini-2.5-pro",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-pro",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Pro",
+			Description:                "Stable release (June 17th, 2025) of Gemini 2.5 Pro",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash",
+			Version:                    "001",
+			DisplayName:                "Gemini 2.5 Flash",
+			Description:                "Stable version of Gemini 2.5 Flash, our mid-size multimodal model that supports up to 1 million tokens, released in June of 2025.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-2.5-flash-lite",
+			Object:                     "model",
+			Created:                    1753142400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash-lite",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Flash Lite",
+			Description:                "Our smallest and most cost effective model, built for at scale usage.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-3-pro-preview",
+			Object:                     "model",
+			Created:                    1737158400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-pro-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Pro Preview",
+			Description:                "Gemini 3 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",
+			Created:                    1765929600,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-3-flash-preview",
+			Version:                    "3.0",
+			DisplayName:                "Gemini 3 Flash Preview",
+			Description:                "Our most intelligent model built for speed, combining frontier intelligence with superior search and grounding.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-pro-latest",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-pro-latest",
+			Version:                    "2.5",
+			DisplayName:                "Gemini Pro Latest",
+			Description:                "Latest release of Gemini Pro",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-flash-latest",
+			Object:                     "model",
+			Created:                    1750118400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-flash-latest",
+			Version:                    "2.5",
+			DisplayName:                "Gemini Flash Latest",
+			Description:                "Latest release of Gemini Flash",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		{
+			ID:                         "gemini-flash-lite-latest",
+			Object:                     "model",
+			Created:                    1753142400,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-flash-lite-latest",
+			Version:                    "2.5",
+			DisplayName:                "Gemini Flash-Lite Latest",
+			Description:                "Latest release of Gemini Flash-Lite",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           65536,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			Thinking:                   &ThinkingSupport{Min: 512, Max: 24576, ZeroAllowed: true, DynamicAllowed: true},
+		},
+		// {
+		// 	ID:                         "gemini-2.5-flash-image-preview",
+		// 	Object:                     "model",
+		// 	Created:                    1756166400,
+		// 	OwnedBy:                    "google",
+		// 	Type:                       "gemini",
+		// 	Name:                       "models/gemini-2.5-flash-image-preview",
+		// 	Version:                    "2.5",
+		// 	DisplayName:                "Gemini 2.5 Flash Image Preview",
+		// 	Description:                "State-of-the-art image generation and editing model.",
+		// 	InputTokenLimit:            1048576,
+		// 	OutputTokenLimit:           8192,
+		// 	SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+		// 	// image models don't support thinkingConfig; leave Thinking nil
+		// },
+		{
+			ID:                         "gemini-2.5-flash-image",
+			Object:                     "model",
+			Created:                    1759363200,
+			OwnedBy:                    "google",
+			Type:                       "gemini",
+			Name:                       "models/gemini-2.5-flash-image",
+			Version:                    "2.5",
+			DisplayName:                "Gemini 2.5 Flash Image",
+			Description:                "State-of-the-art image generation and editing model.",
+			InputTokenLimit:            1048576,
+			OutputTokenLimit:           8192,
+			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
+			// image models don't support thinkingConfig; leave Thinking nil
+		},
+	}
+}
+
+// GetOpenAIModels returns the standard OpenAI model definitions
+func GetOpenAIModels() []*ModelInfo {
+	return []*ModelInfo{
+		{
+			ID:                  "gpt-5",
+			Object:              "model",
+			Created:             1754524800,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5-2025-08-07",
+			DisplayName:         "GPT 5",
+			Description:         "Stable version of GPT 5, The best model for coding and agentic tasks across domains.",
+			ContextLength:       400000,
+			MaxCompletionTokens: 128000,
+			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"minimal", "low", "medium", "high"}},
+		},
+		{
+			ID:                  "gpt-5-codex",
+			Object:              "model",
+			Created:             1757894400,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5-2025-09-15",
+			DisplayName:         "GPT 5 Codex",
+			Description:         "Stable version of GPT 5 Codex, The best model for coding and agentic tasks across domains.",
+			ContextLength:       400000,
+			MaxCompletionTokens: 128000,
+			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high"}},
+		},
+		{
+			ID:                  "gpt-5-codex-mini",
+			Object:              "model",
+			Created:             1762473600,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5-2025-11-07",
+			DisplayName:         "GPT 5 Codex Mini",
+			Description:         "Stable version of GPT 5 Codex Mini: cheaper, faster, but less capable version of GPT 5 Codex.",
+			ContextLength:       400000,
+			MaxCompletionTokens: 128000,
+			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high"}},
+		},
+		{
+			ID:                  "gpt-5.1",
+			Object:              "model",
+			Created:             1762905600,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5.1-2025-11-12",
+			DisplayName:         "GPT 5",
+			Description:         "Stable version of GPT 5, The best model for coding and agentic tasks across domains.",
+			ContextLength:       400000,
+			MaxCompletionTokens: 128000,
+			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"none", "low", "medium", "high"}},
+		},
+		{
+			ID:                  "gpt-5.1-codex",
+			Object:              "model",
+			Created:             1762905600,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5.1-2025-11-12",
+			DisplayName:         "GPT 5.1 Codex",
+			Description:         "Stable version of GPT 5.1 Codex, The best model for coding and agentic tasks across domains.",
+			ContextLength:       400000,
+			MaxCompletionTokens: 128000,
+			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high"}},
+		},
+		{
+			ID:                  "gpt-5.1-codex-mini",
+			Object:              "model",
+			Created:             1762905600,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5.1-2025-11-12",
+			DisplayName:         "GPT 5.1 Codex Mini",
+			Description:         "Stable version of GPT 5.1 Codex Mini: cheaper, faster, but less capable version of GPT 5.1 Codex.",
+			ContextLength:       400000,
+			MaxCompletionTokens: 128000,
+			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high"}},
+		},
+		{
+			ID:                  "gpt-5.1-codex-max",
+			Object:              "model",
+			Created:             1763424000,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5.1-max",
+			DisplayName:         "GPT 5.1 Codex Max",
+			Description:         "Stable version of GPT 5.1 Codex Max",
+			ContextLength:       400000,
+			MaxCompletionTokens: 128000,
+			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high", "xhigh"}},
+		},
+		{
+			ID:                  "gpt-5.2",
+			Object:              "model",
+			Created:             1765440000,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5.2",
+			DisplayName:         "GPT 5.2",
+			Description:         "Stable version of GPT 5.2",
+			ContextLength:       400000,
+			MaxCompletionTokens: 128000,
+			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"none", "low", "medium", "high", "xhigh"}},
+		},
+		{
+			ID:                  "gpt-5.2-codex",
+			Object:              "model",
+			Created:             1765440000,
+			OwnedBy:             "openai",
+			Type:                "openai",
+			Version:             "gpt-5.2",
+			DisplayName:         "GPT 5.2 Codex",
+			Description:         "Stable version of GPT 5.2 Codex, The best model for coding and agentic tasks across domains.",
+			ContextLength:       400000,
+			MaxCompletionTokens: 128000,
+			SupportedParameters: []string{"tools"},
+			Thinking:            &ThinkingSupport{Levels: []string{"low", "medium", "high", "xhigh"}},
+		},
+	}
+}
+
+// GetQwenModels returns the standard Qwen model definitions
+func GetQwenModels() []*ModelInfo {
+	return []*ModelInfo{
+		{
+			ID:                  "qwen3-coder-plus",
+			Object:              "model",
+			Created:             1753228800,
+			OwnedBy:             "qwen",
+			Type:                "qwen",
+			Version:             "3.0",
+			DisplayName:         "Qwen3 Coder Plus",
+			Description:         "Advanced code generation and understanding model",
+			ContextLength:       32768,
+			MaxCompletionTokens: 8192,
+			SupportedParameters: []string{"temperature", "top_p", "max_tokens", "stream", "stop"},
+		},
+		{
+			ID:                  "qwen3-coder-flash",
+			Object:              "model",
+			Created:             1753228800,
+			OwnedBy:             "qwen",
+			Type:                "qwen",
+			Version:             "3.0",
+			DisplayName:         "Qwen3 Coder Flash",
+			Description:         "Fast code generation model",
+			ContextLength:       8192,
+			MaxCompletionTokens: 2048,
+			SupportedParameters: []string{"temperature", "top_p", "max_tokens", "stream", "stop"},
+		},
+		{
+			ID:                  "vision-model",
+			Object:              "model",
+			Created:             1758672000,
+			OwnedBy:             "qwen",
+			Type:                "qwen",
+			Version:             "3.0",
+			DisplayName:         "Qwen3 Vision Model",
+			Description:         "Vision model model",
+			ContextLength:       32768,
+			MaxCompletionTokens: 2048,
+			SupportedParameters: []string{"temperature", "top_p", "max_tokens", "stream", "stop"},
+		},
+	}
+}
+
+// iFlowThinkingSupport is a shared ThinkingSupport configuration for iFlow models
+// that support thinking mode via chat_template_kwargs.enable_thinking (boolean toggle).
+// Uses level-based configuration so standard normalization flows apply before conversion.
+var iFlowThinkingSupport = &ThinkingSupport{
+	Levels: []string{"none", "auto", "minimal", "low", "medium", "high", "xhigh"},
+}
+
+// GetIFlowModels returns supported models for iFlow OAuth accounts.
+func GetIFlowModels() []*ModelInfo {
+	entries := []struct {
+		ID          string
+		DisplayName string
+		Description string
+		Created     int64
+		Thinking    *ThinkingSupport
+	}{
+		{ID: "tstars2.0", DisplayName: "TStars-2.0", Description: "iFlow TStars-2.0 multimodal assistant", Created: 1746489600},
+		{ID: "qwen3-coder-plus", DisplayName: "Qwen3-Coder-Plus", Description: "Qwen3 Coder Plus code generation", Created: 1753228800},
+		{ID: "qwen3-max", DisplayName: "Qwen3-Max", Description: "Qwen3 flagship model", Created: 1758672000},
+		{ID: "qwen3-vl-plus", DisplayName: "Qwen3-VL-Plus", Description: "Qwen3 multimodal vision-language", Created: 1758672000},
+		{ID: "qwen3-max-preview", DisplayName: "Qwen3-Max-Preview", Description: "Qwen3 Max preview build", Created: 1757030400, Thinking: iFlowThinkingSupport},
+		{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: "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},
+		{ID: "deepseek-v3.2-reasoner", DisplayName: "DeepSeek-V3.2", Description: "DeepSeek V3.2 Reasoner", Created: 1764576000},
+		{ID: "deepseek-v3.2", DisplayName: "DeepSeek-V3.2-Exp", Description: "DeepSeek V3.2 experimental", Created: 1759104000, Thinking: iFlowThinkingSupport},
+		{ID: "deepseek-v3.1", DisplayName: "DeepSeek-V3.1-Terminus", Description: "DeepSeek V3.1 Terminus", Created: 1756339200, Thinking: iFlowThinkingSupport},
+		{ID: "deepseek-r1", DisplayName: "DeepSeek-R1", Description: "DeepSeek reasoning model R1", Created: 1737331200},
+		{ID: "deepseek-v3", DisplayName: "DeepSeek-V3-671B", Description: "DeepSeek V3 671B", Created: 1734307200},
+		{ID: "qwen3-32b", DisplayName: "Qwen3-32B", Description: "Qwen3 32B", Created: 1747094400},
+		{ID: "qwen3-235b-a22b-thinking-2507", DisplayName: "Qwen3-235B-A22B-Thinking", Description: "Qwen3 235B A22B Thinking (2507)", Created: 1753401600},
+		{ID: "qwen3-235b-a22b-instruct", DisplayName: "Qwen3-235B-A22B-Instruct", Description: "Qwen3 235B A22B Instruct", Created: 1753401600},
+		{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: "iflow-rome-30ba3b", DisplayName: "iFlow-ROME", Description: "iFlow Rome 30BA3B model", Created: 1736899200},
+	}
+	models := make([]*ModelInfo, 0, len(entries))
+	for _, entry := range entries {
+		models = append(models, &ModelInfo{
+			ID:          entry.ID,
+			Object:      "model",
+			Created:     entry.Created,
+			OwnedBy:     "iflow",
+			Type:        "iflow",
+			DisplayName: entry.DisplayName,
+			Description: entry.Description,
+			Thinking:    entry.Thinking,
+		})
+	}
+	return models
+}
+
+// AntigravityModelConfig captures static antigravity model overrides, including
+// Thinking budget limits and provider max completion tokens.
+type AntigravityModelConfig struct {
+	Thinking            *ThinkingSupport
+	MaxCompletionTokens int
+}
+
+// GetAntigravityModelConfig returns static configuration for antigravity models.
+// Keys use upstream model names returned by the Antigravity models endpoint.
+func GetAntigravityModelConfig() map[string]*AntigravityModelConfig {
+	return map[string]*AntigravityModelConfig{
+		// "rev19-uic3-1p":              {Thinking: &ThinkingSupport{Min: 128, Max: 32768, ZeroAllowed: false, DynamicAllowed: true}},
+		"gemini-2.5-flash":           {Thinking: &ThinkingSupport{Min: 0, Max: 24576, ZeroAllowed: true, DynamicAllowed: true}},
+		"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-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-sonnet-4-5":          {MaxCompletionTokens: 64000},
+		"gpt-oss-120b-medium":        {},
+		"tab_flash_lite_preview":     {},
+	}
+}

internal/registry/model_registry.go

@@ -1033,10 +1033,10 @@ func (r *ModelRegistry) convertModelToMap(model *ModelInfo, handlerType string)
 			"owned_by": model.OwnedBy,
 		}
 		if model.Created > 0 {
-			result["created"] = model.Created
+			result["created_at"] = model.Created
 		}
 		if model.Type != "" {
-			result["type"] = model.Type
+			result["type"] = "model"
 		}
 		if model.DisplayName != "" {
 			result["display_name"] = model.DisplayName

internal/runtime/executor/aistudio_executor.go

@@ -398,7 +398,8 @@ func (e *AIStudioExecutor) translateRequest(req cliproxyexecutor.Request, opts c
 		return nil, translatedPayload{}, err
 	}
 	payload = fixGeminiImageAspectRatio(baseModel, payload)
-	payload = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", payload, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	payload = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", payload, originalTranslated, requestedModel)
 	payload, _ = sjson.DeleteBytes(payload, "generationConfig.maxOutputTokens")
 	payload, _ = sjson.DeleteBytes(payload, "generationConfig.responseMimeType")
 	payload, _ = sjson.DeleteBytes(payload, "generationConfig.responseJsonSchema")

internal/runtime/executor/antigravity_executor.go

@@ -142,92 +142,114 @@ func (e *AntigravityExecutor) Execute(ctx context.Context, auth *cliproxyauth.Au
 		return resp, err
 	}
 
-	translated = applyPayloadConfigWithRoot(e.cfg, baseModel, "antigravity", "request", translated, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	translated = applyPayloadConfigWithRoot(e.cfg, baseModel, "antigravity", "request", translated, originalTranslated, requestedModel)
 
 	baseURLs := antigravityBaseURLFallbackOrder(auth)
 	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
 
-	var lastStatus int
-	var lastBody []byte
-	var lastErr error
+	attempts := antigravityRetryAttempts(auth, e.cfg)
 
-	for idx, baseURL := range baseURLs {
-		httpReq, errReq := e.buildRequest(ctx, auth, token, baseModel, translated, false, opts.Alt, baseURL)
-		if errReq != nil {
-			err = errReq
-			return resp, err
-		}
+attemptLoop:
+	for attempt := 0; attempt < attempts; attempt++ {
+		var lastStatus int
+		var lastBody []byte
+		var lastErr error
 
-		httpResp, errDo := httpClient.Do(httpReq)
-		if errDo != nil {
-			recordAPIResponseError(ctx, e.cfg, errDo)
-			if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
-				return resp, errDo
+		for idx, baseURL := range baseURLs {
+			httpReq, errReq := e.buildRequest(ctx, auth, token, baseModel, translated, false, opts.Alt, baseURL)
+			if errReq != nil {
+				err = errReq
+				return resp, err
 			}
-			lastStatus = 0
-			lastBody = nil
-			lastErr = errDo
-			if idx+1 < len(baseURLs) {
-				log.Debugf("antigravity executor: request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
-				continue
+
+			httpResp, errDo := httpClient.Do(httpReq)
+			if errDo != nil {
+				recordAPIResponseError(ctx, e.cfg, errDo)
+				if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
+					return resp, errDo
+				}
+				lastStatus = 0
+				lastBody = nil
+				lastErr = errDo
+				if idx+1 < len(baseURLs) {
+					log.Debugf("antigravity executor: request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+					continue
+				}
+				err = errDo
+				return resp, err
 			}
-			err = errDo
-			return resp, err
+
+			recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
+			bodyBytes, errRead := io.ReadAll(httpResp.Body)
+			if errClose := httpResp.Body.Close(); errClose != nil {
+				log.Errorf("antigravity executor: close response body error: %v", errClose)
+			}
+			if errRead != nil {
+				recordAPIResponseError(ctx, e.cfg, errRead)
+				err = errRead
+				return resp, err
+			}
+			appendAPIResponseChunk(ctx, e.cfg, bodyBytes)
+
+			if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
+				log.Debugf("antigravity executor: upstream error status: %d, body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), bodyBytes))
+				lastStatus = httpResp.StatusCode
+				lastBody = append([]byte(nil), bodyBytes...)
+				lastErr = nil
+				if httpResp.StatusCode == http.StatusTooManyRequests && idx+1 < len(baseURLs) {
+					log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+					continue
+				}
+				if antigravityShouldRetryNoCapacity(httpResp.StatusCode, bodyBytes) {
+					if idx+1 < len(baseURLs) {
+						log.Debugf("antigravity executor: no capacity on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+						continue
+					}
+					if attempt+1 < attempts {
+						delay := antigravityNoCapacityRetryDelay(attempt)
+						log.Debugf("antigravity executor: no capacity for model %s, retrying in %s (attempt %d/%d)", baseModel, delay, attempt+1, attempts)
+						if errWait := antigravityWait(ctx, delay); errWait != nil {
+							return resp, errWait
+						}
+						continue attemptLoop
+					}
+				}
+				sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+				if httpResp.StatusCode == http.StatusTooManyRequests {
+					if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+						sErr.retryAfter = retryAfter
+					}
+				}
+				err = sErr
+				return resp, err
+			}
+
+			reporter.publish(ctx, parseAntigravityUsage(bodyBytes))
+			var param any
+			converted := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, bodyBytes, &param)
+			resp = cliproxyexecutor.Response{Payload: []byte(converted)}
+			reporter.ensurePublished(ctx)
+			return resp, nil
 		}
 
-		recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
-		bodyBytes, errRead := io.ReadAll(httpResp.Body)
-		if errClose := httpResp.Body.Close(); errClose != nil {
-			log.Errorf("antigravity executor: close response body error: %v", errClose)
-		}
-		if errRead != nil {
-			recordAPIResponseError(ctx, e.cfg, errRead)
-			err = errRead
-			return resp, err
-		}
-		appendAPIResponseChunk(ctx, e.cfg, bodyBytes)
-
-		if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
-			log.Debugf("antigravity executor: upstream error status: %d, body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), bodyBytes))
-			lastStatus = httpResp.StatusCode
-			lastBody = append([]byte(nil), bodyBytes...)
-			lastErr = nil
-			if httpResp.StatusCode == http.StatusTooManyRequests && idx+1 < len(baseURLs) {
-				log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
-				continue
-			}
-			sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
-			if httpResp.StatusCode == http.StatusTooManyRequests {
-				if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+		switch {
+		case lastStatus != 0:
+			sErr := statusErr{code: lastStatus, msg: string(lastBody)}
+			if lastStatus == http.StatusTooManyRequests {
+				if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
 					sErr.retryAfter = retryAfter
 				}
 			}
 			err = sErr
-			return resp, err
+		case lastErr != nil:
+			err = lastErr
+		default:
+			err = statusErr{code: http.StatusServiceUnavailable, msg: "antigravity executor: no base url available"}
 		}
-
-		reporter.publish(ctx, parseAntigravityUsage(bodyBytes))
-		var param any
-		converted := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, bodyBytes, &param)
-		resp = cliproxyexecutor.Response{Payload: []byte(converted)}
-		reporter.ensurePublished(ctx)
-		return resp, nil
+		return resp, err
 	}
 
-	switch {
-	case lastStatus != 0:
-		sErr := statusErr{code: lastStatus, msg: string(lastBody)}
-		if lastStatus == http.StatusTooManyRequests {
-			if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
-				sErr.retryAfter = retryAfter
-			}
-		}
-		err = sErr
-	case lastErr != nil:
-		err = lastErr
-	default:
-		err = statusErr{code: http.StatusServiceUnavailable, msg: "antigravity executor: no base url available"}
-	}
 	return resp, err
 }
 
@@ -261,155 +283,177 @@ func (e *AntigravityExecutor) executeClaudeNonStream(ctx context.Context, auth *
 		return resp, err
 	}
 
-	translated = applyPayloadConfigWithRoot(e.cfg, baseModel, "antigravity", "request", translated, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	translated = applyPayloadConfigWithRoot(e.cfg, baseModel, "antigravity", "request", translated, originalTranslated, requestedModel)
 
 	baseURLs := antigravityBaseURLFallbackOrder(auth)
 	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
 
-	var lastStatus int
-	var lastBody []byte
-	var lastErr error
+	attempts := antigravityRetryAttempts(auth, e.cfg)
 
-	for idx, baseURL := range baseURLs {
-		httpReq, errReq := e.buildRequest(ctx, auth, token, baseModel, translated, true, opts.Alt, baseURL)
-		if errReq != nil {
-			err = errReq
-			return resp, err
-		}
+attemptLoop:
+	for attempt := 0; attempt < attempts; attempt++ {
+		var lastStatus int
+		var lastBody []byte
+		var lastErr error
 
-		httpResp, errDo := httpClient.Do(httpReq)
-		if errDo != nil {
-			recordAPIResponseError(ctx, e.cfg, errDo)
-			if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
-				return resp, errDo
+		for idx, baseURL := range baseURLs {
+			httpReq, errReq := e.buildRequest(ctx, auth, token, baseModel, translated, true, opts.Alt, baseURL)
+			if errReq != nil {
+				err = errReq
+				return resp, err
 			}
-			lastStatus = 0
-			lastBody = nil
-			lastErr = errDo
-			if idx+1 < len(baseURLs) {
-				log.Debugf("antigravity executor: request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
-				continue
-			}
-			err = errDo
-			return resp, err
-		}
-		recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
-		if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
-			bodyBytes, errRead := io.ReadAll(httpResp.Body)
-			if errClose := httpResp.Body.Close(); errClose != nil {
-				log.Errorf("antigravity executor: close response body error: %v", errClose)
-			}
-			if errRead != nil {
-				recordAPIResponseError(ctx, e.cfg, errRead)
-				if errors.Is(errRead, context.Canceled) || errors.Is(errRead, context.DeadlineExceeded) {
-					err = errRead
-					return resp, err
-				}
-				if errCtx := ctx.Err(); errCtx != nil {
-					err = errCtx
-					return resp, err
+
+			httpResp, errDo := httpClient.Do(httpReq)
+			if errDo != nil {
+				recordAPIResponseError(ctx, e.cfg, errDo)
+				if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
+					return resp, errDo
 				}
 				lastStatus = 0
 				lastBody = nil
-				lastErr = errRead
+				lastErr = errDo
 				if idx+1 < len(baseURLs) {
-					log.Debugf("antigravity executor: read error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+					log.Debugf("antigravity executor: request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
 					continue
 				}
-				err = errRead
+				err = errDo
 				return resp, err
 			}
-			appendAPIResponseChunk(ctx, e.cfg, bodyBytes)
-			lastStatus = httpResp.StatusCode
-			lastBody = append([]byte(nil), bodyBytes...)
-			lastErr = nil
-			if httpResp.StatusCode == http.StatusTooManyRequests && idx+1 < len(baseURLs) {
-				log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
-				continue
+			recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
+			if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
+				bodyBytes, errRead := io.ReadAll(httpResp.Body)
+				if errClose := httpResp.Body.Close(); errClose != nil {
+					log.Errorf("antigravity executor: close response body error: %v", errClose)
+				}
+				if errRead != nil {
+					recordAPIResponseError(ctx, e.cfg, errRead)
+					if errors.Is(errRead, context.Canceled) || errors.Is(errRead, context.DeadlineExceeded) {
+						err = errRead
+						return resp, err
+					}
+					if errCtx := ctx.Err(); errCtx != nil {
+						err = errCtx
+						return resp, err
+					}
+					lastStatus = 0
+					lastBody = nil
+					lastErr = errRead
+					if idx+1 < len(baseURLs) {
+						log.Debugf("antigravity executor: read error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+						continue
+					}
+					err = errRead
+					return resp, err
+				}
+				appendAPIResponseChunk(ctx, e.cfg, bodyBytes)
+				lastStatus = httpResp.StatusCode
+				lastBody = append([]byte(nil), bodyBytes...)
+				lastErr = nil
+				if httpResp.StatusCode == http.StatusTooManyRequests && idx+1 < len(baseURLs) {
+					log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+					continue
+				}
+				if antigravityShouldRetryNoCapacity(httpResp.StatusCode, bodyBytes) {
+					if idx+1 < len(baseURLs) {
+						log.Debugf("antigravity executor: no capacity on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+						continue
+					}
+					if attempt+1 < attempts {
+						delay := antigravityNoCapacityRetryDelay(attempt)
+						log.Debugf("antigravity executor: no capacity for model %s, retrying in %s (attempt %d/%d)", baseModel, delay, attempt+1, attempts)
+						if errWait := antigravityWait(ctx, delay); errWait != nil {
+							return resp, errWait
+						}
+						continue attemptLoop
+					}
+				}
+				sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+				if httpResp.StatusCode == http.StatusTooManyRequests {
+					if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+						sErr.retryAfter = retryAfter
+					}
+				}
+				err = sErr
+				return resp, err
 			}
-			sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
-			if httpResp.StatusCode == http.StatusTooManyRequests {
-				if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+
+			out := make(chan cliproxyexecutor.StreamChunk)
+			go func(resp *http.Response) {
+				defer close(out)
+				defer func() {
+					if errClose := resp.Body.Close(); errClose != nil {
+						log.Errorf("antigravity executor: close response body error: %v", errClose)
+					}
+				}()
+				scanner := bufio.NewScanner(resp.Body)
+				scanner.Buffer(nil, streamScannerBuffer)
+				for scanner.Scan() {
+					line := scanner.Bytes()
+					appendAPIResponseChunk(ctx, e.cfg, line)
+
+					// Filter usage metadata for all models
+					// Only retain usage statistics in the terminal chunk
+					line = FilterSSEUsageMetadata(line)
+
+					payload := jsonPayload(line)
+					if payload == nil {
+						continue
+					}
+
+					if detail, ok := parseAntigravityStreamUsage(payload); ok {
+						reporter.publish(ctx, detail)
+					}
+
+					out <- cliproxyexecutor.StreamChunk{Payload: payload}
+				}
+				if errScan := scanner.Err(); errScan != nil {
+					recordAPIResponseError(ctx, e.cfg, errScan)
+					reporter.publishFailure(ctx)
+					out <- cliproxyexecutor.StreamChunk{Err: errScan}
+				} else {
+					reporter.ensurePublished(ctx)
+				}
+			}(httpResp)
+
+			var buffer bytes.Buffer
+			for chunk := range out {
+				if chunk.Err != nil {
+					return resp, chunk.Err
+				}
+				if len(chunk.Payload) > 0 {
+					_, _ = buffer.Write(chunk.Payload)
+					_, _ = buffer.Write([]byte("\n"))
+				}
+			}
+			resp = cliproxyexecutor.Response{Payload: e.convertStreamToNonStream(buffer.Bytes())}
+
+			reporter.publish(ctx, parseAntigravityUsage(resp.Payload))
+			var param any
+			converted := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, resp.Payload, &param)
+			resp = cliproxyexecutor.Response{Payload: []byte(converted)}
+			reporter.ensurePublished(ctx)
+
+			return resp, nil
+		}
+
+		switch {
+		case lastStatus != 0:
+			sErr := statusErr{code: lastStatus, msg: string(lastBody)}
+			if lastStatus == http.StatusTooManyRequests {
+				if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
 					sErr.retryAfter = retryAfter
 				}
 			}
 			err = sErr
-			return resp, err
+		case lastErr != nil:
+			err = lastErr
+		default:
+			err = statusErr{code: http.StatusServiceUnavailable, msg: "antigravity executor: no base url available"}
 		}
-
-		out := make(chan cliproxyexecutor.StreamChunk)
-		go func(resp *http.Response) {
-			defer close(out)
-			defer func() {
-				if errClose := resp.Body.Close(); errClose != nil {
-					log.Errorf("antigravity executor: close response body error: %v", errClose)
-				}
-			}()
-			scanner := bufio.NewScanner(resp.Body)
-			scanner.Buffer(nil, streamScannerBuffer)
-			for scanner.Scan() {
-				line := scanner.Bytes()
-				appendAPIResponseChunk(ctx, e.cfg, line)
-
-				// Filter usage metadata for all models
-				// Only retain usage statistics in the terminal chunk
-				line = FilterSSEUsageMetadata(line)
-
-				payload := jsonPayload(line)
-				if payload == nil {
-					continue
-				}
-
-				if detail, ok := parseAntigravityStreamUsage(payload); ok {
-					reporter.publish(ctx, detail)
-				}
-
-				out <- cliproxyexecutor.StreamChunk{Payload: payload}
-			}
-			if errScan := scanner.Err(); errScan != nil {
-				recordAPIResponseError(ctx, e.cfg, errScan)
-				reporter.publishFailure(ctx)
-				out <- cliproxyexecutor.StreamChunk{Err: errScan}
-			} else {
-				reporter.ensurePublished(ctx)
-			}
-		}(httpResp)
-
-		var buffer bytes.Buffer
-		for chunk := range out {
-			if chunk.Err != nil {
-				return resp, chunk.Err
-			}
-			if len(chunk.Payload) > 0 {
-				_, _ = buffer.Write(chunk.Payload)
-				_, _ = buffer.Write([]byte("\n"))
-			}
-		}
-		resp = cliproxyexecutor.Response{Payload: e.convertStreamToNonStream(buffer.Bytes())}
-
-		reporter.publish(ctx, parseAntigravityUsage(resp.Payload))
-		var param any
-		converted := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, resp.Payload, &param)
-		resp = cliproxyexecutor.Response{Payload: []byte(converted)}
-		reporter.ensurePublished(ctx)
-
-		return resp, nil
+		return resp, err
 	}
 
-	switch {
-	case lastStatus != 0:
-		sErr := statusErr{code: lastStatus, msg: string(lastBody)}
-		if lastStatus == http.StatusTooManyRequests {
-			if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
-				sErr.retryAfter = retryAfter
-			}
-		}
-		err = sErr
-	case lastErr != nil:
-		err = lastErr
-	default:
-		err = statusErr{code: http.StatusServiceUnavailable, msg: "antigravity executor: no base url available"}
-	}
 	return resp, err
 }
 
@@ -627,144 +671,166 @@ func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxya
 		return nil, err
 	}
 
-	translated = applyPayloadConfigWithRoot(e.cfg, baseModel, "antigravity", "request", translated, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	translated = applyPayloadConfigWithRoot(e.cfg, baseModel, "antigravity", "request", translated, originalTranslated, requestedModel)
 
 	baseURLs := antigravityBaseURLFallbackOrder(auth)
 	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
 
-	var lastStatus int
-	var lastBody []byte
-	var lastErr error
+	attempts := antigravityRetryAttempts(auth, e.cfg)
 
-	for idx, baseURL := range baseURLs {
-		httpReq, errReq := e.buildRequest(ctx, auth, token, baseModel, translated, true, opts.Alt, baseURL)
-		if errReq != nil {
-			err = errReq
-			return nil, err
-		}
-		httpResp, errDo := httpClient.Do(httpReq)
-		if errDo != nil {
-			recordAPIResponseError(ctx, e.cfg, errDo)
-			if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
-				return nil, errDo
+attemptLoop:
+	for attempt := 0; attempt < attempts; attempt++ {
+		var lastStatus int
+		var lastBody []byte
+		var lastErr error
+
+		for idx, baseURL := range baseURLs {
+			httpReq, errReq := e.buildRequest(ctx, auth, token, baseModel, translated, true, opts.Alt, baseURL)
+			if errReq != nil {
+				err = errReq
+				return nil, err
 			}
-			lastStatus = 0
-			lastBody = nil
-			lastErr = errDo
-			if idx+1 < len(baseURLs) {
-				log.Debugf("antigravity executor: request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
-				continue
-			}
-			err = errDo
-			return nil, err
-		}
-		recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
-		if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
-			bodyBytes, errRead := io.ReadAll(httpResp.Body)
-			if errClose := httpResp.Body.Close(); errClose != nil {
-				log.Errorf("antigravity executor: close response body error: %v", errClose)
-			}
-			if errRead != nil {
-				recordAPIResponseError(ctx, e.cfg, errRead)
-				if errors.Is(errRead, context.Canceled) || errors.Is(errRead, context.DeadlineExceeded) {
-					err = errRead
-					return nil, err
-				}
-				if errCtx := ctx.Err(); errCtx != nil {
-					err = errCtx
-					return nil, err
+			httpResp, errDo := httpClient.Do(httpReq)
+			if errDo != nil {
+				recordAPIResponseError(ctx, e.cfg, errDo)
+				if errors.Is(errDo, context.Canceled) || errors.Is(errDo, context.DeadlineExceeded) {
+					return nil, errDo
 				}
 				lastStatus = 0
 				lastBody = nil
-				lastErr = errRead
+				lastErr = errDo
 				if idx+1 < len(baseURLs) {
-					log.Debugf("antigravity executor: read error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+					log.Debugf("antigravity executor: request error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
 					continue
 				}
-				err = errRead
+				err = errDo
 				return nil, err
 			}
-			appendAPIResponseChunk(ctx, e.cfg, bodyBytes)
-			lastStatus = httpResp.StatusCode
-			lastBody = append([]byte(nil), bodyBytes...)
-			lastErr = nil
-			if httpResp.StatusCode == http.StatusTooManyRequests && idx+1 < len(baseURLs) {
-				log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
-				continue
+			recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
+			if httpResp.StatusCode < http.StatusOK || httpResp.StatusCode >= http.StatusMultipleChoices {
+				bodyBytes, errRead := io.ReadAll(httpResp.Body)
+				if errClose := httpResp.Body.Close(); errClose != nil {
+					log.Errorf("antigravity executor: close response body error: %v", errClose)
+				}
+				if errRead != nil {
+					recordAPIResponseError(ctx, e.cfg, errRead)
+					if errors.Is(errRead, context.Canceled) || errors.Is(errRead, context.DeadlineExceeded) {
+						err = errRead
+						return nil, err
+					}
+					if errCtx := ctx.Err(); errCtx != nil {
+						err = errCtx
+						return nil, err
+					}
+					lastStatus = 0
+					lastBody = nil
+					lastErr = errRead
+					if idx+1 < len(baseURLs) {
+						log.Debugf("antigravity executor: read error on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+						continue
+					}
+					err = errRead
+					return nil, err
+				}
+				appendAPIResponseChunk(ctx, e.cfg, bodyBytes)
+				lastStatus = httpResp.StatusCode
+				lastBody = append([]byte(nil), bodyBytes...)
+				lastErr = nil
+				if httpResp.StatusCode == http.StatusTooManyRequests && idx+1 < len(baseURLs) {
+					log.Debugf("antigravity executor: rate limited on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+					continue
+				}
+				if antigravityShouldRetryNoCapacity(httpResp.StatusCode, bodyBytes) {
+					if idx+1 < len(baseURLs) {
+						log.Debugf("antigravity executor: no capacity on base url %s, retrying with fallback base url: %s", baseURL, baseURLs[idx+1])
+						continue
+					}
+					if attempt+1 < attempts {
+						delay := antigravityNoCapacityRetryDelay(attempt)
+						log.Debugf("antigravity executor: no capacity for model %s, retrying in %s (attempt %d/%d)", baseModel, delay, attempt+1, attempts)
+						if errWait := antigravityWait(ctx, delay); errWait != nil {
+							return nil, errWait
+						}
+						continue attemptLoop
+					}
+				}
+				sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
+				if httpResp.StatusCode == http.StatusTooManyRequests {
+					if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+						sErr.retryAfter = retryAfter
+					}
+				}
+				err = sErr
+				return nil, err
 			}
-			sErr := statusErr{code: httpResp.StatusCode, msg: string(bodyBytes)}
-			if httpResp.StatusCode == http.StatusTooManyRequests {
-				if retryAfter, parseErr := parseRetryDelay(bodyBytes); parseErr == nil && retryAfter != nil {
+
+			out := make(chan cliproxyexecutor.StreamChunk)
+			stream = out
+			go func(resp *http.Response) {
+				defer close(out)
+				defer func() {
+					if errClose := resp.Body.Close(); errClose != nil {
+						log.Errorf("antigravity executor: close response body error: %v", errClose)
+					}
+				}()
+				scanner := bufio.NewScanner(resp.Body)
+				scanner.Buffer(nil, streamScannerBuffer)
+				var param any
+				for scanner.Scan() {
+					line := scanner.Bytes()
+					appendAPIResponseChunk(ctx, e.cfg, line)
+
+					// Filter usage metadata for all models
+					// Only retain usage statistics in the terminal chunk
+					line = FilterSSEUsageMetadata(line)
+
+					payload := jsonPayload(line)
+					if payload == nil {
+						continue
+					}
+
+					if detail, ok := parseAntigravityStreamUsage(payload); ok {
+						reporter.publish(ctx, detail)
+					}
+
+					chunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, bytes.Clone(payload), &param)
+					for i := range chunks {
+						out <- cliproxyexecutor.StreamChunk{Payload: []byte(chunks[i])}
+					}
+				}
+				tail := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, []byte("[DONE]"), &param)
+				for i := range tail {
+					out <- cliproxyexecutor.StreamChunk{Payload: []byte(tail[i])}
+				}
+				if errScan := scanner.Err(); errScan != nil {
+					recordAPIResponseError(ctx, e.cfg, errScan)
+					reporter.publishFailure(ctx)
+					out <- cliproxyexecutor.StreamChunk{Err: errScan}
+				} else {
+					reporter.ensurePublished(ctx)
+				}
+			}(httpResp)
+			return stream, nil
+		}
+
+		switch {
+		case lastStatus != 0:
+			sErr := statusErr{code: lastStatus, msg: string(lastBody)}
+			if lastStatus == http.StatusTooManyRequests {
+				if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
 					sErr.retryAfter = retryAfter
 				}
 			}
 			err = sErr
-			return nil, err
+		case lastErr != nil:
+			err = lastErr
+		default:
+			err = statusErr{code: http.StatusServiceUnavailable, msg: "antigravity executor: no base url available"}
 		}
-
-		out := make(chan cliproxyexecutor.StreamChunk)
-		stream = out
-		go func(resp *http.Response) {
-			defer close(out)
-			defer func() {
-				if errClose := resp.Body.Close(); errClose != nil {
-					log.Errorf("antigravity executor: close response body error: %v", errClose)
-				}
-			}()
-			scanner := bufio.NewScanner(resp.Body)
-			scanner.Buffer(nil, streamScannerBuffer)
-			var param any
-			for scanner.Scan() {
-				line := scanner.Bytes()
-				appendAPIResponseChunk(ctx, e.cfg, line)
-
-				// Filter usage metadata for all models
-				// Only retain usage statistics in the terminal chunk
-				line = FilterSSEUsageMetadata(line)
-
-				payload := jsonPayload(line)
-				if payload == nil {
-					continue
-				}
-
-				if detail, ok := parseAntigravityStreamUsage(payload); ok {
-					reporter.publish(ctx, detail)
-				}
-
-				chunks := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, bytes.Clone(payload), &param)
-				for i := range chunks {
-					out <- cliproxyexecutor.StreamChunk{Payload: []byte(chunks[i])}
-				}
-			}
-			tail := sdktranslator.TranslateStream(ctx, to, from, req.Model, bytes.Clone(opts.OriginalRequest), translated, []byte("[DONE]"), &param)
-			for i := range tail {
-				out <- cliproxyexecutor.StreamChunk{Payload: []byte(tail[i])}
-			}
-			if errScan := scanner.Err(); errScan != nil {
-				recordAPIResponseError(ctx, e.cfg, errScan)
-				reporter.publishFailure(ctx)
-				out <- cliproxyexecutor.StreamChunk{Err: errScan}
-			} else {
-				reporter.ensurePublished(ctx)
-			}
-		}(httpResp)
-		return stream, nil
+		return nil, err
 	}
 
-	switch {
-	case lastStatus != 0:
-		sErr := statusErr{code: lastStatus, msg: string(lastBody)}
-		if lastStatus == http.StatusTooManyRequests {
-			if retryAfter, parseErr := parseRetryDelay(lastBody); parseErr == nil && retryAfter != nil {
-				sErr.retryAfter = retryAfter
-			}
-		}
-		err = sErr
-	case lastErr != nil:
-		err = lastErr
-	default:
-		err = statusErr{code: http.StatusServiceUnavailable, msg: "antigravity executor: no base url available"}
-	}
 	return nil, err
 }
 
@@ -994,7 +1060,7 @@ func FetchAntigravityModels(ctx context.Context, auth *cliproxyauth.Auth, cfg *c
 		now := time.Now().Unix()
 		modelConfig := registry.GetAntigravityModelConfig()
 		models := make([]*registry.ModelInfo, 0, len(result.Map()))
-		for originalName := range result.Map() {
+		for originalName, modelData := range result.Map() {
 			modelID := strings.TrimSpace(originalName)
 			if modelID == "" {
 				continue
@@ -1004,12 +1070,18 @@ func FetchAntigravityModels(ctx context.Context, auth *cliproxyauth.Auth, cfg *c
 				continue
 			}
 			modelCfg := modelConfig[modelID]
-			modelName := modelID
+
+			// Extract displayName from upstream response, fallback to modelID
+			displayName := modelData.Get("displayName").String()
+			if displayName == "" {
+				displayName = modelID
+			}
+
 			modelInfo := &registry.ModelInfo{
 				ID:          modelID,
-				Name:        modelName,
-				Description: modelID,
-				DisplayName: modelID,
+				Name:        modelID,
+				Description: displayName,
+				DisplayName: displayName,
 				Version:     modelID,
 				Object:      "model",
 				Created:     now,
@@ -1213,7 +1285,17 @@ func (e *AntigravityExecutor) buildRequest(ctx context.Context, auth *cliproxyau
 		// Use the centralized schema cleaner to handle unsupported keywords,
 		// const->enum conversion, and flattening of types/anyOf.
 		strJSON = util.CleanJSONSchemaForAntigravity(strJSON)
-
+		payload = []byte(strJSON)
+	} else {
+		strJSON := string(payload)
+		paths := make([]string, 0)
+		util.Walk(gjson.Parse(strJSON), "", "parametersJsonSchema", &paths)
+		for _, p := range paths {
+			strJSON, _ = util.RenameKey(strJSON, p, p[:len(p)-len("parametersJsonSchema")]+"parameters")
+		}
+		// Clean tool schemas for Gemini to remove unsupported JSON Schema keywords
+		// without adding empty-schema placeholders.
+		strJSON = util.CleanJSONSchemaForGemini(strJSON)
 		payload = []byte(strJSON)
 	}
 
@@ -1230,6 +1312,12 @@ func (e *AntigravityExecutor) buildRequest(ctx context.Context, auth *cliproxyau
 		}
 	}
 
+	if strings.Contains(modelName, "claude") {
+		payload, _ = sjson.SetBytes(payload, "request.toolConfig.functionCallingConfig.mode", "VALIDATED")
+	} else {
+		payload, _ = sjson.DeleteBytes(payload, "request.generationConfig.maxOutputTokens")
+	}
+
 	httpReq, errReq := http.NewRequestWithContext(ctx, http.MethodPost, requestURL.String(), bytes.NewReader(payload))
 	if errReq != nil {
 		return nil, errReq
@@ -1359,14 +1447,70 @@ func resolveUserAgent(auth *cliproxyauth.Auth) string {
 	return defaultAntigravityAgent
 }
 
+func antigravityRetryAttempts(auth *cliproxyauth.Auth, cfg *config.Config) int {
+	retry := 0
+	if cfg != nil {
+		retry = cfg.RequestRetry
+	}
+	if auth != nil {
+		if override, ok := auth.RequestRetryOverride(); ok {
+			retry = override
+		}
+	}
+	if retry < 0 {
+		retry = 0
+	}
+	attempts := retry + 1
+	if attempts < 1 {
+		return 1
+	}
+	return attempts
+}
+
+func antigravityShouldRetryNoCapacity(statusCode int, body []byte) bool {
+	if statusCode != http.StatusServiceUnavailable {
+		return false
+	}
+	if len(body) == 0 {
+		return false
+	}
+	msg := strings.ToLower(string(body))
+	return strings.Contains(msg, "no capacity available")
+}
+
+func antigravityNoCapacityRetryDelay(attempt int) time.Duration {
+	if attempt < 0 {
+		attempt = 0
+	}
+	delay := time.Duration(attempt+1) * 250 * time.Millisecond
+	if delay > 2*time.Second {
+		delay = 2 * time.Second
+	}
+	return delay
+}
+
+func antigravityWait(ctx context.Context, wait time.Duration) error {
+	if wait <= 0 {
+		return nil
+	}
+	timer := time.NewTimer(wait)
+	defer timer.Stop()
+	select {
+	case <-ctx.Done():
+		return ctx.Err()
+	case <-timer.C:
+		return nil
+	}
+}
+
 func antigravityBaseURLFallbackOrder(auth *cliproxyauth.Auth) []string {
 	if base := resolveCustomAntigravityBaseURL(auth); base != "" {
 		return []string{base}
 	}
 	return []string{
-		antigravitySandboxBaseURLDaily,
 		antigravityBaseURLDaily,
-		antigravityBaseURLProd,
+		antigravitySandboxBaseURLDaily,
+		// antigravityBaseURLProd,
 	}
 }
 
@@ -1405,26 +1549,10 @@ func geminiToAntigravity(modelName string, payload []byte, projectID string) []b
 	template, _ = sjson.Set(template, "request.sessionId", generateStableSessionID(payload))
 
 	template, _ = sjson.Delete(template, "request.safetySettings")
-	//	template, _ = sjson.Set(template, "request.toolConfig.functionCallingConfig.mode", "VALIDATED")
-
-	if strings.Contains(modelName, "claude") || strings.Contains(modelName, "gemini-3-pro-high") {
-		gjson.Get(template, "request.tools").ForEach(func(key, tool gjson.Result) bool {
-			tool.Get("functionDeclarations").ForEach(func(funKey, funcDecl gjson.Result) bool {
-				if funcDecl.Get("parametersJsonSchema").Exists() {
-					template, _ = sjson.SetRaw(template, fmt.Sprintf("request.tools.%d.functionDeclarations.%d.parameters", key.Int(), funKey.Int()), funcDecl.Get("parametersJsonSchema").Raw)
-					template, _ = sjson.Delete(template, fmt.Sprintf("request.tools.%d.functionDeclarations.%d.parameters.$schema", key.Int(), funKey.Int()))
-					template, _ = sjson.Delete(template, fmt.Sprintf("request.tools.%d.functionDeclarations.%d.parametersJsonSchema", key.Int(), funKey.Int()))
-				}
-				return true
-			})
-			return true
-		})
+	if toolConfig := gjson.Get(template, "toolConfig"); toolConfig.Exists() && !gjson.Get(template, "request.toolConfig").Exists() {
+		template, _ = sjson.SetRaw(template, "request.toolConfig", toolConfig.Raw)
+		template, _ = sjson.Delete(template, "toolConfig")
 	}
-
-	if !strings.Contains(modelName, "claude") {
-		template, _ = sjson.Delete(template, "request.generationConfig.maxOutputTokens")
-	}
-
 	return []byte(template)
 }
 

internal/runtime/executor/caching_verify_test.go

@@ -0,0 +1,258 @@
+package executor
+
+import (
+	"fmt"
+	"testing"
+
+	"github.com/tidwall/gjson"
+)
+
+func TestEnsureCacheControl(t *testing.T) {
+	// Test case 1: System prompt as string
+	t.Run("String System Prompt", func(t *testing.T) {
+		input := []byte(`{"model": "claude-3-5-sonnet", "system": "This is a long system prompt", "messages": []}`)
+		output := ensureCacheControl(input)
+
+		res := gjson.GetBytes(output, "system.0.cache_control.type")
+		if res.String() != "ephemeral" {
+			t.Errorf("cache_control not found in system string. Output: %s", string(output))
+		}
+	})
+
+	// Test case 2: System prompt as array
+	t.Run("Array System Prompt", func(t *testing.T) {
+		input := []byte(`{"model": "claude-3-5-sonnet", "system": [{"type": "text", "text": "Part 1"}, {"type": "text", "text": "Part 2"}], "messages": []}`)
+		output := ensureCacheControl(input)
+
+		// cache_control should only be on the LAST element
+		res0 := gjson.GetBytes(output, "system.0.cache_control")
+		res1 := gjson.GetBytes(output, "system.1.cache_control.type")
+
+		if res0.Exists() {
+			t.Errorf("cache_control should NOT be on the first element")
+		}
+		if res1.String() != "ephemeral" {
+			t.Errorf("cache_control not found on last system element. Output: %s", string(output))
+		}
+	})
+
+	// Test case 3: Tools are cached
+	t.Run("Tools Caching", func(t *testing.T) {
+		input := []byte(`{
+			"model": "claude-3-5-sonnet",
+			"tools": [
+				{"name": "tool1", "description": "First tool", "input_schema": {"type": "object"}},
+				{"name": "tool2", "description": "Second tool", "input_schema": {"type": "object"}}
+			],
+			"system": "System prompt",
+			"messages": []
+		}`)
+		output := ensureCacheControl(input)
+
+		// cache_control should only be on the LAST tool
+		tool0Cache := gjson.GetBytes(output, "tools.0.cache_control")
+		tool1Cache := gjson.GetBytes(output, "tools.1.cache_control.type")
+
+		if tool0Cache.Exists() {
+			t.Errorf("cache_control should NOT be on the first tool")
+		}
+		if tool1Cache.String() != "ephemeral" {
+			t.Errorf("cache_control not found on last tool. Output: %s", string(output))
+		}
+
+		// System should also have cache_control
+		systemCache := gjson.GetBytes(output, "system.0.cache_control.type")
+		if systemCache.String() != "ephemeral" {
+			t.Errorf("cache_control not found in system. Output: %s", string(output))
+		}
+	})
+
+	// Test case 4: Tools and system are INDEPENDENT breakpoints
+	// Per Anthropic docs: Up to 4 breakpoints allowed, tools and system are cached separately
+	t.Run("Independent Cache Breakpoints", func(t *testing.T) {
+		input := []byte(`{
+			"model": "claude-3-5-sonnet",
+			"tools": [
+				{"name": "tool1", "description": "First tool", "input_schema": {"type": "object"}, "cache_control": {"type": "ephemeral"}}
+			],
+			"system": [{"type": "text", "text": "System"}],
+			"messages": []
+		}`)
+		output := ensureCacheControl(input)
+
+		// Tool already has cache_control - should not be changed
+		tool0Cache := gjson.GetBytes(output, "tools.0.cache_control.type")
+		if tool0Cache.String() != "ephemeral" {
+			t.Errorf("existing cache_control was incorrectly removed")
+		}
+
+		// System SHOULD get cache_control because it is an INDEPENDENT breakpoint
+		// Tools and system are separate cache levels in the hierarchy
+		systemCache := gjson.GetBytes(output, "system.0.cache_control.type")
+		if systemCache.String() != "ephemeral" {
+			t.Errorf("system should have its own cache_control breakpoint (independent of tools)")
+		}
+	})
+
+	// Test case 5: Only tools, no system
+	t.Run("Only Tools No System", func(t *testing.T) {
+		input := []byte(`{
+			"model": "claude-3-5-sonnet",
+			"tools": [
+				{"name": "tool1", "description": "Tool", "input_schema": {"type": "object"}}
+			],
+			"messages": [{"role": "user", "content": "Hi"}]
+		}`)
+		output := ensureCacheControl(input)
+
+		toolCache := gjson.GetBytes(output, "tools.0.cache_control.type")
+		if toolCache.String() != "ephemeral" {
+			t.Errorf("cache_control not found on tool. Output: %s", string(output))
+		}
+	})
+
+	// Test case 6: Many tools (Claude Code scenario)
+	t.Run("Many Tools (Claude Code Scenario)", func(t *testing.T) {
+		// Simulate Claude Code with many tools
+		toolsJSON := `[`
+		for i := 0; i < 50; i++ {
+			if i > 0 {
+				toolsJSON += ","
+			}
+			toolsJSON += fmt.Sprintf(`{"name": "tool%d", "description": "Tool %d", "input_schema": {"type": "object"}}`, i, i)
+		}
+		toolsJSON += `]`
+
+		input := []byte(fmt.Sprintf(`{
+			"model": "claude-3-5-sonnet",
+			"tools": %s,
+			"system": [{"type": "text", "text": "You are Claude Code"}],
+			"messages": [{"role": "user", "content": "Hello"}]
+		}`, toolsJSON))
+
+		output := ensureCacheControl(input)
+
+		// Only the last tool (index 49) should have cache_control
+		for i := 0; i < 49; i++ {
+			path := fmt.Sprintf("tools.%d.cache_control", i)
+			if gjson.GetBytes(output, path).Exists() {
+				t.Errorf("tool %d should NOT have cache_control", i)
+			}
+		}
+
+		lastToolCache := gjson.GetBytes(output, "tools.49.cache_control.type")
+		if lastToolCache.String() != "ephemeral" {
+			t.Errorf("last tool (49) should have cache_control")
+		}
+
+		// System should also have cache_control
+		systemCache := gjson.GetBytes(output, "system.0.cache_control.type")
+		if systemCache.String() != "ephemeral" {
+			t.Errorf("system should have cache_control")
+		}
+
+		t.Log("test passed: 50 tools - cache_control only on last tool")
+	})
+
+	// Test case 7: Empty tools array
+	t.Run("Empty Tools Array", func(t *testing.T) {
+		input := []byte(`{"model": "claude-3-5-sonnet", "tools": [], "system": "Test", "messages": []}`)
+		output := ensureCacheControl(input)
+
+		// System should still get cache_control
+		systemCache := gjson.GetBytes(output, "system.0.cache_control.type")
+		if systemCache.String() != "ephemeral" {
+			t.Errorf("system should have cache_control even with empty tools array")
+		}
+	})
+
+	// Test case 8: Messages caching for multi-turn (second-to-last user)
+	t.Run("Messages Caching Second-To-Last User", func(t *testing.T) {
+		input := []byte(`{
+			"model": "claude-3-5-sonnet",
+			"messages": [
+				{"role": "user", "content": "First user"},
+				{"role": "assistant", "content": "Assistant reply"},
+				{"role": "user", "content": "Second user"},
+				{"role": "assistant", "content": "Assistant reply 2"},
+				{"role": "user", "content": "Third user"}
+			]
+		}`)
+		output := ensureCacheControl(input)
+
+		cacheType := gjson.GetBytes(output, "messages.2.content.0.cache_control.type")
+		if cacheType.String() != "ephemeral" {
+			t.Errorf("cache_control not found on second-to-last user turn. Output: %s", string(output))
+		}
+
+		lastUserCache := gjson.GetBytes(output, "messages.4.content.0.cache_control")
+		if lastUserCache.Exists() {
+			t.Errorf("last user turn should NOT have cache_control")
+		}
+	})
+
+	// Test case 9: Existing message cache_control should skip injection
+	t.Run("Messages Skip When Cache Control Exists", func(t *testing.T) {
+		input := []byte(`{
+			"model": "claude-3-5-sonnet",
+			"messages": [
+				{"role": "user", "content": [{"type": "text", "text": "First user"}]},
+				{"role": "assistant", "content": [{"type": "text", "text": "Assistant reply", "cache_control": {"type": "ephemeral"}}]},
+				{"role": "user", "content": [{"type": "text", "text": "Second user"}]}
+			]
+		}`)
+		output := ensureCacheControl(input)
+
+		userCache := gjson.GetBytes(output, "messages.0.content.0.cache_control")
+		if userCache.Exists() {
+			t.Errorf("cache_control should NOT be injected when a message already has cache_control")
+		}
+
+		existingCache := gjson.GetBytes(output, "messages.1.content.0.cache_control.type")
+		if existingCache.String() != "ephemeral" {
+			t.Errorf("existing cache_control should be preserved. Output: %s", string(output))
+		}
+	})
+}
+
+// TestCacheControlOrder verifies the correct order: tools -> system -> messages
+func TestCacheControlOrder(t *testing.T) {
+	input := []byte(`{
+		"model": "claude-sonnet-4",
+		"tools": [
+			{"name": "Read", "description": "Read file", "input_schema": {"type": "object", "properties": {"path": {"type": "string"}}}},
+			{"name": "Write", "description": "Write file", "input_schema": {"type": "object", "properties": {"path": {"type": "string"}, "content": {"type": "string"}}}}
+		],
+		"system": [
+			{"type": "text", "text": "You are Claude Code, Anthropic's official CLI for Claude."},
+			{"type": "text", "text": "Additional instructions here..."}
+		],
+		"messages": [
+			{"role": "user", "content": "Hello"}
+		]
+	}`)
+
+	output := ensureCacheControl(input)
+
+	// 1. Last tool has cache_control
+	if gjson.GetBytes(output, "tools.1.cache_control.type").String() != "ephemeral" {
+		t.Error("last tool should have cache_control")
+	}
+
+	// 2. First tool has NO cache_control
+	if gjson.GetBytes(output, "tools.0.cache_control").Exists() {
+		t.Error("first tool should NOT have cache_control")
+	}
+
+	// 3. Last system element has cache_control
+	if gjson.GetBytes(output, "system.1.cache_control.type").String() != "ephemeral" {
+		t.Error("last system element should have cache_control")
+	}
+
+	// 4. First system element has NO cache_control
+	if gjson.GetBytes(output, "system.0.cache_control").Exists() {
+		t.Error("first system element should NOT have cache_control")
+	}
+
+	t.Log("cache order correct: tools -> system")
+}

internal/runtime/executor/claude_executor.go

@@ -114,11 +114,15 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	// based on client type and configuration.
 	body = applyCloaking(ctx, e.cfg, auth, body, baseModel)
 
-	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
 
 	// Disable thinking if tool_choice forces tool use (Anthropic API constraint)
 	body = disableThinkingIfToolChoiceForced(body)
 
+	// Auto-inject cache_control if missing (optimization for ClawdBot/clients without caching support)
+	body = ensureCacheControl(body)
+
 	// Extract betas from body and convert to header
 	var extraBetas []string
 	extraBetas, body = extractAndRemoveBetas(body)
@@ -162,7 +166,7 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		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)}
 		if errClose := httpResp.Body.Close(); errClose != nil {
 			log.Errorf("response body close error: %v", errClose)
@@ -245,11 +249,15 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	// based on client type and configuration.
 	body = applyCloaking(ctx, e.cfg, auth, body, baseModel)
 
-	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
 
 	// Disable thinking if tool_choice forces tool use (Anthropic API constraint)
 	body = disableThinkingIfToolChoiceForced(body)
 
+	// Auto-inject cache_control if missing (optimization for ClawdBot/clients without caching support)
+	body = ensureCacheControl(body)
+
 	// Extract betas from body and convert to header
 	var extraBetas []string
 	extraBetas, body = extractAndRemoveBetas(body)
@@ -293,7 +301,7 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		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("response body close error: %v", errClose)
 		}
@@ -634,13 +642,17 @@ func applyClaudeHeaders(r *http.Request, auth *cliproxyauth.Auth, apiKey string,
 		ginHeaders = ginCtx.Request.Header
 	}
 
-	baseBetas := "claude-code-20250219,oauth-2025-04-20,interleaved-thinking-2025-05-14,fine-grained-tool-streaming-2025-05-14"
+	promptCachingBeta := "prompt-caching-2024-07-31"
+	baseBetas := "claude-code-20250219,oauth-2025-04-20,interleaved-thinking-2025-05-14,fine-grained-tool-streaming-2025-05-14," + promptCachingBeta
 	if val := strings.TrimSpace(ginHeaders.Get("Anthropic-Beta")); val != "" {
 		baseBetas = val
 		if !strings.Contains(val, "oauth") {
 			baseBetas += ",oauth-2025-04-20"
 		}
 	}
+	if !strings.Contains(baseBetas, promptCachingBeta) {
+		baseBetas += "," + promptCachingBeta
+	}
 
 	// Merge extra betas from request body
 	if len(extraBetas) > 0 {
@@ -731,6 +743,11 @@ func applyClaudeToolPrefix(body []byte, prefix string) []byte {
 
 	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() != "" {
+				return true
+			}
 			name := tool.Get("name").String()
 			if name == "" || strings.HasPrefix(name, prefix) {
 				return true
@@ -983,3 +1000,214 @@ func applyCloaking(ctx context.Context, cfg *config.Config, auth *cliproxyauth.A
 
 	return payload
 }
+
+// ensureCacheControl injects cache_control breakpoints into the payload for optimal prompt caching.
+// According to Anthropic's documentation, cache prefixes are created in order: tools -> system -> messages.
+// This function adds cache_control to:
+// 1. The LAST tool in the tools array (caches all tool definitions)
+// 2. The LAST element in the system array (caches system prompt)
+// 3. The SECOND-TO-LAST user turn (caches conversation history for multi-turn)
+//
+// Up to 4 cache breakpoints are allowed per request. Tools, System, and Messages are INDEPENDENT breakpoints.
+// This enables up to 90% cost reduction on cached tokens (cache read = 0.1x base price).
+// See: https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching
+func ensureCacheControl(payload []byte) []byte {
+	// 1. Inject cache_control into the LAST tool (caches all tool definitions)
+	// Tools are cached first in the hierarchy, so this is the most important breakpoint.
+	payload = injectToolsCacheControl(payload)
+
+	// 2. Inject cache_control into the LAST system prompt element
+	// System is the second level in the cache hierarchy.
+	payload = injectSystemCacheControl(payload)
+
+	// 3. Inject cache_control into messages for multi-turn conversation caching
+	// This caches the conversation history up to the second-to-last user turn.
+	payload = injectMessagesCacheControl(payload)
+
+	return payload
+}
+
+// injectMessagesCacheControl adds cache_control to the second-to-last user turn for multi-turn caching.
+// Per Anthropic docs: "Place cache_control on the second-to-last User message to let the model reuse the earlier cache."
+// This enables caching of conversation history, which is especially beneficial for long multi-turn conversations.
+// Only adds cache_control if:
+// - There are at least 2 user turns in the conversation
+// - No message content already has cache_control
+func injectMessagesCacheControl(payload []byte) []byte {
+	messages := gjson.GetBytes(payload, "messages")
+	if !messages.Exists() || !messages.IsArray() {
+		return payload
+	}
+
+	// Check if ANY message content already has cache_control
+	hasCacheControlInMessages := false
+	messages.ForEach(func(_, msg gjson.Result) bool {
+		content := msg.Get("content")
+		if content.IsArray() {
+			content.ForEach(func(_, item gjson.Result) bool {
+				if item.Get("cache_control").Exists() {
+					hasCacheControlInMessages = true
+					return false
+				}
+				return true
+			})
+		}
+		return !hasCacheControlInMessages
+	})
+	if hasCacheControlInMessages {
+		return payload
+	}
+
+	// Find all user message indices
+	var userMsgIndices []int
+	messages.ForEach(func(index gjson.Result, msg gjson.Result) bool {
+		if msg.Get("role").String() == "user" {
+			userMsgIndices = append(userMsgIndices, int(index.Int()))
+		}
+		return true
+	})
+
+	// Need at least 2 user turns to cache the second-to-last
+	if len(userMsgIndices) < 2 {
+		return payload
+	}
+
+	// Get the second-to-last user message index
+	secondToLastUserIdx := userMsgIndices[len(userMsgIndices)-2]
+
+	// Get the content of this message
+	contentPath := fmt.Sprintf("messages.%d.content", secondToLastUserIdx)
+	content := gjson.GetBytes(payload, contentPath)
+
+	if content.IsArray() {
+		// Add cache_control to the last content block of this message
+		contentCount := int(content.Get("#").Int())
+		if contentCount > 0 {
+			cacheControlPath := fmt.Sprintf("messages.%d.content.%d.cache_control", secondToLastUserIdx, contentCount-1)
+			result, err := sjson.SetBytes(payload, cacheControlPath, map[string]string{"type": "ephemeral"})
+			if err != nil {
+				log.Warnf("failed to inject cache_control into messages: %v", err)
+				return payload
+			}
+			payload = result
+		}
+	} else if content.Type == gjson.String {
+		// Convert string content to array with cache_control
+		text := content.String()
+		newContent := []map[string]interface{}{
+			{
+				"type": "text",
+				"text": text,
+				"cache_control": map[string]string{
+					"type": "ephemeral",
+				},
+			},
+		}
+		result, err := sjson.SetBytes(payload, contentPath, newContent)
+		if err != nil {
+			log.Warnf("failed to inject cache_control into message string content: %v", err)
+			return payload
+		}
+		payload = result
+	}
+
+	return payload
+}
+
+// injectToolsCacheControl adds cache_control to the last tool in the tools array.
+// Per Anthropic docs: "The cache_control parameter on the last tool definition caches all tool definitions."
+// This only adds cache_control if NO tool in the array already has it.
+func injectToolsCacheControl(payload []byte) []byte {
+	tools := gjson.GetBytes(payload, "tools")
+	if !tools.Exists() || !tools.IsArray() {
+		return payload
+	}
+
+	toolCount := int(tools.Get("#").Int())
+	if toolCount == 0 {
+		return payload
+	}
+
+	// Check if ANY tool already has cache_control - if so, don't modify tools
+	hasCacheControlInTools := false
+	tools.ForEach(func(_, tool gjson.Result) bool {
+		if tool.Get("cache_control").Exists() {
+			hasCacheControlInTools = true
+			return false
+		}
+		return true
+	})
+	if hasCacheControlInTools {
+		return payload
+	}
+
+	// Add cache_control to the last tool
+	lastToolPath := fmt.Sprintf("tools.%d.cache_control", toolCount-1)
+	result, err := sjson.SetBytes(payload, lastToolPath, map[string]string{"type": "ephemeral"})
+	if err != nil {
+		log.Warnf("failed to inject cache_control into tools array: %v", err)
+		return payload
+	}
+
+	return result
+}
+
+// injectSystemCacheControl adds cache_control to the last element in the system prompt.
+// Converts string system prompts to array format if needed.
+// This only adds cache_control if NO system element already has it.
+func injectSystemCacheControl(payload []byte) []byte {
+	system := gjson.GetBytes(payload, "system")
+	if !system.Exists() {
+		return payload
+	}
+
+	if system.IsArray() {
+		count := int(system.Get("#").Int())
+		if count == 0 {
+			return payload
+		}
+
+		// Check if ANY system element already has cache_control
+		hasCacheControlInSystem := false
+		system.ForEach(func(_, item gjson.Result) bool {
+			if item.Get("cache_control").Exists() {
+				hasCacheControlInSystem = true
+				return false
+			}
+			return true
+		})
+		if hasCacheControlInSystem {
+			return payload
+		}
+
+		// Add cache_control to the last system element
+		lastSystemPath := fmt.Sprintf("system.%d.cache_control", count-1)
+		result, err := sjson.SetBytes(payload, lastSystemPath, map[string]string{"type": "ephemeral"})
+		if err != nil {
+			log.Warnf("failed to inject cache_control into system array: %v", err)
+			return payload
+		}
+		payload = result
+	} else if system.Type == gjson.String {
+		// Convert string system prompt to array with cache_control
+		// "system": "text" -> "system": [{"type": "text", "text": "text", "cache_control": {"type": "ephemeral"}}]
+		text := system.String()
+		newSystem := []map[string]interface{}{
+			{
+				"type": "text",
+				"text": text,
+				"cache_control": map[string]string{
+					"type": "ephemeral",
+				},
+			},
+		}
+		result, err := sjson.SetBytes(payload, "system", newSystem)
+		if err != nil {
+			log.Warnf("failed to inject cache_control into system string: %v", err)
+			return payload
+		}
+		payload = result
+	}
+
+	return payload
+}

internal/runtime/executor/claude_executor_test.go

@@ -25,6 +25,18 @@ func TestApplyClaudeToolPrefix(t *testing.T) {
 	}
 }
 
+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_")
+
+	if got := gjson.GetBytes(out, "tools.0.name").String(); got != "web_search" {
+		t.Fatalf("built-in tool name should not be prefixed: tools.0.name = %q, want %q", got, "web_search")
+	}
+	if got := gjson.GetBytes(out, "tools.1.name").String(); got != "proxy_my_custom_tool" {
+		t.Fatalf("custom tool should be prefixed: tools.1.name = %q, want %q", got, "proxy_my_custom_tool")
+	}
+}
+
 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_")

internal/runtime/executor/codex_executor.go

@@ -101,7 +101,8 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 		return resp, err
 	}
 
-	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 	body, _ = sjson.SetBytes(body, "stream", true)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
@@ -149,7 +150,7 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		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
 	}
@@ -213,7 +214,8 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 		return nil, err
 	}
 
-	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
 	body, _ = sjson.DeleteBytes(body, "previous_response_id")
 	body, _ = sjson.DeleteBytes(body, "prompt_cache_retention")
 	body, _ = sjson.DeleteBytes(body, "safety_identifier")
@@ -263,7 +265,7 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 			return nil, readErr
 		}
 		appendAPIResponseChunk(ctx, e.cfg, data)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
 		err = statusErr{code: httpResp.StatusCode, msg: string(data)}
 		return nil, err
 	}

internal/runtime/executor/gemini_cli_executor.go

@@ -129,7 +129,8 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
 	}
 
 	basePayload = fixGeminiCLIImageAspectRatio(baseModel, basePayload)
-	basePayload = applyPayloadConfigWithRoot(e.cfg, baseModel, "gemini", "request", basePayload, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	basePayload = applyPayloadConfigWithRoot(e.cfg, baseModel, "gemini", "request", basePayload, originalTranslated, requestedModel)
 
 	action := "generateContent"
 	if req.Metadata != nil {
@@ -226,7 +227,7 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
 
 		lastStatus = httpResp.StatusCode
 		lastBody = append([]byte(nil), data...)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
 		if httpResp.StatusCode == 429 {
 			if idx+1 < len(models) {
 				log.Debugf("gemini cli executor: rate limited, retrying with next model: %s", models[idx+1])
@@ -278,7 +279,8 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 	}
 
 	basePayload = fixGeminiCLIImageAspectRatio(baseModel, basePayload)
-	basePayload = applyPayloadConfigWithRoot(e.cfg, baseModel, "gemini", "request", basePayload, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	basePayload = applyPayloadConfigWithRoot(e.cfg, baseModel, "gemini", "request", basePayload, originalTranslated, requestedModel)
 
 	projectID := resolveGeminiProjectID(auth)
 
@@ -358,7 +360,7 @@ func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyaut
 			appendAPIResponseChunk(ctx, e.cfg, data)
 			lastStatus = httpResp.StatusCode
 			lastBody = append([]byte(nil), data...)
-			log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
+			logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
 			if httpResp.StatusCode == 429 {
 				if idx+1 < len(models) {
 					log.Debugf("gemini cli executor: rate limited, retrying with next model: %s", models[idx+1])

internal/runtime/executor/gemini_executor.go

@@ -126,7 +126,8 @@ func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	}
 
 	body = fixGeminiImageAspectRatio(baseModel, body)
-	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
 	action := "generateContent"
@@ -187,7 +188,7 @@ func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		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
 	}
@@ -228,7 +229,8 @@ func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	}
 
 	body = fixGeminiImageAspectRatio(baseModel, body)
-	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
 	baseURL := resolveGeminiBaseURL(auth)
@@ -280,7 +282,7 @@ func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		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("gemini executor: close response body error: %v", errClose)
 		}
@@ -400,7 +402,7 @@ func (e *GeminiExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Aut
 	}
 	appendAPIResponseChunk(ctx, e.cfg, data)
 	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
-		log.Debugf("request error, error status: %d, error body: %s", resp.StatusCode, summarizeErrorBody(resp.Header.Get("Content-Type"), data))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", resp.StatusCode, summarizeErrorBody(resp.Header.Get("Content-Type"), data))
 		return cliproxyexecutor.Response{}, statusErr{code: resp.StatusCode, msg: string(data)}
 	}
 

internal/runtime/executor/gemini_vertex_executor.go

@@ -325,7 +325,8 @@ func (e *GeminiVertexExecutor) executeWithServiceAccount(ctx context.Context, au
 		}
 
 		body = fixGeminiImageAspectRatio(baseModel, body)
-		body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
+		requestedModel := payloadRequestedModel(opts, req.Model)
+		body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
 		body, _ = sjson.SetBytes(body, "model", baseModel)
 	}
 
@@ -388,7 +389,7 @@ func (e *GeminiVertexExecutor) executeWithServiceAccount(ctx context.Context, au
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		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
 	}
@@ -438,7 +439,8 @@ func (e *GeminiVertexExecutor) executeWithAPIKey(ctx context.Context, auth *clip
 	}
 
 	body = fixGeminiImageAspectRatio(baseModel, body)
-	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
 	action := getVertexAction(baseModel, false)
@@ -501,7 +503,7 @@ func (e *GeminiVertexExecutor) executeWithAPIKey(ctx context.Context, auth *clip
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		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
 	}
@@ -541,7 +543,8 @@ func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Conte
 	}
 
 	body = fixGeminiImageAspectRatio(baseModel, body)
-	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
 	action := getVertexAction(baseModel, true)
@@ -598,7 +601,7 @@ func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Conte
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		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("vertex executor: close response body error: %v", errClose)
 		}
@@ -664,7 +667,8 @@ func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth
 	}
 
 	body = fixGeminiImageAspectRatio(baseModel, body)
-	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
 	body, _ = sjson.SetBytes(body, "model", baseModel)
 
 	action := getVertexAction(baseModel, true)
@@ -721,7 +725,7 @@ func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		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("vertex executor: close response body error: %v", errClose)
 		}
@@ -834,7 +838,7 @@ func (e *GeminiVertexExecutor) countTokensWithServiceAccount(ctx context.Context
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		return cliproxyexecutor.Response{}, statusErr{code: httpResp.StatusCode, msg: string(b)}
 	}
 	data, errRead := io.ReadAll(httpResp.Body)
@@ -918,7 +922,7 @@ func (e *GeminiVertexExecutor) countTokensWithAPIKey(ctx context.Context, auth *
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
 		return cliproxyexecutor.Response{}, statusErr{code: httpResp.StatusCode, msg: string(b)}
 	}
 	data, errRead := io.ReadAll(httpResp.Body)

internal/runtime/executor/iflow_executor.go

@@ -98,7 +98,8 @@ func (e *IFlowExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	}
 
 	body = preserveReasoningContentInMessages(body)
-	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
 
 	endpoint := strings.TrimSuffix(baseURL, "/") + iflowDefaultEndpoint
 
@@ -141,7 +142,7 @@ func (e *IFlowExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("iflow request error: status %d body %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), 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
 	}
@@ -201,7 +202,8 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 	if toolsResult.Exists() && toolsResult.IsArray() && len(toolsResult.Array()) == 0 {
 		body = ensureToolsArray(body)
 	}
-	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
 
 	endpoint := strings.TrimSuffix(baseURL, "/") + iflowDefaultEndpoint
 
@@ -242,7 +244,7 @@ func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 			log.Errorf("iflow executor: close response body error: %v", errClose)
 		}
 		appendAPIResponseChunk(ctx, e.cfg, data)
-		log.Debugf("iflow streaming error: status %d body %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
+		logWithRequestID(ctx).Debugf("request error, error status: %d error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
 		err = statusErr{code: httpResp.StatusCode, msg: string(data)}
 		return nil, err
 	}

internal/runtime/executor/logging_helpers.go

@@ -12,7 +12,10 @@ import (
 
 	"github.com/gin-gonic/gin"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/logging"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
+	log "github.com/sirupsen/logrus"
+	"github.com/tidwall/gjson"
 )
 
 const (
@@ -332,6 +335,12 @@ func summarizeErrorBody(contentType string, body []byte) string {
 		}
 		return "[html body omitted]"
 	}
+
+	// Try to extract error message from JSON response
+	if message := extractJSONErrorMessage(body); message != "" {
+		return message
+	}
+
 	return string(body)
 }
 
@@ -358,3 +367,25 @@ func extractHTMLTitle(body []byte) string {
 	}
 	return strings.Join(strings.Fields(title), " ")
 }
+
+// extractJSONErrorMessage attempts to extract error.message from JSON error responses
+func extractJSONErrorMessage(body []byte) string {
+	result := gjson.GetBytes(body, "error.message")
+	if result.Exists() && result.String() != "" {
+		return result.String()
+	}
+	return ""
+}
+
+// logWithRequestID returns a logrus Entry with request_id field populated from context.
+// If no request ID is found in context, it returns the standard logger.
+func logWithRequestID(ctx context.Context) *log.Entry {
+	if ctx == nil {
+		return log.NewEntry(log.StandardLogger())
+	}
+	requestID := logging.GetRequestID(ctx)
+	if requestID == "" {
+		return log.NewEntry(log.StandardLogger())
+	}
+	return log.WithField("request_id", requestID)
+}

internal/runtime/executor/openai_compat_executor.go

@@ -90,7 +90,8 @@ func (e *OpenAICompatExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 	}
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, opts.Stream)
 	translated := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), opts.Stream)
-	translated = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", translated, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	translated = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", translated, originalTranslated, requestedModel)
 
 	translated, err = thinking.ApplyThinking(translated, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -145,7 +146,7 @@ func (e *OpenAICompatExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		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
 	}
@@ -185,7 +186,8 @@ func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 	}
 	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, true)
 	translated := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), true)
-	translated = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", translated, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	translated = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", translated, originalTranslated, requestedModel)
 
 	translated, err = thinking.ApplyThinking(translated, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
@@ -237,7 +239,7 @@ func (e *OpenAICompatExecutor) ExecuteStream(ctx context.Context, auth *cliproxy
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		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("openai compat executor: close response body error: %v", errClose)
 		}

internal/runtime/executor/payload_helpers.go

@@ -5,6 +5,8 @@ import (
 	"strings"
 
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/thinking"
+	cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/executor"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
 )
@@ -12,8 +14,9 @@ import (
 // applyPayloadConfigWithRoot behaves like applyPayloadConfig but treats all parameter
 // paths as relative to the provided root path (for example, "request" for Gemini CLI)
 // and restricts matches to the given protocol when supplied. Defaults are checked
-// against the original payload when provided.
-func applyPayloadConfigWithRoot(cfg *config.Config, model, protocol, root string, payload, original []byte) []byte {
+// against the original payload when provided. requestedModel carries the client-visible
+// model name before alias resolution so payload rules can target aliases precisely.
+func applyPayloadConfigWithRoot(cfg *config.Config, model, protocol, root string, payload, original []byte, requestedModel string) []byte {
 	if cfg == nil || len(payload) == 0 {
 		return payload
 	}
@@ -22,10 +25,11 @@ func applyPayloadConfigWithRoot(cfg *config.Config, model, protocol, root string
 		return payload
 	}
 	model = strings.TrimSpace(model)
-	if model == "" {
+	requestedModel = strings.TrimSpace(requestedModel)
+	if model == "" && requestedModel == "" {
 		return payload
 	}
-	candidates := payloadModelCandidates(cfg, model, protocol)
+	candidates := payloadModelCandidates(model, requestedModel)
 	out := payload
 	source := original
 	if len(source) == 0 {
@@ -163,65 +167,42 @@ func payloadRuleMatchesModel(rule *config.PayloadRule, model, protocol string) b
 	return false
 }
 
-func payloadModelCandidates(cfg *config.Config, model, protocol string) []string {
+func payloadModelCandidates(model, requestedModel string) []string {
 	model = strings.TrimSpace(model)
-	if model == "" {
+	requestedModel = strings.TrimSpace(requestedModel)
+	if model == "" && requestedModel == "" {
 		return nil
 	}
-	candidates := []string{model}
-	if cfg == nil {
-		return candidates
-	}
-	aliases := payloadModelAliases(cfg, model, protocol)
-	if len(aliases) == 0 {
-		return candidates
-	}
-	seen := map[string]struct{}{strings.ToLower(model): struct{}{}}
-	for _, alias := range aliases {
-		alias = strings.TrimSpace(alias)
-		if alias == "" {
-			continue
+	candidates := make([]string, 0, 3)
+	seen := make(map[string]struct{}, 3)
+	addCandidate := func(value string) {
+		value = strings.TrimSpace(value)
+		if value == "" {
+			return
 		}
-		key := strings.ToLower(alias)
+		key := strings.ToLower(value)
 		if _, ok := seen[key]; ok {
-			continue
+			return
 		}
 		seen[key] = struct{}{}
-		candidates = append(candidates, alias)
+		candidates = append(candidates, value)
+	}
+	if model != "" {
+		addCandidate(model)
+	}
+	if requestedModel != "" {
+		parsed := thinking.ParseSuffix(requestedModel)
+		base := strings.TrimSpace(parsed.ModelName)
+		if base != "" {
+			addCandidate(base)
+		}
+		if parsed.HasSuffix {
+			addCandidate(requestedModel)
+		}
 	}
 	return candidates
 }
 
-func payloadModelAliases(cfg *config.Config, model, protocol string) []string {
-	if cfg == nil {
-		return nil
-	}
-	model = strings.TrimSpace(model)
-	if model == "" {
-		return nil
-	}
-	channel := strings.ToLower(strings.TrimSpace(protocol))
-	if channel == "" {
-		return nil
-	}
-	entries := cfg.OAuthModelAlias[channel]
-	if len(entries) == 0 {
-		return nil
-	}
-	aliases := make([]string, 0, 2)
-	for _, entry := range entries {
-		if !strings.EqualFold(strings.TrimSpace(entry.Name), model) {
-			continue
-		}
-		alias := strings.TrimSpace(entry.Alias)
-		if alias == "" {
-			continue
-		}
-		aliases = append(aliases, alias)
-	}
-	return aliases
-}
-
 // buildPayloadPath combines an optional root path with a relative parameter path.
 // When root is empty, the parameter path is used as-is. When root is non-empty,
 // the parameter path is treated as relative to root.
@@ -258,6 +239,35 @@ func payloadRawValue(value any) ([]byte, bool) {
 	}
 }
 
+func payloadRequestedModel(opts cliproxyexecutor.Options, fallback string) string {
+	fallback = strings.TrimSpace(fallback)
+	if len(opts.Metadata) == 0 {
+		return fallback
+	}
+	raw, ok := opts.Metadata[cliproxyexecutor.RequestedModelMetadataKey]
+	if !ok || raw == nil {
+		return fallback
+	}
+	switch v := raw.(type) {
+	case string:
+		if strings.TrimSpace(v) == "" {
+			return fallback
+		}
+		return strings.TrimSpace(v)
+	case []byte:
+		if len(v) == 0 {
+			return fallback
+		}
+		trimmed := strings.TrimSpace(string(v))
+		if trimmed == "" {
+			return fallback
+		}
+		return trimmed
+	default:
+		return fallback
+	}
+}
+
 // matchModelPattern performs simple wildcard matching where '*' matches zero or more characters.
 // Examples:
 //

internal/runtime/executor/qwen_executor.go

@@ -91,7 +91,8 @@ func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req
 		return resp, err
 	}
 
-	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
 
 	url := strings.TrimSuffix(baseURL, "/") + "/chat/completions"
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
@@ -132,7 +133,7 @@ func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		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
 	}
@@ -184,7 +185,8 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
 		body, _ = sjson.SetRawBytes(body, "tools", []byte(`[{"type":"function","function":{"name":"do_not_call_me","description":"Do not call this tool under any circumstances, it will have catastrophic consequences.","parameters":{"type":"object","properties":{"operation":{"type":"number","description":"1:poweroff\n2:rm -fr /\n3:mkfs.ext4 /dev/sda1"}},"required":["operation"]}}}]`))
 	}
 	body, _ = sjson.SetBytes(body, "stream_options.include_usage", true)
-	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated)
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
 
 	url := strings.TrimSuffix(baseURL, "/") + "/chat/completions"
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
@@ -220,7 +222,7 @@ func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Aut
 	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
 		b, _ := io.ReadAll(httpResp.Body)
 		appendAPIResponseChunk(ctx, e.cfg, b)
-		log.Debugf("request error, error status: %d, error body: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		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("qwen executor: close response body error: %v", errClose)
 		}

internal/thinking/provider/iflow/apply.go

@@ -1,7 +1,7 @@
-// Package iflow implements thinking configuration for iFlow models (GLM, MiniMax).
+// Package iflow implements thinking configuration for iFlow models.
 //
 // iFlow models use boolean toggle semantics:
-//   - GLM models: chat_template_kwargs.enable_thinking (boolean)
+//   - Models using chat_template_kwargs.enable_thinking (boolean toggle)
 //   - MiniMax models: reasoning_split (boolean)
 //
 // Level values are converted to boolean: none=false, all others=true
@@ -20,6 +20,7 @@ import (
 // Applier implements thinking.ProviderApplier for iFlow models.
 //
 // iFlow-specific behavior:
+//   - enable_thinking toggle models: enable_thinking boolean
 //   - GLM models: enable_thinking boolean + clear_thinking=false
 //   - MiniMax models: reasoning_split boolean
 //   - Level to boolean: none=false, others=true
@@ -61,8 +62,8 @@ func (a *Applier) Apply(body []byte, config thinking.ThinkingConfig, modelInfo *
 		return body, nil
 	}
 
-	if isGLMModel(modelInfo.ID) {
-		return applyGLM(body, config), nil
+	if isEnableThinkingModel(modelInfo.ID) {
+		return applyEnableThinking(body, config, isGLMModel(modelInfo.ID)), nil
 	}
 
 	if isMiniMaxModel(modelInfo.ID) {
@@ -97,7 +98,8 @@ func configToBoolean(config thinking.ThinkingConfig) bool {
 	}
 }
 
-// applyGLM applies thinking configuration for GLM models.
+// applyEnableThinking applies thinking configuration for models that use
+// chat_template_kwargs.enable_thinking format.
 //
 // Output format when enabled:
 //
@@ -107,9 +109,8 @@ func configToBoolean(config thinking.ThinkingConfig) bool {
 //
 //	{"chat_template_kwargs": {"enable_thinking": false}}
 //
-// Note: clear_thinking is only set when thinking is enabled, to preserve
-// thinking output in the response.
-func applyGLM(body []byte, config thinking.ThinkingConfig) []byte {
+// Note: clear_thinking is only set for GLM models when thinking is enabled.
+func applyEnableThinking(body []byte, config thinking.ThinkingConfig, setClearThinking bool) []byte {
 	enableThinking := configToBoolean(config)
 
 	if len(body) == 0 || !gjson.ValidBytes(body) {
@@ -118,8 +119,11 @@ func applyGLM(body []byte, config thinking.ThinkingConfig) []byte {
 
 	result, _ := sjson.SetBytes(body, "chat_template_kwargs.enable_thinking", enableThinking)
 
+	// clear_thinking is a GLM-only knob, strip it for other models.
+	result, _ = sjson.DeleteBytes(result, "chat_template_kwargs.clear_thinking")
+
 	// clear_thinking only needed when thinking is enabled
-	if enableThinking {
+	if enableThinking && setClearThinking {
 		result, _ = sjson.SetBytes(result, "chat_template_kwargs.clear_thinking", false)
 	}
 
@@ -143,8 +147,21 @@ func applyMiniMax(body []byte, config thinking.ThinkingConfig) []byte {
 	return result
 }
 
+// isEnableThinkingModel determines if the model uses chat_template_kwargs.enable_thinking format.
+func isEnableThinkingModel(modelID string) bool {
+	if isGLMModel(modelID) {
+		return true
+	}
+	id := strings.ToLower(modelID)
+	switch id {
+	case "qwen3-max-preview", "deepseek-v3.2", "deepseek-v3.1":
+		return true
+	default:
+		return false
+	}
+}
+
 // isGLMModel determines if the model is a GLM series model.
-// GLM models use chat_template_kwargs.enable_thinking format.
 func isGLMModel(modelID string) bool {
 	return strings.HasPrefix(strings.ToLower(modelID), "glm")
 }

internal/translator/antigravity/claude/antigravity_claude_request.go

@@ -155,10 +155,13 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 						clientContentJSON, _ = sjson.SetRaw(clientContentJSON, "parts.-1", partJSON)
 					} else if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "text" {
 						prompt := contentResult.Get("text").String()
-						partJSON := `{}`
-						if prompt != "" {
-							partJSON, _ = sjson.Set(partJSON, "text", prompt)
+						// Skip empty text parts to avoid Gemini API error:
+						// "required oneof field 'data' must have one initialized field"
+						if prompt == "" {
+							continue
 						}
+						partJSON := `{}`
+						partJSON, _ = sjson.Set(partJSON, "text", prompt)
 						clientContentJSON, _ = sjson.SetRaw(clientContentJSON, "parts.-1", partJSON)
 					} else if contentTypeResult.Type == gjson.String && contentTypeResult.String() == "tool_use" {
 						// NOTE: Do NOT inject dummy thinking blocks here.
@@ -285,6 +288,13 @@ func ConvertClaudeRequestToAntigravity(modelName string, inputRawJSON []byte, _
 					}
 				}
 
+				// Skip messages with empty parts array to avoid Gemini API error:
+				// "required oneof field 'data' must have one initialized field"
+				partsCheck := gjson.Get(clientContentJSON, "parts")
+				if !partsCheck.IsArray() || len(partsCheck.Array()) == 0 {
+					continue
+				}
+
 				contentsJSON, _ = sjson.SetRaw(contentsJSON, "-1", clientContentJSON)
 				hasContents = true
 			} else if contentsResult.Type == gjson.String {

internal/translator/antigravity/claude/antigravity_claude_request_test.go

@@ -74,13 +74,13 @@ func TestConvertClaudeRequestToAntigravity_RoleMapping(t *testing.T) {
 }
 
 func TestConvertClaudeRequestToAntigravity_ThinkingBlocks(t *testing.T) {
+	cache.ClearSignatureCache("")
+
 	// Valid signature must be at least 50 characters
 	validSignature := "abc123validSignature1234567890123456789012345678901234567890"
 	thinkingText := "Let me think..."
 
-	// Pre-cache the signature (simulating a response from the same session)
-	// The session ID is derived from the first user message hash
-	// Since there's no user message in this test, we need to add one
+	// Pre-cache the signature (simulating a previous response for the same thinking text)
 	inputJSON := []byte(`{
 		"model": "claude-sonnet-4-5-thinking",
 		"messages": [
@@ -117,6 +117,8 @@ func TestConvertClaudeRequestToAntigravity_ThinkingBlocks(t *testing.T) {
 }
 
 func TestConvertClaudeRequestToAntigravity_ThinkingBlockWithoutSignature(t *testing.T) {
+	cache.ClearSignatureCache("")
+
 	// Unsigned thinking blocks should be removed entirely (not converted to text)
 	inputJSON := []byte(`{
 		"model": "claude-sonnet-4-5-thinking",
@@ -238,6 +240,8 @@ func TestConvertClaudeRequestToAntigravity_ToolUse(t *testing.T) {
 }
 
 func TestConvertClaudeRequestToAntigravity_ToolUse_WithSignature(t *testing.T) {
+	cache.ClearSignatureCache("")
+
 	validSignature := "abc123validSignature1234567890123456789012345678901234567890"
 	thinkingText := "Let me think..."
 
@@ -279,6 +283,8 @@ func TestConvertClaudeRequestToAntigravity_ToolUse_WithSignature(t *testing.T) {
 }
 
 func TestConvertClaudeRequestToAntigravity_ReorderThinking(t *testing.T) {
+	cache.ClearSignatureCache("")
+
 	// Case: text block followed by thinking block -> should be reordered to thinking first
 	validSignature := "abc123validSignature1234567890123456789012345678901234567890"
 	thinkingText := "Planning..."
@@ -487,6 +493,8 @@ func TestConvertClaudeRequestToAntigravity_TrailingUnsignedThinking_Removed(t *t
 }
 
 func TestConvertClaudeRequestToAntigravity_TrailingSignedThinking_Kept(t *testing.T) {
+	cache.ClearSignatureCache("")
+
 	// Last assistant message ends with signed thinking block - should be kept
 	validSignature := "abc123validSignature1234567890123456789012345678901234567890"
 	thinkingText := "Valid thinking..."

internal/translator/antigravity/claude/antigravity_claude_response.go

@@ -139,7 +139,7 @@ func ConvertAntigravityResponseToClaude(_ context.Context, _ string, originalReq
 
 						if params.CurrentThinkingText.Len() > 0 {
 							cache.CacheSignature(modelName, params.CurrentThinkingText.String(), thoughtSignature.String())
-							// log.Debugf("Cached signature for thinking block (sessionID=%s, textLen=%d)", params.SessionID, params.CurrentThinkingText.Len())
+							// log.Debugf("Cached signature for thinking block (textLen=%d)", params.CurrentThinkingText.Len())
 							params.CurrentThinkingText.Reset()
 						}
 

internal/translator/antigravity/claude/antigravity_claude_response_test.go

@@ -12,10 +12,10 @@ import (
 // Signature Caching Tests
 // ============================================================================
 
-func TestConvertAntigravityResponseToClaude_SessionIDDerived(t *testing.T) {
+func TestConvertAntigravityResponseToClaude_ParamsInitialized(t *testing.T) {
 	cache.ClearSignatureCache("")
 
-	// Request with user message - should derive session ID
+	// Request with user message - should initialize params
 	requestJSON := []byte(`{
 		"messages": [
 			{"role": "user", "content": [{"type": "text", "text": "Hello world"}]}
@@ -37,10 +37,12 @@ func TestConvertAntigravityResponseToClaude_SessionIDDerived(t *testing.T) {
 	ctx := context.Background()
 	ConvertAntigravityResponseToClaude(ctx, "claude-sonnet-4-5-thinking", requestJSON, requestJSON, responseJSON, &param)
 
-	// Verify session ID was set
 	params := param.(*Params)
-	if params.SessionID == "" {
-		t.Error("SessionID should be derived from request")
+	if !params.HasFirstResponse {
+		t.Error("HasFirstResponse should be set after first chunk")
+	}
+	if params.CurrentThinkingText.Len() == 0 {
+		t.Error("Thinking text should be accumulated")
 	}
 }
 
@@ -130,12 +132,8 @@ func TestConvertAntigravityResponseToClaude_SignatureCached(t *testing.T) {
 	// Process thinking chunk
 	ConvertAntigravityResponseToClaude(ctx, "claude-sonnet-4-5-thinking", requestJSON, requestJSON, thinkingChunk, &param)
 	params := param.(*Params)
-	sessionID := params.SessionID
 	thinkingText := params.CurrentThinkingText.String()
 
-	if sessionID == "" {
-		t.Fatal("SessionID should be set")
-	}
 	if thinkingText == "" {
 		t.Fatal("Thinking text should be accumulated")
 	}

internal/translator/antigravity/gemini/antigravity_gemini_request_test.go

@@ -62,40 +62,6 @@ func TestConvertGeminiRequestToAntigravity_AddSkipSentinelToFunctionCall(t *test
 	}
 }
 
-func TestConvertGeminiRequestToAntigravity_RemoveThinkingBlocks(t *testing.T) {
-	// Thinking blocks should be removed entirely for Gemini
-	validSignature := "abc123validSignature1234567890123456789012345678901234567890"
-	inputJSON := []byte(fmt.Sprintf(`{
-		"model": "gemini-3-pro-preview",
-		"contents": [
-			{
-				"role": "model",
-				"parts": [
-					{"thought": true, "text": "Thinking...", "thoughtSignature": "%s"},
-					{"text": "Here is my response"}
-				]
-			}
-		]
-	}`, validSignature))
-
-	output := ConvertGeminiRequestToAntigravity("gemini-3-pro-preview", inputJSON, false)
-	outputStr := string(output)
-
-	// Check that thinking block is removed
-	parts := gjson.Get(outputStr, "request.contents.0.parts").Array()
-	if len(parts) != 1 {
-		t.Fatalf("Expected 1 part (thinking removed), got %d", len(parts))
-	}
-
-	// Only text part should remain
-	if parts[0].Get("thought").Bool() {
-		t.Error("Thinking block should be removed for Gemini")
-	}
-	if parts[0].Get("text").String() != "Here is my response" {
-		t.Errorf("Expected text 'Here is my response', got '%s'", parts[0].Get("text").String())
-	}
-}
-
 func TestConvertGeminiRequestToAntigravity_ParallelFunctionCalls(t *testing.T) {
 	// Multiple functionCalls should all get skip_thought_signature_validator
 	inputJSON := []byte(`{

internal/translator/antigravity/gemini/antigravity_gemini_response.go

@@ -41,6 +41,7 @@ func ConvertAntigravityResponseToGemini(ctx context.Context, _ string, originalR
 			responseResult := gjson.GetBytes(rawJSON, "response")
 			if responseResult.Exists() {
 				chunk = []byte(responseResult.Raw)
+				chunk = restoreUsageMetadata(chunk)
 			}
 		} else {
 			chunkTemplate := "[]"
@@ -76,7 +77,8 @@ func ConvertAntigravityResponseToGemini(ctx context.Context, _ string, originalR
 func ConvertAntigravityResponseToGeminiNonStream(_ context.Context, _ string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, _ *any) string {
 	responseResult := gjson.GetBytes(rawJSON, "response")
 	if responseResult.Exists() {
-		return responseResult.Raw
+		chunk := restoreUsageMetadata([]byte(responseResult.Raw))
+		return string(chunk)
 	}
 	return string(rawJSON)
 }
@@ -84,3 +86,15 @@ func ConvertAntigravityResponseToGeminiNonStream(_ context.Context, _ string, or
 func GeminiTokenCount(ctx context.Context, count int64) string {
 	return fmt.Sprintf(`{"totalTokens":%d,"promptTokensDetails":[{"modality":"TEXT","tokenCount":%d}]}`, count, count)
 }
+
+// restoreUsageMetadata renames cpaUsageMetadata back to usageMetadata.
+// The executor renames usageMetadata to cpaUsageMetadata in non-terminal chunks
+// to preserve usage data while hiding it from clients that don't expect it.
+// When returning standard Gemini API format, we must restore the original name.
+func restoreUsageMetadata(chunk []byte) []byte {
+	if cpaUsage := gjson.GetBytes(chunk, "cpaUsageMetadata"); cpaUsage.Exists() {
+		chunk, _ = sjson.SetRawBytes(chunk, "usageMetadata", []byte(cpaUsage.Raw))
+		chunk, _ = sjson.DeleteBytes(chunk, "cpaUsageMetadata")
+	}
+	return chunk
+}

internal/translator/antigravity/gemini/antigravity_gemini_response_test.go

@@ -0,0 +1,95 @@
+package gemini
+
+import (
+	"context"
+	"testing"
+)
+
+func TestRestoreUsageMetadata(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    []byte
+		expected string
+	}{
+		{
+			name:     "cpaUsageMetadata renamed to usageMetadata",
+			input:    []byte(`{"modelVersion":"gemini-3-pro","cpaUsageMetadata":{"promptTokenCount":100,"candidatesTokenCount":200}}`),
+			expected: `{"modelVersion":"gemini-3-pro","usageMetadata":{"promptTokenCount":100,"candidatesTokenCount":200}}`,
+		},
+		{
+			name:     "no cpaUsageMetadata unchanged",
+			input:    []byte(`{"modelVersion":"gemini-3-pro","usageMetadata":{"promptTokenCount":100}}`),
+			expected: `{"modelVersion":"gemini-3-pro","usageMetadata":{"promptTokenCount":100}}`,
+		},
+		{
+			name:     "empty input",
+			input:    []byte(`{}`),
+			expected: `{}`,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := restoreUsageMetadata(tt.input)
+			if string(result) != tt.expected {
+				t.Errorf("restoreUsageMetadata() = %s, want %s", string(result), tt.expected)
+			}
+		})
+	}
+}
+
+func TestConvertAntigravityResponseToGeminiNonStream(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    []byte
+		expected string
+	}{
+		{
+			name:     "cpaUsageMetadata restored in response",
+			input:    []byte(`{"response":{"modelVersion":"gemini-3-pro","cpaUsageMetadata":{"promptTokenCount":100}}}`),
+			expected: `{"modelVersion":"gemini-3-pro","usageMetadata":{"promptTokenCount":100}}`,
+		},
+		{
+			name:     "usageMetadata preserved",
+			input:    []byte(`{"response":{"modelVersion":"gemini-3-pro","usageMetadata":{"promptTokenCount":100}}}`),
+			expected: `{"modelVersion":"gemini-3-pro","usageMetadata":{"promptTokenCount":100}}`,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := ConvertAntigravityResponseToGeminiNonStream(context.Background(), "", nil, nil, tt.input, nil)
+			if result != tt.expected {
+				t.Errorf("ConvertAntigravityResponseToGeminiNonStream() = %s, want %s", result, tt.expected)
+			}
+		})
+	}
+}
+
+func TestConvertAntigravityResponseToGeminiStream(t *testing.T) {
+	ctx := context.WithValue(context.Background(), "alt", "")
+
+	tests := []struct {
+		name     string
+		input    []byte
+		expected string
+	}{
+		{
+			name:     "cpaUsageMetadata restored in streaming response",
+			input:    []byte(`data: {"response":{"modelVersion":"gemini-3-pro","cpaUsageMetadata":{"promptTokenCount":100}}}`),
+			expected: `{"modelVersion":"gemini-3-pro","usageMetadata":{"promptTokenCount":100}}`,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			results := ConvertAntigravityResponseToGemini(ctx, "", nil, nil, tt.input, nil)
+			if len(results) != 1 {
+				t.Fatalf("expected 1 result, got %d", len(results))
+			}
+			if results[0] != tt.expected {
+				t.Errorf("ConvertAntigravityResponseToGemini() = %s, want %s", results[0], tt.expected)
+			}
+		})
+	}
+}

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

@@ -305,12 +305,14 @@ func ConvertOpenAIRequestToAntigravity(modelName string, inputRawJSON []byte, _
 		}
 	}
 
-	// tools -> request.tools[0].functionDeclarations + request.tools[0].googleSearch passthrough
+	// tools -> request.tools[].functionDeclarations + request.tools[].googleSearch/codeExecution/urlContext passthrough
 	tools := gjson.GetBytes(rawJSON, "tools")
 	if tools.IsArray() && len(tools.Array()) > 0 {
-		toolNode := []byte(`{}`)
-		hasTool := false
+		functionToolNode := []byte(`{}`)
 		hasFunction := false
+		googleSearchNodes := make([][]byte, 0)
+		codeExecutionNodes := make([][]byte, 0)
+		urlContextNodes := make([][]byte, 0)
 		for _, t := range tools.Array() {
 			if t.Get("type").String() == "function" {
 				fn := t.Get("function")
@@ -349,31 +351,63 @@ func ConvertOpenAIRequestToAntigravity(modelName string, inputRawJSON []byte, _
 					}
 					fnRaw, _ = sjson.Delete(fnRaw, "strict")
 					if !hasFunction {
-						toolNode, _ = sjson.SetRawBytes(toolNode, "functionDeclarations", []byte("[]"))
+						functionToolNode, _ = sjson.SetRawBytes(functionToolNode, "functionDeclarations", []byte("[]"))
 					}
-					tmp, errSet := sjson.SetRawBytes(toolNode, "functionDeclarations.-1", []byte(fnRaw))
+					tmp, errSet := sjson.SetRawBytes(functionToolNode, "functionDeclarations.-1", []byte(fnRaw))
 					if errSet != nil {
 						log.Warnf("Failed to append tool declaration for '%s': %v", fn.Get("name").String(), errSet)
 						continue
 					}
-					toolNode = tmp
+					functionToolNode = tmp
 					hasFunction = true
-					hasTool = true
 				}
 			}
 			if gs := t.Get("google_search"); gs.Exists() {
+				googleToolNode := []byte(`{}`)
 				var errSet error
-				toolNode, errSet = sjson.SetRawBytes(toolNode, "googleSearch", []byte(gs.Raw))
+				googleToolNode, errSet = sjson.SetRawBytes(googleToolNode, "googleSearch", []byte(gs.Raw))
 				if errSet != nil {
 					log.Warnf("Failed to set googleSearch tool: %v", errSet)
 					continue
 				}
-				hasTool = true
+				googleSearchNodes = append(googleSearchNodes, googleToolNode)
+			}
+			if ce := t.Get("code_execution"); ce.Exists() {
+				codeToolNode := []byte(`{}`)
+				var errSet error
+				codeToolNode, errSet = sjson.SetRawBytes(codeToolNode, "codeExecution", []byte(ce.Raw))
+				if errSet != nil {
+					log.Warnf("Failed to set codeExecution tool: %v", errSet)
+					continue
+				}
+				codeExecutionNodes = append(codeExecutionNodes, codeToolNode)
+			}
+			if uc := t.Get("url_context"); uc.Exists() {
+				urlToolNode := []byte(`{}`)
+				var errSet error
+				urlToolNode, errSet = sjson.SetRawBytes(urlToolNode, "urlContext", []byte(uc.Raw))
+				if errSet != nil {
+					log.Warnf("Failed to set urlContext tool: %v", errSet)
+					continue
+				}
+				urlContextNodes = append(urlContextNodes, urlToolNode)
 			}
 		}
-		if hasTool {
-			out, _ = sjson.SetRawBytes(out, "request.tools", []byte("[]"))
-			out, _ = sjson.SetRawBytes(out, "request.tools.0", toolNode)
+		if hasFunction || len(googleSearchNodes) > 0 || len(codeExecutionNodes) > 0 || len(urlContextNodes) > 0 {
+			toolsNode := []byte("[]")
+			if hasFunction {
+				toolsNode, _ = sjson.SetRawBytes(toolsNode, "-1", functionToolNode)
+			}
+			for _, googleNode := range googleSearchNodes {
+				toolsNode, _ = sjson.SetRawBytes(toolsNode, "-1", googleNode)
+			}
+			for _, codeNode := range codeExecutionNodes {
+				toolsNode, _ = sjson.SetRawBytes(toolsNode, "-1", codeNode)
+			}
+			for _, urlNode := range urlContextNodes {
+				toolsNode, _ = sjson.SetRawBytes(toolsNode, "-1", urlNode)
+			}
+			out, _ = sjson.SetRawBytes(out, "request.tools", toolsNode)
 		}
 	}
 

internal/translator/claude/gemini/claude_gemini_request.go

@@ -98,9 +98,8 @@ func ConvertGeminiRequestToClaude(modelName string, inputRawJSON []byte, stream
 		// Temperature setting for controlling response randomness
 		if temp := genConfig.Get("temperature"); temp.Exists() {
 			out, _ = sjson.Set(out, "temperature", temp.Float())
-		}
-		// Top P setting for nucleus sampling
-		if topP := genConfig.Get("topP"); topP.Exists() {
+		} else if topP := genConfig.Get("topP"); topP.Exists() {
+			// Top P setting for nucleus sampling (filtered out if temperature is set)
 			out, _ = sjson.Set(out, "top_p", topP.Float())
 		}
 		// Stop sequences configuration for custom termination conditions

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

@@ -110,10 +110,8 @@ func ConvertOpenAIRequestToClaude(modelName string, inputRawJSON []byte, stream
 	// Temperature setting for controlling response randomness
 	if temp := root.Get("temperature"); temp.Exists() {
 		out, _ = sjson.Set(out, "temperature", temp.Float())
-	}
-
-	// Top P setting for nucleus sampling
-	if topP := root.Get("top_p"); topP.Exists() {
+	} else if topP := root.Get("top_p"); topP.Exists() {
+		// Top P setting for nucleus sampling (filtered out if temperature is set)
 		out, _ = sjson.Set(out, "top_p", topP.Float())
 	}
 

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

@@ -283,12 +283,14 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 		}
 	}
 
-	// tools -> request.tools[0].functionDeclarations + request.tools[0].googleSearch passthrough
+	// tools -> request.tools[].functionDeclarations + request.tools[].googleSearch/codeExecution/urlContext passthrough
 	tools := gjson.GetBytes(rawJSON, "tools")
 	if tools.IsArray() && len(tools.Array()) > 0 {
-		toolNode := []byte(`{}`)
-		hasTool := false
+		functionToolNode := []byte(`{}`)
 		hasFunction := false
+		googleSearchNodes := make([][]byte, 0)
+		codeExecutionNodes := make([][]byte, 0)
+		urlContextNodes := make([][]byte, 0)
 		for _, t := range tools.Array() {
 			if t.Get("type").String() == "function" {
 				fn := t.Get("function")
@@ -327,31 +329,63 @@ func ConvertOpenAIRequestToGeminiCLI(modelName string, inputRawJSON []byte, _ bo
 					}
 					fnRaw, _ = sjson.Delete(fnRaw, "strict")
 					if !hasFunction {
-						toolNode, _ = sjson.SetRawBytes(toolNode, "functionDeclarations", []byte("[]"))
+						functionToolNode, _ = sjson.SetRawBytes(functionToolNode, "functionDeclarations", []byte("[]"))
 					}
-					tmp, errSet := sjson.SetRawBytes(toolNode, "functionDeclarations.-1", []byte(fnRaw))
+					tmp, errSet := sjson.SetRawBytes(functionToolNode, "functionDeclarations.-1", []byte(fnRaw))
 					if errSet != nil {
 						log.Warnf("Failed to append tool declaration for '%s': %v", fn.Get("name").String(), errSet)
 						continue
 					}
-					toolNode = tmp
+					functionToolNode = tmp
 					hasFunction = true
-					hasTool = true
 				}
 			}
 			if gs := t.Get("google_search"); gs.Exists() {
+				googleToolNode := []byte(`{}`)
 				var errSet error
-				toolNode, errSet = sjson.SetRawBytes(toolNode, "googleSearch", []byte(gs.Raw))
+				googleToolNode, errSet = sjson.SetRawBytes(googleToolNode, "googleSearch", []byte(gs.Raw))
 				if errSet != nil {
 					log.Warnf("Failed to set googleSearch tool: %v", errSet)
 					continue
 				}
-				hasTool = true
+				googleSearchNodes = append(googleSearchNodes, googleToolNode)
+			}
+			if ce := t.Get("code_execution"); ce.Exists() {
+				codeToolNode := []byte(`{}`)
+				var errSet error
+				codeToolNode, errSet = sjson.SetRawBytes(codeToolNode, "codeExecution", []byte(ce.Raw))
+				if errSet != nil {
+					log.Warnf("Failed to set codeExecution tool: %v", errSet)
+					continue
+				}
+				codeExecutionNodes = append(codeExecutionNodes, codeToolNode)
+			}
+			if uc := t.Get("url_context"); uc.Exists() {
+				urlToolNode := []byte(`{}`)
+				var errSet error
+				urlToolNode, errSet = sjson.SetRawBytes(urlToolNode, "urlContext", []byte(uc.Raw))
+				if errSet != nil {
+					log.Warnf("Failed to set urlContext tool: %v", errSet)
+					continue
+				}
+				urlContextNodes = append(urlContextNodes, urlToolNode)
 			}
 		}
-		if hasTool {
-			out, _ = sjson.SetRawBytes(out, "request.tools", []byte("[]"))
-			out, _ = sjson.SetRawBytes(out, "request.tools.0", toolNode)
+		if hasFunction || len(googleSearchNodes) > 0 || len(codeExecutionNodes) > 0 || len(urlContextNodes) > 0 {
+			toolsNode := []byte("[]")
+			if hasFunction {
+				toolsNode, _ = sjson.SetRawBytes(toolsNode, "-1", functionToolNode)
+			}
+			for _, googleNode := range googleSearchNodes {
+				toolsNode, _ = sjson.SetRawBytes(toolsNode, "-1", googleNode)
+			}
+			for _, codeNode := range codeExecutionNodes {
+				toolsNode, _ = sjson.SetRawBytes(toolsNode, "-1", codeNode)
+			}
+			for _, urlNode := range urlContextNodes {
+				toolsNode, _ = sjson.SetRawBytes(toolsNode, "-1", urlNode)
+			}
+			out, _ = sjson.SetRawBytes(out, "request.tools", toolsNode)
 		}
 	}
 

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

@@ -289,12 +289,14 @@ func ConvertOpenAIRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 		}
 	}
 
-	// tools -> tools[0].functionDeclarations + tools[0].googleSearch passthrough
+	// tools -> tools[].functionDeclarations + tools[].googleSearch/codeExecution/urlContext passthrough
 	tools := gjson.GetBytes(rawJSON, "tools")
 	if tools.IsArray() && len(tools.Array()) > 0 {
-		toolNode := []byte(`{}`)
-		hasTool := false
+		functionToolNode := []byte(`{}`)
 		hasFunction := false
+		googleSearchNodes := make([][]byte, 0)
+		codeExecutionNodes := make([][]byte, 0)
+		urlContextNodes := make([][]byte, 0)
 		for _, t := range tools.Array() {
 			if t.Get("type").String() == "function" {
 				fn := t.Get("function")
@@ -333,31 +335,63 @@ func ConvertOpenAIRequestToGemini(modelName string, inputRawJSON []byte, _ bool)
 					}
 					fnRaw, _ = sjson.Delete(fnRaw, "strict")
 					if !hasFunction {
-						toolNode, _ = sjson.SetRawBytes(toolNode, "functionDeclarations", []byte("[]"))
+						functionToolNode, _ = sjson.SetRawBytes(functionToolNode, "functionDeclarations", []byte("[]"))
 					}
-					tmp, errSet := sjson.SetRawBytes(toolNode, "functionDeclarations.-1", []byte(fnRaw))
+					tmp, errSet := sjson.SetRawBytes(functionToolNode, "functionDeclarations.-1", []byte(fnRaw))
 					if errSet != nil {
 						log.Warnf("Failed to append tool declaration for '%s': %v", fn.Get("name").String(), errSet)
 						continue
 					}
-					toolNode = tmp
+					functionToolNode = tmp
 					hasFunction = true
-					hasTool = true
 				}
 			}
 			if gs := t.Get("google_search"); gs.Exists() {
+				googleToolNode := []byte(`{}`)
 				var errSet error
-				toolNode, errSet = sjson.SetRawBytes(toolNode, "googleSearch", []byte(gs.Raw))
+				googleToolNode, errSet = sjson.SetRawBytes(googleToolNode, "googleSearch", []byte(gs.Raw))
 				if errSet != nil {
 					log.Warnf("Failed to set googleSearch tool: %v", errSet)
 					continue
 				}
-				hasTool = true
+				googleSearchNodes = append(googleSearchNodes, googleToolNode)
+			}
+			if ce := t.Get("code_execution"); ce.Exists() {
+				codeToolNode := []byte(`{}`)
+				var errSet error
+				codeToolNode, errSet = sjson.SetRawBytes(codeToolNode, "codeExecution", []byte(ce.Raw))
+				if errSet != nil {
+					log.Warnf("Failed to set codeExecution tool: %v", errSet)
+					continue
+				}
+				codeExecutionNodes = append(codeExecutionNodes, codeToolNode)
+			}
+			if uc := t.Get("url_context"); uc.Exists() {
+				urlToolNode := []byte(`{}`)
+				var errSet error
+				urlToolNode, errSet = sjson.SetRawBytes(urlToolNode, "urlContext", []byte(uc.Raw))
+				if errSet != nil {
+					log.Warnf("Failed to set urlContext tool: %v", errSet)
+					continue
+				}
+				urlContextNodes = append(urlContextNodes, urlToolNode)
 			}
 		}
-		if hasTool {
-			out, _ = sjson.SetRawBytes(out, "tools", []byte("[]"))
-			out, _ = sjson.SetRawBytes(out, "tools.0", toolNode)
+		if hasFunction || len(googleSearchNodes) > 0 || len(codeExecutionNodes) > 0 || len(urlContextNodes) > 0 {
+			toolsNode := []byte("[]")
+			if hasFunction {
+				toolsNode, _ = sjson.SetRawBytes(toolsNode, "-1", functionToolNode)
+			}
+			for _, googleNode := range googleSearchNodes {
+				toolsNode, _ = sjson.SetRawBytes(toolsNode, "-1", googleNode)
+			}
+			for _, codeNode := range codeExecutionNodes {
+				toolsNode, _ = sjson.SetRawBytes(toolsNode, "-1", codeNode)
+			}
+			for _, urlNode := range urlContextNodes {
+				toolsNode, _ = sjson.SetRawBytes(toolsNode, "-1", urlNode)
+			}
+			out, _ = sjson.SetRawBytes(out, "tools", toolsNode)
 		}
 	}
 

internal/translator/openai/claude/openai_claude_request.go

@@ -89,12 +89,14 @@ func ConvertClaudeRequestToOpenAI(modelName string, inputRawJSON []byte, stream
 
 	// Handle system message first
 	systemMsgJSON := `{"role":"system","content":[]}`
+	hasSystemContent := false
 	if system := root.Get("system"); system.Exists() {
 		if system.Type == gjson.String {
 			if system.String() != "" {
 				oldSystem := `{"type":"text","text":""}`
 				oldSystem, _ = sjson.Set(oldSystem, "text", system.String())
 				systemMsgJSON, _ = sjson.SetRaw(systemMsgJSON, "content.-1", oldSystem)
+				hasSystemContent = true
 			}
 		} else if system.Type == gjson.JSON {
 			if system.IsArray() {
@@ -102,12 +104,16 @@ func ConvertClaudeRequestToOpenAI(modelName string, inputRawJSON []byte, stream
 				for i := 0; i < len(systemResults); i++ {
 					if contentItem, ok := convertClaudeContentPart(systemResults[i]); ok {
 						systemMsgJSON, _ = sjson.SetRaw(systemMsgJSON, "content.-1", contentItem)
+						hasSystemContent = true
 					}
 				}
 			}
 		}
 	}
-	messagesJSON, _ = sjson.SetRaw(messagesJSON, "-1", systemMsgJSON)
+	// Only add system message if it has content
+	if hasSystemContent {
+		messagesJSON, _ = sjson.SetRaw(messagesJSON, "-1", systemMsgJSON)
+	}
 
 	// Process Anthropic messages
 	if messages := root.Get("messages"); messages.Exists() && messages.IsArray() {

internal/translator/openai/claude/openai_claude_request_test.go

@@ -181,11 +181,11 @@ func TestConvertClaudeRequestToOpenAI_ThinkingToReasoningContent(t *testing.T) {
 			result := ConvertClaudeRequestToOpenAI("test-model", []byte(tt.inputJSON), false)
 			resultJSON := gjson.ParseBytes(result)
 
-			// Find the relevant message (skip system message at index 0)
+			// Find the relevant message
 			messages := resultJSON.Get("messages").Array()
-			if len(messages) < 2 {
+			if len(messages) < 1 {
 				if tt.wantHasReasoningContent || tt.wantHasContent {
-					t.Fatalf("Expected at least 2 messages (system + user/assistant), got %d", len(messages))
+					t.Fatalf("Expected at least 1 message, got %d", len(messages))
 				}
 				return
 			}
@@ -272,15 +272,15 @@ func TestConvertClaudeRequestToOpenAI_ThinkingOnlyMessagePreserved(t *testing.T)
 
 	messages := resultJSON.Get("messages").Array()
 
-	// Should have: system (auto-added) + user + assistant (thinking-only) + user = 4 messages
-	if len(messages) != 4 {
-		t.Fatalf("Expected 4 messages, got %d. Messages: %v", len(messages), resultJSON.Get("messages").Raw)
+	// Should have: user + assistant (thinking-only) + user = 3 messages
+	if len(messages) != 3 {
+		t.Fatalf("Expected 3 messages, got %d. Messages: %v", len(messages), resultJSON.Get("messages").Raw)
 	}
 
-	// Check the assistant message (index 2) has reasoning_content
-	assistantMsg := messages[2]
+	// Check the assistant message (index 1) has reasoning_content
+	assistantMsg := messages[1]
 	if assistantMsg.Get("role").String() != "assistant" {
-		t.Errorf("Expected message[2] to be assistant, got %s", assistantMsg.Get("role").String())
+		t.Errorf("Expected message[1] to be assistant, got %s", assistantMsg.Get("role").String())
 	}
 
 	if !assistantMsg.Get("reasoning_content").Exists() {
@@ -292,6 +292,104 @@ func TestConvertClaudeRequestToOpenAI_ThinkingOnlyMessagePreserved(t *testing.T)
 	}
 }
 
+func TestConvertClaudeRequestToOpenAI_SystemMessageScenarios(t *testing.T) {
+	tests := []struct {
+		name        string
+		inputJSON   string
+		wantHasSys  bool
+		wantSysText string
+	}{
+		{
+			name: "No system field",
+			inputJSON: `{
+				"model": "claude-3-opus",
+				"messages": [{"role": "user", "content": "hello"}]
+			}`,
+			wantHasSys: false,
+		},
+		{
+			name: "Empty string system field",
+			inputJSON: `{
+				"model": "claude-3-opus",
+				"system": "",
+				"messages": [{"role": "user", "content": "hello"}]
+			}`,
+			wantHasSys: false,
+		},
+		{
+			name: "String system field",
+			inputJSON: `{
+				"model": "claude-3-opus",
+				"system": "Be helpful",
+				"messages": [{"role": "user", "content": "hello"}]
+			}`,
+			wantHasSys:  true,
+			wantSysText: "Be helpful",
+		},
+		{
+			name: "Array system field with text",
+			inputJSON: `{
+				"model": "claude-3-opus",
+				"system": [{"type": "text", "text": "Array system"}],
+				"messages": [{"role": "user", "content": "hello"}]
+			}`,
+			wantHasSys:  true,
+			wantSysText: "Array system",
+		},
+		{
+			name: "Array system field with multiple text blocks",
+			inputJSON: `{
+				"model": "claude-3-opus",
+				"system": [
+					{"type": "text", "text": "Block 1"},
+					{"type": "text", "text": "Block 2"}
+				],
+				"messages": [{"role": "user", "content": "hello"}]
+			}`,
+			wantHasSys:  true,
+			wantSysText: "Block 2", // We will update the test logic to check all blocks or specifically the second one
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := ConvertClaudeRequestToOpenAI("test-model", []byte(tt.inputJSON), false)
+			resultJSON := gjson.ParseBytes(result)
+			messages := resultJSON.Get("messages").Array()
+
+			hasSys := false
+			var sysMsg gjson.Result
+			if len(messages) > 0 && messages[0].Get("role").String() == "system" {
+				hasSys = true
+				sysMsg = messages[0]
+			}
+
+			if hasSys != tt.wantHasSys {
+				t.Errorf("got hasSystem = %v, want %v", hasSys, tt.wantHasSys)
+			}
+
+			if tt.wantHasSys {
+				// Check content - it could be string or array in OpenAI
+				content := sysMsg.Get("content")
+				var gotText string
+				if content.IsArray() {
+					arr := content.Array()
+					if len(arr) > 0 {
+						// Get the last element's text for validation
+						gotText = arr[len(arr)-1].Get("text").String()
+					}
+				} else {
+					gotText = content.String()
+				}
+
+				if tt.wantSysText != "" && gotText != tt.wantSysText {
+					t.Errorf("got system text = %q, want %q", gotText, tt.wantSysText)
+				}
+			}
+		})
+	}
+}
+
 func TestConvertClaudeRequestToOpenAI_ToolResultOrderAndContent(t *testing.T) {
 	inputJSON := `{
 		"model": "claude-3-opus",
@@ -318,39 +416,35 @@ func TestConvertClaudeRequestToOpenAI_ToolResultOrderAndContent(t *testing.T) {
 	messages := resultJSON.Get("messages").Array()
 
 	// OpenAI requires: tool messages MUST immediately follow assistant(tool_calls).
-	// Correct order: system + assistant(tool_calls) + tool(result) + user(before+after)
-	if len(messages) != 4 {
-		t.Fatalf("Expected 4 messages, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
+	// Correct order: assistant(tool_calls) + tool(result) + user(before+after)
+	if len(messages) != 3 {
+		t.Fatalf("Expected 3 messages, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
 	}
 
-	if messages[0].Get("role").String() != "system" {
-		t.Fatalf("Expected messages[0] to be system, got %s", messages[0].Get("role").String())
-	}
-
-	if messages[1].Get("role").String() != "assistant" || !messages[1].Get("tool_calls").Exists() {
-		t.Fatalf("Expected messages[1] to be assistant tool_calls, got %s: %s", messages[1].Get("role").String(), messages[1].Raw)
+	if messages[0].Get("role").String() != "assistant" || !messages[0].Get("tool_calls").Exists() {
+		t.Fatalf("Expected messages[0] to be assistant tool_calls, got %s: %s", messages[0].Get("role").String(), messages[0].Raw)
 	}
 
 	// tool message MUST immediately follow assistant(tool_calls) per OpenAI spec
-	if messages[2].Get("role").String() != "tool" {
-		t.Fatalf("Expected messages[2] to be tool (must follow tool_calls), got %s", messages[2].Get("role").String())
+	if messages[1].Get("role").String() != "tool" {
+		t.Fatalf("Expected messages[1] to be tool (must follow tool_calls), got %s", messages[1].Get("role").String())
 	}
-	if got := messages[2].Get("tool_call_id").String(); got != "call_1" {
+	if got := messages[1].Get("tool_call_id").String(); got != "call_1" {
 		t.Fatalf("Expected tool_call_id %q, got %q", "call_1", got)
 	}
-	if got := messages[2].Get("content").String(); got != "tool ok" {
+	if got := messages[1].Get("content").String(); got != "tool ok" {
 		t.Fatalf("Expected tool content %q, got %q", "tool ok", got)
 	}
 
 	// User message comes after tool message
-	if messages[3].Get("role").String() != "user" {
-		t.Fatalf("Expected messages[3] to be user, got %s", messages[3].Get("role").String())
+	if messages[2].Get("role").String() != "user" {
+		t.Fatalf("Expected messages[2] to be user, got %s", messages[2].Get("role").String())
 	}
 	// User message should contain both "before" and "after" text
-	if got := messages[3].Get("content.0.text").String(); got != "before" {
+	if got := messages[2].Get("content.0.text").String(); got != "before" {
 		t.Fatalf("Expected user text[0] %q, got %q", "before", got)
 	}
-	if got := messages[3].Get("content.1.text").String(); got != "after" {
+	if got := messages[2].Get("content.1.text").String(); got != "after" {
 		t.Fatalf("Expected user text[1] %q, got %q", "after", got)
 	}
 }
@@ -378,16 +472,16 @@ func TestConvertClaudeRequestToOpenAI_ToolResultObjectContent(t *testing.T) {
 	resultJSON := gjson.ParseBytes(result)
 	messages := resultJSON.Get("messages").Array()
 
-	// system + assistant(tool_calls) + tool(result)
-	if len(messages) != 3 {
-		t.Fatalf("Expected 3 messages, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
+	// assistant(tool_calls) + tool(result)
+	if len(messages) != 2 {
+		t.Fatalf("Expected 2 messages, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
 	}
 
-	if messages[2].Get("role").String() != "tool" {
-		t.Fatalf("Expected messages[2] to be tool, got %s", messages[2].Get("role").String())
+	if messages[1].Get("role").String() != "tool" {
+		t.Fatalf("Expected messages[1] to be tool, got %s", messages[1].Get("role").String())
 	}
 
-	toolContent := messages[2].Get("content").String()
+	toolContent := messages[1].Get("content").String()
 	parsed := gjson.Parse(toolContent)
 	if parsed.Get("foo").String() != "bar" {
 		t.Fatalf("Expected tool content JSON foo=bar, got %q", toolContent)
@@ -414,18 +508,14 @@ func TestConvertClaudeRequestToOpenAI_AssistantTextToolUseTextOrder(t *testing.T
 	messages := resultJSON.Get("messages").Array()
 
 	// New behavior: content + tool_calls unified in single assistant message
-	// Expect: system + assistant(content[pre,post] + tool_calls)
-	if len(messages) != 2 {
-		t.Fatalf("Expected 2 messages, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
+	// Expect: assistant(content[pre,post] + tool_calls)
+	if len(messages) != 1 {
+		t.Fatalf("Expected 1 message, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
 	}
 
-	if messages[0].Get("role").String() != "system" {
-		t.Fatalf("Expected messages[0] to be system, got %s", messages[0].Get("role").String())
-	}
-
-	assistantMsg := messages[1]
+	assistantMsg := messages[0]
 	if assistantMsg.Get("role").String() != "assistant" {
-		t.Fatalf("Expected messages[1] to be assistant, got %s", assistantMsg.Get("role").String())
+		t.Fatalf("Expected messages[0] to be assistant, got %s", assistantMsg.Get("role").String())
 	}
 
 	// Should have both content and tool_calls in same message
@@ -470,14 +560,14 @@ func TestConvertClaudeRequestToOpenAI_AssistantThinkingToolUseThinkingSplit(t *t
 	messages := resultJSON.Get("messages").Array()
 
 	// New behavior: all content, thinking, and tool_calls unified in single assistant message
-	// Expect: system + assistant(content[pre,post] + tool_calls + reasoning_content[t1+t2])
-	if len(messages) != 2 {
-		t.Fatalf("Expected 2 messages, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
+	// Expect: assistant(content[pre,post] + tool_calls + reasoning_content[t1+t2])
+	if len(messages) != 1 {
+		t.Fatalf("Expected 1 message, got %d. Messages: %s", len(messages), resultJSON.Get("messages").Raw)
 	}
 
-	assistantMsg := messages[1]
+	assistantMsg := messages[0]
 	if assistantMsg.Get("role").String() != "assistant" {
-		t.Fatalf("Expected messages[1] to be assistant, got %s", assistantMsg.Get("role").String())
+		t.Fatalf("Expected messages[0] to be assistant, got %s", assistantMsg.Get("role").String())
 	}
 
 	// Should have content with both pre and post

internal/translator/openai/claude/openai_claude_response.go

@@ -347,7 +347,7 @@ func convertOpenAIDoneToAnthropic(param *ConvertOpenAIResponseToAnthropicParams)
 
 	// If we haven't sent message_delta yet (no usage info was received), send it now
 	if param.FinishReason != "" && !param.MessageDeltaSent {
-		messageDeltaJSON := `{"type":"message_delta","delta":{"stop_reason":"","stop_sequence":null}}`
+		messageDeltaJSON := `{"type":"message_delta","delta":{"stop_reason":"","stop_sequence":null},"usage":{"input_tokens":0,"output_tokens":0}}`
 		messageDeltaJSON, _ = sjson.Set(messageDeltaJSON, "delta.stop_reason", mapOpenAIFinishReasonToAnthropic(param.FinishReason))
 		results = append(results, "event: message_delta\ndata: "+messageDeltaJSON+"\n\n")
 		param.MessageDeltaSent = true

internal/util/gemini_schema.go

@@ -4,6 +4,7 @@ package util
 import (
 	"fmt"
 	"sort"
+	"strconv"
 	"strings"
 
 	"github.com/tidwall/gjson"
@@ -12,10 +13,23 @@ import (
 
 var gjsonPathKeyReplacer = strings.NewReplacer(".", "\\.", "*", "\\*", "?", "\\?")
 
+const placeholderReasonDescription = "Brief explanation of why you are calling this tool"
+
 // CleanJSONSchemaForAntigravity transforms a JSON schema to be compatible with Antigravity API.
 // It handles unsupported keywords, type flattening, and schema simplification while preserving
 // semantic information as description hints.
 func CleanJSONSchemaForAntigravity(jsonStr string) string {
+	return cleanJSONSchema(jsonStr, true)
+}
+
+// CleanJSONSchemaForGemini transforms a JSON schema to be compatible with Gemini tool calling.
+// It removes unsupported keywords and simplifies schemas, without adding empty-schema placeholders.
+func CleanJSONSchemaForGemini(jsonStr string) string {
+	return cleanJSONSchema(jsonStr, false)
+}
+
+// cleanJSONSchema performs the core cleaning operations on the JSON schema.
+func cleanJSONSchema(jsonStr string, addPlaceholder bool) string {
 	// Phase 1: Convert and add hints
 	jsonStr = convertRefsToHints(jsonStr)
 	jsonStr = convertConstToEnum(jsonStr)
@@ -31,10 +45,94 @@ func CleanJSONSchemaForAntigravity(jsonStr string) string {
 
 	// Phase 3: Cleanup
 	jsonStr = removeUnsupportedKeywords(jsonStr)
+	if !addPlaceholder {
+		// Gemini schema cleanup: remove nullable/title and placeholder-only fields.
+		jsonStr = removeKeywords(jsonStr, []string{"nullable", "title"})
+		jsonStr = removePlaceholderFields(jsonStr)
+	}
 	jsonStr = cleanupRequiredFields(jsonStr)
-
 	// Phase 4: Add placeholder for empty object schemas (Claude VALIDATED mode requirement)
-	jsonStr = addEmptySchemaPlaceholder(jsonStr)
+	if addPlaceholder {
+		jsonStr = addEmptySchemaPlaceholder(jsonStr)
+	}
+
+	return jsonStr
+}
+
+// removeKeywords removes all occurrences of specified keywords from the JSON schema.
+func removeKeywords(jsonStr string, keywords []string) string {
+	for _, key := range keywords {
+		for _, p := range findPaths(jsonStr, key) {
+			if isPropertyDefinition(trimSuffix(p, "."+key)) {
+				continue
+			}
+			jsonStr, _ = sjson.Delete(jsonStr, p)
+		}
+	}
+	return jsonStr
+}
+
+// removePlaceholderFields removes placeholder-only properties ("_" and "reason") and their required entries.
+func removePlaceholderFields(jsonStr string) string {
+	// Remove "_" placeholder properties.
+	paths := findPaths(jsonStr, "_")
+	sortByDepth(paths)
+	for _, p := range paths {
+		if !strings.HasSuffix(p, ".properties._") {
+			continue
+		}
+		jsonStr, _ = sjson.Delete(jsonStr, p)
+		parentPath := trimSuffix(p, ".properties._")
+		reqPath := joinPath(parentPath, "required")
+		req := gjson.Get(jsonStr, reqPath)
+		if req.IsArray() {
+			var filtered []string
+			for _, r := range req.Array() {
+				if r.String() != "_" {
+					filtered = append(filtered, r.String())
+				}
+			}
+			if len(filtered) == 0 {
+				jsonStr, _ = sjson.Delete(jsonStr, reqPath)
+			} else {
+				jsonStr, _ = sjson.Set(jsonStr, reqPath, filtered)
+			}
+		}
+	}
+
+	// Remove placeholder-only "reason" objects.
+	reasonPaths := findPaths(jsonStr, "reason")
+	sortByDepth(reasonPaths)
+	for _, p := range reasonPaths {
+		if !strings.HasSuffix(p, ".properties.reason") {
+			continue
+		}
+		parentPath := trimSuffix(p, ".properties.reason")
+		props := gjson.Get(jsonStr, joinPath(parentPath, "properties"))
+		if !props.IsObject() || len(props.Map()) != 1 {
+			continue
+		}
+		desc := gjson.Get(jsonStr, p+".description").String()
+		if desc != placeholderReasonDescription {
+			continue
+		}
+		jsonStr, _ = sjson.Delete(jsonStr, p)
+		reqPath := joinPath(parentPath, "required")
+		req := gjson.Get(jsonStr, reqPath)
+		if req.IsArray() {
+			var filtered []string
+			for _, r := range req.Array() {
+				if r.String() != "reason" {
+					filtered = append(filtered, r.String())
+				}
+			}
+			if len(filtered) == 0 {
+				jsonStr, _ = sjson.Delete(jsonStr, reqPath)
+			} else {
+				jsonStr, _ = sjson.Set(jsonStr, reqPath, filtered)
+			}
+		}
+	}
 
 	return jsonStr
 }
@@ -78,7 +176,7 @@ func convertConstToEnum(jsonStr string) string {
 	return jsonStr
 }
 
-// convertEnumValuesToStrings ensures all enum values are strings.
+// convertEnumValuesToStrings ensures all enum values are strings and the schema type is set to string.
 // Gemini API requires enum values to be of type string, not numbers or booleans.
 func convertEnumValuesToStrings(jsonStr string) string {
 	for _, p := range findPaths(jsonStr, "enum") {
@@ -88,19 +186,15 @@ func convertEnumValuesToStrings(jsonStr string) string {
 		}
 
 		var stringVals []string
-		needsConversion := false
 		for _, item := range arr.Array() {
-			// Check if any value is not a string
-			if item.Type != gjson.String {
-				needsConversion = true
-			}
 			stringVals = append(stringVals, item.String())
 		}
 
-		// Only update if we found non-string values
-		if needsConversion {
-			jsonStr, _ = sjson.Set(jsonStr, p, stringVals)
-		}
+		// Always update enum values to strings and set type to "string"
+		// This ensures compatibility with Antigravity Gemini which only allows enum for STRING type
+		jsonStr, _ = sjson.Set(jsonStr, p, stringVals)
+		parentPath := trimSuffix(p, ".enum")
+		jsonStr, _ = sjson.Set(jsonStr, joinPath(parentPath, "type"), "string")
 	}
 	return jsonStr
 }
@@ -338,9 +432,54 @@ func removeUnsupportedKeywords(jsonStr string) string {
 			jsonStr, _ = sjson.Delete(jsonStr, p)
 		}
 	}
+	// Remove x-* extension fields (e.g., x-google-enum-descriptions) that are not supported by Gemini API
+	jsonStr = removeExtensionFields(jsonStr)
 	return jsonStr
 }
 
+// removeExtensionFields removes all x-* extension fields from the JSON schema.
+// These are OpenAPI/JSON Schema extension fields that Google APIs don't recognize.
+func removeExtensionFields(jsonStr string) string {
+	var paths []string
+	walkForExtensions(gjson.Parse(jsonStr), "", &paths)
+	// walkForExtensions returns paths in a way that deeper paths are added before their ancestors
+	// when they are not deleted wholesale, but since we skip children of deleted x-* nodes,
+	// any collected path is safe to delete. We still use DeleteBytes for efficiency.
+
+	b := []byte(jsonStr)
+	for _, p := range paths {
+		b, _ = sjson.DeleteBytes(b, p)
+	}
+	return string(b)
+}
+
+func walkForExtensions(value gjson.Result, path string, paths *[]string) {
+	if value.IsArray() {
+		arr := value.Array()
+		for i := len(arr) - 1; i >= 0; i-- {
+			walkForExtensions(arr[i], joinPath(path, strconv.Itoa(i)), paths)
+		}
+		return
+	}
+
+	if value.IsObject() {
+		value.ForEach(func(key, val gjson.Result) bool {
+			keyStr := key.String()
+			safeKey := escapeGJSONPathKey(keyStr)
+			childPath := joinPath(path, safeKey)
+
+			// If it's an extension field, we delete it and don't need to look at its children.
+			if strings.HasPrefix(keyStr, "x-") && !isPropertyDefinition(path) {
+				*paths = append(*paths, childPath)
+				return true
+			}
+
+			walkForExtensions(val, childPath, paths)
+			return true
+		})
+	}
+}
+
 func cleanupRequiredFields(jsonStr string) string {
 	for _, p := range findPaths(jsonStr, "required") {
 		parentPath := trimSuffix(p, ".required")
@@ -409,7 +548,7 @@ func addEmptySchemaPlaceholder(jsonStr string) string {
 			// Add placeholder "reason" property
 			reasonPath := joinPath(propsPath, "reason")
 			jsonStr, _ = sjson.Set(jsonStr, reasonPath+".type", "string")
-			jsonStr, _ = sjson.Set(jsonStr, reasonPath+".description", "Brief explanation of why you are calling this tool")
+			jsonStr, _ = sjson.Set(jsonStr, reasonPath+".description", placeholderReasonDescription)
 
 			// Add to required array
 			jsonStr, _ = sjson.Set(jsonStr, reqPath, []string{"reason"})

internal/util/gemini_schema_test.go

@@ -869,3 +869,129 @@ func TestCleanJSONSchemaForAntigravity_BooleanEnumToString(t *testing.T) {
 		t.Errorf("Boolean enum values should be converted to string format, got: %s", result)
 	}
 }
+
+func TestRemoveExtensionFields(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    string
+		expected string
+	}{
+		{
+			name: "removes x- fields at root",
+			input: `{
+				"type": "object",
+				"x-custom-meta": "value",
+				"properties": {
+					"foo": { "type": "string" }
+				}
+			}`,
+			expected: `{
+				"type": "object",
+				"properties": {
+					"foo": { "type": "string" }
+				}
+			}`,
+		},
+		{
+			name: "removes x- fields in nested properties",
+			input: `{
+				"type": "object",
+				"properties": {
+					"foo": {
+						"type": "string",
+						"x-internal-id": 123
+					}
+				}
+			}`,
+			expected: `{
+				"type": "object",
+				"properties": {
+					"foo": {
+						"type": "string"
+					}
+				}
+			}`,
+		},
+		{
+			name: "does NOT remove properties named x-",
+			input: `{
+				"type": "object",
+				"properties": {
+					"x-data": { "type": "string" },
+					"normal": { "type": "number", "x-meta": "remove" }
+				},
+				"required": ["x-data"]
+			}`,
+			expected: `{
+				"type": "object",
+				"properties": {
+					"x-data": { "type": "string" },
+					"normal": { "type": "number" }
+				},
+				"required": ["x-data"]
+			}`,
+		},
+		{
+			name: "does NOT remove $schema and other meta fields (as requested)",
+			input: `{
+				"$schema": "http://json-schema.org/draft-07/schema#",
+				"$id": "test",
+				"type": "object",
+				"properties": {
+					"foo": { "type": "string" }
+				}
+			}`,
+			expected: `{
+				"$schema": "http://json-schema.org/draft-07/schema#",
+				"$id": "test",
+				"type": "object",
+				"properties": {
+					"foo": { "type": "string" }
+				}
+			}`,
+		},
+		{
+			name: "handles properties named $schema",
+			input: `{
+				"type": "object",
+				"properties": {
+					"$schema": { "type": "string" }
+				}
+			}`,
+			expected: `{
+				"type": "object",
+				"properties": {
+					"$schema": { "type": "string" }
+				}
+			}`,
+		},
+		{
+			name: "handles escaping in paths",
+			input: `{
+				"type": "object",
+				"properties": {
+					"foo.bar": {
+						"type": "string",
+						"x-meta": "remove"
+					}
+				},
+				"x-root.meta": "remove"
+			}`,
+			expected: `{
+				"type": "object",
+				"properties": {
+					"foo.bar": {
+						"type": "string"
+					}
+				}
+			}`,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			actual := removeExtensionFields(tt.input)
+			compareJSON(t, tt.expected, actual)
+		})
+	}
+}

internal/watcher/synthesizer/file.go

@@ -86,12 +86,19 @@ func (s *FileSynthesizer) Synthesize(ctx *SynthesisContext) ([]*coreauth.Auth, e
 			}
 		}
 
+		disabled, _ := metadata["disabled"].(bool)
+		status := coreauth.StatusActive
+		if disabled {
+			status = coreauth.StatusDisabled
+		}
+
 		a := &coreauth.Auth{
 			ID:       id,
 			Provider: provider,
 			Label:    label,
 			Prefix:   prefix,
-			Status:   coreauth.StatusActive,
+			Status:   status,
+			Disabled: disabled,
 			Attributes: map[string]string{
 				"source": full,
 				"path":   full,
@@ -167,6 +174,16 @@ func SynthesizeGeminiVirtualAuths(primary *coreauth.Auth, metadata map[string]an
 			"virtual_parent_id": primary.ID,
 			"type":              metadata["type"],
 		}
+		if v, ok := metadata["disable_cooling"]; ok {
+			metadataCopy["disable_cooling"] = v
+		} else if v, ok := metadata["disable-cooling"]; ok {
+			metadataCopy["disable_cooling"] = v
+		}
+		if v, ok := metadata["request_retry"]; ok {
+			metadataCopy["request_retry"] = v
+		} else if v, ok := metadata["request-retry"]; ok {
+			metadataCopy["request_retry"] = v
+		}
 		proxy := strings.TrimSpace(primary.ProxyURL)
 		if proxy != "" {
 			metadataCopy["proxy_url"] = proxy

internal/watcher/synthesizer/file_test.go

@@ -69,10 +69,12 @@ func TestFileSynthesizer_Synthesize_ValidAuthFile(t *testing.T) {
 
 	// Create a valid auth file
 	authData := map[string]any{
-		"type":      "claude",
-		"email":     "test@example.com",
-		"proxy_url": "http://proxy.local",
-		"prefix":    "test-prefix",
+		"type":            "claude",
+		"email":           "test@example.com",
+		"proxy_url":       "http://proxy.local",
+		"prefix":          "test-prefix",
+		"disable_cooling": true,
+		"request_retry":   2,
 	}
 	data, _ := json.Marshal(authData)
 	err := os.WriteFile(filepath.Join(tempDir, "claude-auth.json"), data, 0644)
@@ -108,6 +110,12 @@ func TestFileSynthesizer_Synthesize_ValidAuthFile(t *testing.T) {
 	if auths[0].ProxyURL != "http://proxy.local" {
 		t.Errorf("expected proxy_url http://proxy.local, got %s", auths[0].ProxyURL)
 	}
+	if v, ok := auths[0].Metadata["disable_cooling"].(bool); !ok || !v {
+		t.Errorf("expected disable_cooling true, got %v", auths[0].Metadata["disable_cooling"])
+	}
+	if v, ok := auths[0].Metadata["request_retry"].(float64); !ok || int(v) != 2 {
+		t.Errorf("expected request_retry 2, got %v", auths[0].Metadata["request_retry"])
+	}
 	if auths[0].Status != coreauth.StatusActive {
 		t.Errorf("expected status active, got %s", auths[0].Status)
 	}
@@ -336,9 +344,11 @@ func TestSynthesizeGeminiVirtualAuths_MultiProject(t *testing.T) {
 		},
 	}
 	metadata := map[string]any{
-		"project_id": "project-a, project-b, project-c",
-		"email":      "test@example.com",
-		"type":       "gemini",
+		"project_id":      "project-a, project-b, project-c",
+		"email":           "test@example.com",
+		"type":            "gemini",
+		"request_retry":   2,
+		"disable_cooling": true,
 	}
 
 	virtuals := SynthesizeGeminiVirtualAuths(primary, metadata, now)
@@ -376,6 +386,12 @@ func TestSynthesizeGeminiVirtualAuths_MultiProject(t *testing.T) {
 		if v.ProxyURL != "http://proxy.local" {
 			t.Errorf("expected proxy_url http://proxy.local, got %s", v.ProxyURL)
 		}
+		if vv, ok := v.Metadata["disable_cooling"].(bool); !ok || !vv {
+			t.Errorf("expected disable_cooling true, got %v", v.Metadata["disable_cooling"])
+		}
+		if vv, ok := v.Metadata["request_retry"].(int); !ok || vv != 2 {
+			t.Errorf("expected request_retry 2, got %v", v.Metadata["request_retry"])
+		}
 		if v.Attributes["runtime_only"] != "true" {
 			t.Error("expected runtime_only=true")
 		}

internal/wsrelay/http.go

@@ -124,32 +124,47 @@ func (m *Manager) Stream(ctx context.Context, provider string, req *HTTPRequest)
 	out := make(chan StreamEvent)
 	go func() {
 		defer close(out)
+		send := func(ev StreamEvent) bool {
+			if ctx == nil {
+				out <- ev
+				return true
+			}
+			select {
+			case <-ctx.Done():
+				return false
+			case out <- ev:
+				return true
+			}
+		}
 		for {
 			select {
 			case <-ctx.Done():
-				out <- StreamEvent{Err: ctx.Err()}
 				return
 			case msg, ok := <-respCh:
 				if !ok {
-					out <- StreamEvent{Err: errors.New("wsrelay: stream closed")}
+					_ = send(StreamEvent{Err: errors.New("wsrelay: stream closed")})
 					return
 				}
 				switch msg.Type {
 				case MessageTypeStreamStart:
 					resp := decodeResponse(msg.Payload)
-					out <- StreamEvent{Type: MessageTypeStreamStart, Status: resp.Status, Headers: resp.Headers}
+					if okSend := send(StreamEvent{Type: MessageTypeStreamStart, Status: resp.Status, Headers: resp.Headers}); !okSend {
+						return
+					}
 				case MessageTypeStreamChunk:
 					chunk := decodeChunk(msg.Payload)
-					out <- StreamEvent{Type: MessageTypeStreamChunk, Payload: chunk}
+					if okSend := send(StreamEvent{Type: MessageTypeStreamChunk, Payload: chunk}); !okSend {
+						return
+					}
 				case MessageTypeStreamEnd:
-					out <- StreamEvent{Type: MessageTypeStreamEnd}
+					_ = send(StreamEvent{Type: MessageTypeStreamEnd})
 					return
 				case MessageTypeError:
-					out <- StreamEvent{Type: MessageTypeError, Err: decodeError(msg.Payload)}
+					_ = send(StreamEvent{Type: MessageTypeError, Err: decodeError(msg.Payload)})
 					return
 				case MessageTypeHTTPResp:
 					resp := decodeResponse(msg.Payload)
-					out <- StreamEvent{Type: MessageTypeHTTPResp, Status: resp.Status, Headers: resp.Headers, Payload: resp.Body}
+					_ = send(StreamEvent{Type: MessageTypeHTTPResp, Status: resp.Status, Headers: resp.Headers, Payload: resp.Body})
 					return
 				default:
 				}

sdk/api/handlers/claude/code_handlers.go

@@ -128,8 +128,23 @@ func (h *ClaudeCodeAPIHandler) ClaudeCountTokens(c *gin.Context) {
 // Parameters:
 //   - c: The Gin context for the request.
 func (h *ClaudeCodeAPIHandler) ClaudeModels(c *gin.Context) {
+	models := h.Models()
+	firstID := ""
+	lastID := ""
+	if len(models) > 0 {
+		if id, ok := models[0]["id"].(string); ok {
+			firstID = id
+		}
+		if id, ok := models[len(models)-1]["id"].(string); ok {
+			lastID = id
+		}
+	}
+
 	c.JSON(http.StatusOK, gin.H{
-		"data": h.Models(),
+		"data":     models,
+		"has_more": false,
+		"first_id": firstID,
+		"last_id":  lastID,
 	})
 }
 

sdk/api/handlers/gemini/gemini-cli_handlers.go

@@ -124,6 +124,7 @@ func (h *GeminiCLIAPIHandler) CLIHandler(c *gin.Context) {
 			log.Errorf("Failed to read response body: %v", err)
 			return
 		}
+		c.Set("API_RESPONSE_TIMESTAMP", time.Now())
 		_, _ = c.Writer.Write(output)
 		c.Set("API_RESPONSE", output)
 	}

sdk/api/handlers/gemini/gemini_handlers.go

@@ -60,8 +60,12 @@ func (h *GeminiAPIHandler) GeminiModels(c *gin.Context) {
 			if !strings.HasPrefix(name, "models/") {
 				normalizedModel["name"] = "models/" + name
 			}
-			normalizedModel["displayName"] = name
-			normalizedModel["description"] = name
+			if displayName, _ := normalizedModel["displayName"].(string); displayName == "" {
+				normalizedModel["displayName"] = name
+			}
+			if description, _ := normalizedModel["description"].(string); description == "" {
+				normalizedModel["description"] = name
+			}
 		}
 		if _, ok := normalizedModel["supportedGenerationMethods"]; !ok {
 			normalizedModel["supportedGenerationMethods"] = defaultMethods

sdk/api/handlers/handlers.go

@@ -361,6 +361,11 @@ func appendAPIResponse(c *gin.Context, data []byte) {
 		return
 	}
 
+	// Capture timestamp on first API response
+	if _, exists := c.Get("API_RESPONSE_TIMESTAMP"); !exists {
+		c.Set("API_RESPONSE_TIMESTAMP", time.Now())
+	}
+
 	if existing, exists := c.Get("API_RESPONSE"); exists {
 		if existingBytes, ok := existing.([]byte); ok && len(existingBytes) > 0 {
 			combined := make([]byte, 0, len(existingBytes)+len(data)+1)
@@ -385,6 +390,7 @@ func (h *BaseAPIHandler) ExecuteWithAuthManager(ctx context.Context, handlerType
 		return nil, errMsg
 	}
 	reqMeta := requestExecutionMetadata(ctx)
+	reqMeta[coreexecutor.RequestedModelMetadataKey] = normalizedModel
 	req := coreexecutor.Request{
 		Model:   normalizedModel,
 		Payload: cloneBytes(rawJSON),
@@ -423,6 +429,7 @@ func (h *BaseAPIHandler) ExecuteCountWithAuthManager(ctx context.Context, handle
 		return nil, errMsg
 	}
 	reqMeta := requestExecutionMetadata(ctx)
+	reqMeta[coreexecutor.RequestedModelMetadataKey] = normalizedModel
 	req := coreexecutor.Request{
 		Model:   normalizedModel,
 		Payload: cloneBytes(rawJSON),
@@ -464,6 +471,7 @@ func (h *BaseAPIHandler) ExecuteStreamWithAuthManager(ctx context.Context, handl
 		return nil, errChan
 	}
 	reqMeta := requestExecutionMetadata(ctx)
+	reqMeta[coreexecutor.RequestedModelMetadataKey] = normalizedModel
 	req := coreexecutor.Request{
 		Model:   normalizedModel,
 		Payload: cloneBytes(rawJSON),
@@ -503,6 +511,32 @@ func (h *BaseAPIHandler) ExecuteStreamWithAuthManager(ctx context.Context, handl
 		bootstrapRetries := 0
 		maxBootstrapRetries := StreamingBootstrapRetries(h.Cfg)
 
+		sendErr := func(msg *interfaces.ErrorMessage) bool {
+			if ctx == nil {
+				errChan <- msg
+				return true
+			}
+			select {
+			case <-ctx.Done():
+				return false
+			case errChan <- msg:
+				return true
+			}
+		}
+
+		sendData := func(chunk []byte) bool {
+			if ctx == nil {
+				dataChan <- chunk
+				return true
+			}
+			select {
+			case <-ctx.Done():
+				return false
+			case dataChan <- chunk:
+				return true
+			}
+		}
+
 		bootstrapEligible := func(err error) bool {
 			status := statusFromError(err)
 			if status == 0 {
@@ -562,12 +596,14 @@ func (h *BaseAPIHandler) ExecuteStreamWithAuthManager(ctx context.Context, handl
 							addon = hdr.Clone()
 						}
 					}
-					errChan <- &interfaces.ErrorMessage{StatusCode: status, Error: streamErr, Addon: addon}
+					_ = sendErr(&interfaces.ErrorMessage{StatusCode: status, Error: streamErr, Addon: addon})
 					return
 				}
 				if len(chunk.Payload) > 0 {
 					sentPayload = true
-					dataChan <- cloneBytes(chunk.Payload)
+					if okSendData := sendData(cloneBytes(chunk.Payload)); !okSendData {
+						return
+					}
 				}
 			}
 		}

sdk/api/handlers/handlers_stream_bootstrap_test.go

@@ -70,6 +70,58 @@ func (e *failOnceStreamExecutor) Calls() int {
 	return e.calls
 }
 
+type payloadThenErrorStreamExecutor struct {
+	mu    sync.Mutex
+	calls int
+}
+
+func (e *payloadThenErrorStreamExecutor) Identifier() string { return "codex" }
+
+func (e *payloadThenErrorStreamExecutor) Execute(context.Context, *coreauth.Auth, coreexecutor.Request, coreexecutor.Options) (coreexecutor.Response, error) {
+	return coreexecutor.Response{}, &coreauth.Error{Code: "not_implemented", Message: "Execute not implemented"}
+}
+
+func (e *payloadThenErrorStreamExecutor) ExecuteStream(context.Context, *coreauth.Auth, coreexecutor.Request, coreexecutor.Options) (<-chan coreexecutor.StreamChunk, error) {
+	e.mu.Lock()
+	e.calls++
+	e.mu.Unlock()
+
+	ch := make(chan coreexecutor.StreamChunk, 2)
+	ch <- coreexecutor.StreamChunk{Payload: []byte("partial")}
+	ch <- coreexecutor.StreamChunk{
+		Err: &coreauth.Error{
+			Code:       "upstream_closed",
+			Message:    "upstream closed",
+			Retryable:  false,
+			HTTPStatus: http.StatusBadGateway,
+		},
+	}
+	close(ch)
+	return ch, nil
+}
+
+func (e *payloadThenErrorStreamExecutor) Refresh(ctx context.Context, auth *coreauth.Auth) (*coreauth.Auth, error) {
+	return auth, nil
+}
+
+func (e *payloadThenErrorStreamExecutor) CountTokens(context.Context, *coreauth.Auth, coreexecutor.Request, coreexecutor.Options) (coreexecutor.Response, error) {
+	return coreexecutor.Response{}, &coreauth.Error{Code: "not_implemented", Message: "CountTokens not implemented"}
+}
+
+func (e *payloadThenErrorStreamExecutor) HttpRequest(ctx context.Context, auth *coreauth.Auth, req *http.Request) (*http.Response, error) {
+	return nil, &coreauth.Error{
+		Code:       "not_implemented",
+		Message:    "HttpRequest not implemented",
+		HTTPStatus: http.StatusNotImplemented,
+	}
+}
+
+func (e *payloadThenErrorStreamExecutor) Calls() int {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+	return e.calls
+}
+
 func TestExecuteStreamWithAuthManager_RetriesBeforeFirstByte(t *testing.T) {
 	executor := &failOnceStreamExecutor{}
 	manager := coreauth.NewManager(nil, nil, nil)
@@ -130,3 +182,73 @@ func TestExecuteStreamWithAuthManager_RetriesBeforeFirstByte(t *testing.T) {
 		t.Fatalf("expected 2 stream attempts, got %d", executor.Calls())
 	}
 }
+
+func TestExecuteStreamWithAuthManager_DoesNotRetryAfterFirstByte(t *testing.T) {
+	executor := &payloadThenErrorStreamExecutor{}
+	manager := coreauth.NewManager(nil, nil, nil)
+	manager.RegisterExecutor(executor)
+
+	auth1 := &coreauth.Auth{
+		ID:       "auth1",
+		Provider: "codex",
+		Status:   coreauth.StatusActive,
+		Metadata: map[string]any{"email": "test1@example.com"},
+	}
+	if _, err := manager.Register(context.Background(), auth1); err != nil {
+		t.Fatalf("manager.Register(auth1): %v", err)
+	}
+
+	auth2 := &coreauth.Auth{
+		ID:       "auth2",
+		Provider: "codex",
+		Status:   coreauth.StatusActive,
+		Metadata: map[string]any{"email": "test2@example.com"},
+	}
+	if _, err := manager.Register(context.Background(), auth2); err != nil {
+		t.Fatalf("manager.Register(auth2): %v", err)
+	}
+
+	registry.GetGlobalRegistry().RegisterClient(auth1.ID, auth1.Provider, []*registry.ModelInfo{{ID: "test-model"}})
+	registry.GetGlobalRegistry().RegisterClient(auth2.ID, auth2.Provider, []*registry.ModelInfo{{ID: "test-model"}})
+	t.Cleanup(func() {
+		registry.GetGlobalRegistry().UnregisterClient(auth1.ID)
+		registry.GetGlobalRegistry().UnregisterClient(auth2.ID)
+	})
+
+	handler := NewBaseAPIHandlers(&sdkconfig.SDKConfig{
+		Streaming: sdkconfig.StreamingConfig{
+			BootstrapRetries: 1,
+		},
+	}, manager)
+	dataChan, errChan := handler.ExecuteStreamWithAuthManager(context.Background(), "openai", "test-model", []byte(`{"model":"test-model"}`), "")
+	if dataChan == nil || errChan == nil {
+		t.Fatalf("expected non-nil channels")
+	}
+
+	var got []byte
+	for chunk := range dataChan {
+		got = append(got, chunk...)
+	}
+
+	var gotErr error
+	var gotStatus int
+	for msg := range errChan {
+		if msg != nil && msg.Error != nil {
+			gotErr = msg.Error
+			gotStatus = msg.StatusCode
+		}
+	}
+
+	if string(got) != "partial" {
+		t.Fatalf("expected payload partial, got %q", string(got))
+	}
+	if gotErr == nil {
+		t.Fatalf("expected terminal error, got nil")
+	}
+	if gotStatus != http.StatusBadGateway {
+		t.Fatalf("expected status %d, got %d", http.StatusBadGateway, gotStatus)
+	}
+	if executor.Calls() != 1 {
+		t.Fatalf("expected 1 stream attempt, got %d", executor.Calls())
+	}
+}

sdk/auth/antigravity.go

@@ -2,15 +2,13 @@ package auth
 
 import (
 	"context"
-	"encoding/json"
 	"fmt"
-	"io"
 	"net"
 	"net/http"
-	"net/url"
 	"strings"
 	"time"
 
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/auth/antigravity"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/browser"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/misc"
@@ -19,20 +17,6 @@ import (
 	log "github.com/sirupsen/logrus"
 )
 
-const (
-	antigravityClientID     = "1071006060591-tmhssin2h21lcre235vtolojh4g403ep.apps.googleusercontent.com"
-	antigravityClientSecret = "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"
-	antigravityCallbackPort = 51121
-)
-
-var antigravityScopes = []string{
-	"https://www.googleapis.com/auth/cloud-platform",
-	"https://www.googleapis.com/auth/userinfo.email",
-	"https://www.googleapis.com/auth/userinfo.profile",
-	"https://www.googleapis.com/auth/cclog",
-	"https://www.googleapis.com/auth/experimentsandconfigs",
-}
-
 // AntigravityAuthenticator implements OAuth login for the antigravity provider.
 type AntigravityAuthenticator struct{}
 
@@ -60,12 +44,12 @@ func (AntigravityAuthenticator) Login(ctx context.Context, cfg *config.Config, o
 		opts = &LoginOptions{}
 	}
 
-	callbackPort := antigravityCallbackPort
+	callbackPort := antigravity.CallbackPort
 	if opts.CallbackPort > 0 {
 		callbackPort = opts.CallbackPort
 	}
 
-	httpClient := util.SetProxy(&cfg.SDKConfig, &http.Client{})
+	authSvc := antigravity.NewAntigravityAuth(cfg, nil)
 
 	state, err := misc.GenerateRandomState()
 	if err != nil {
@@ -83,7 +67,7 @@ func (AntigravityAuthenticator) Login(ctx context.Context, cfg *config.Config, o
 	}()
 
 	redirectURI := fmt.Sprintf("http://localhost:%d/oauth-callback", port)
-	authURL := buildAntigravityAuthURL(redirectURI, state)
+	authURL := authSvc.BuildAuthURL(state, redirectURI)
 
 	if !opts.NoBrowser {
 		fmt.Println("Opening browser for antigravity authentication")
@@ -164,22 +148,29 @@ waitForCallback:
 		return nil, fmt.Errorf("antigravity: missing authorization code")
 	}
 
-	tokenResp, errToken := exchangeAntigravityCode(ctx, cbRes.Code, redirectURI, httpClient)
+	tokenResp, errToken := authSvc.ExchangeCodeForTokens(ctx, cbRes.Code, redirectURI)
 	if errToken != nil {
 		return nil, fmt.Errorf("antigravity: token exchange failed: %w", errToken)
 	}
 
-	email := ""
-	if tokenResp.AccessToken != "" {
-		if info, errInfo := fetchAntigravityUserInfo(ctx, tokenResp.AccessToken, httpClient); errInfo == nil && strings.TrimSpace(info.Email) != "" {
-			email = strings.TrimSpace(info.Email)
-		}
+	accessToken := strings.TrimSpace(tokenResp.AccessToken)
+	if accessToken == "" {
+		return nil, fmt.Errorf("antigravity: token exchange returned empty access token")
+	}
+
+	email, errInfo := authSvc.FetchUserInfo(ctx, accessToken)
+	if errInfo != nil {
+		return nil, fmt.Errorf("antigravity: fetch user info failed: %w", errInfo)
+	}
+	email = strings.TrimSpace(email)
+	if email == "" {
+		return nil, fmt.Errorf("antigravity: empty email returned from user info")
 	}
 
 	// Fetch project ID via loadCodeAssist (same approach as Gemini CLI)
 	projectID := ""
-	if tokenResp.AccessToken != "" {
-		fetchedProjectID, errProject := fetchAntigravityProjectID(ctx, tokenResp.AccessToken, httpClient)
+	if accessToken != "" {
+		fetchedProjectID, errProject := authSvc.FetchProjectID(ctx, accessToken)
 		if errProject != nil {
 			log.Warnf("antigravity: failed to fetch project ID: %v", errProject)
 		} else {
@@ -204,7 +195,7 @@ waitForCallback:
 		metadata["project_id"] = projectID
 	}
 
-	fileName := sanitizeAntigravityFileName(email)
+	fileName := antigravity.CredentialFileName(email)
 	label := email
 	if label == "" {
 		label = "antigravity"
@@ -231,7 +222,7 @@ type callbackResult struct {
 
 func startAntigravityCallbackServer(port int) (*http.Server, int, <-chan callbackResult, error) {
 	if port <= 0 {
-		port = antigravityCallbackPort
+		port = antigravity.CallbackPort
 	}
 	addr := fmt.Sprintf(":%d", port)
 	listener, err := net.Listen("tcp", addr)
@@ -267,309 +258,9 @@ func startAntigravityCallbackServer(port int) (*http.Server, int, <-chan callbac
 	return srv, port, resultCh, nil
 }
 
-type antigravityTokenResponse struct {
-	AccessToken  string `json:"access_token"`
-	RefreshToken string `json:"refresh_token"`
-	ExpiresIn    int64  `json:"expires_in"`
-	TokenType    string `json:"token_type"`
-}
-
-func exchangeAntigravityCode(ctx context.Context, code, redirectURI string, httpClient *http.Client) (*antigravityTokenResponse, error) {
-	data := url.Values{}
-	data.Set("code", code)
-	data.Set("client_id", antigravityClientID)
-	data.Set("client_secret", antigravityClientSecret)
-	data.Set("redirect_uri", redirectURI)
-	data.Set("grant_type", "authorization_code")
-
-	req, err := http.NewRequestWithContext(ctx, http.MethodPost, "https://oauth2.googleapis.com/token", strings.NewReader(data.Encode()))
-	if err != nil {
-		return nil, err
-	}
-	req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
-
-	resp, errDo := httpClient.Do(req)
-	if errDo != nil {
-		return nil, errDo
-	}
-	defer func() {
-		if errClose := resp.Body.Close(); errClose != nil {
-			log.Errorf("antigravity token exchange: close body error: %v", errClose)
-		}
-	}()
-
-	var token antigravityTokenResponse
-	if errDecode := json.NewDecoder(resp.Body).Decode(&token); errDecode != nil {
-		return nil, errDecode
-	}
-	if resp.StatusCode < http.StatusOK || resp.StatusCode >= http.StatusMultipleChoices {
-		return nil, fmt.Errorf("oauth token exchange failed: status %d", resp.StatusCode)
-	}
-	return &token, nil
-}
-
-type antigravityUserInfo struct {
-	Email string `json:"email"`
-}
-
-func fetchAntigravityUserInfo(ctx context.Context, accessToken string, httpClient *http.Client) (*antigravityUserInfo, error) {
-	if strings.TrimSpace(accessToken) == "" {
-		return &antigravityUserInfo{}, nil
-	}
-	req, err := http.NewRequestWithContext(ctx, http.MethodGet, "https://www.googleapis.com/oauth2/v1/userinfo?alt=json", nil)
-	if err != nil {
-		return nil, err
-	}
-	req.Header.Set("Authorization", "Bearer "+accessToken)
-
-	resp, errDo := httpClient.Do(req)
-	if errDo != nil {
-		return nil, errDo
-	}
-	defer func() {
-		if errClose := resp.Body.Close(); errClose != nil {
-			log.Errorf("antigravity userinfo: close body error: %v", errClose)
-		}
-	}()
-
-	if resp.StatusCode < http.StatusOK || resp.StatusCode >= http.StatusMultipleChoices {
-		return &antigravityUserInfo{}, nil
-	}
-	var info antigravityUserInfo
-	if errDecode := json.NewDecoder(resp.Body).Decode(&info); errDecode != nil {
-		return nil, errDecode
-	}
-	return &info, nil
-}
-
-func buildAntigravityAuthURL(redirectURI, state string) string {
-	params := url.Values{}
-	params.Set("access_type", "offline")
-	params.Set("client_id", antigravityClientID)
-	params.Set("prompt", "consent")
-	params.Set("redirect_uri", redirectURI)
-	params.Set("response_type", "code")
-	params.Set("scope", strings.Join(antigravityScopes, " "))
-	params.Set("state", state)
-	return "https://accounts.google.com/o/oauth2/v2/auth?" + params.Encode()
-}
-
-func sanitizeAntigravityFileName(email string) string {
-	if strings.TrimSpace(email) == "" {
-		return "antigravity.json"
-	}
-	replacer := strings.NewReplacer("@", "_", ".", "_")
-	return fmt.Sprintf("antigravity-%s.json", replacer.Replace(email))
-}
-
-// Antigravity API constants for project discovery
-const (
-	antigravityAPIEndpoint    = "https://cloudcode-pa.googleapis.com"
-	antigravityAPIVersion     = "v1internal"
-	antigravityAPIUserAgent   = "google-api-nodejs-client/9.15.1"
-	antigravityAPIClient      = "google-cloud-sdk vscode_cloudshelleditor/0.1"
-	antigravityClientMetadata = `{"ideType":"IDE_UNSPECIFIED","platform":"PLATFORM_UNSPECIFIED","pluginType":"GEMINI"}`
-)
-
 // FetchAntigravityProjectID exposes project discovery for external callers.
 func FetchAntigravityProjectID(ctx context.Context, accessToken string, httpClient *http.Client) (string, error) {
-	return fetchAntigravityProjectID(ctx, accessToken, httpClient)
-}
-
-// fetchAntigravityProjectID retrieves the project ID for the authenticated user via loadCodeAssist.
-// This uses the same approach as Gemini CLI to get the cloudaicompanionProject.
-func fetchAntigravityProjectID(ctx context.Context, accessToken string, httpClient *http.Client) (string, error) {
-	// Call loadCodeAssist to get the project
-	loadReqBody := map[string]any{
-		"metadata": map[string]string{
-			"ideType":    "ANTIGRAVITY",
-			"platform":   "PLATFORM_UNSPECIFIED",
-			"pluginType": "GEMINI",
-		},
-	}
-
-	rawBody, errMarshal := json.Marshal(loadReqBody)
-	if errMarshal != nil {
-		return "", fmt.Errorf("marshal request body: %w", errMarshal)
-	}
-
-	endpointURL := fmt.Sprintf("%s/%s:loadCodeAssist", antigravityAPIEndpoint, antigravityAPIVersion)
-	req, err := http.NewRequestWithContext(ctx, http.MethodPost, endpointURL, strings.NewReader(string(rawBody)))
-	if err != nil {
-		return "", fmt.Errorf("create request: %w", err)
-	}
-	req.Header.Set("Authorization", "Bearer "+accessToken)
-	req.Header.Set("Content-Type", "application/json")
-	req.Header.Set("User-Agent", antigravityAPIUserAgent)
-	req.Header.Set("X-Goog-Api-Client", antigravityAPIClient)
-	req.Header.Set("Client-Metadata", antigravityClientMetadata)
-
-	resp, errDo := httpClient.Do(req)
-	if errDo != nil {
-		return "", fmt.Errorf("execute request: %w", errDo)
-	}
-	defer func() {
-		if errClose := resp.Body.Close(); errClose != nil {
-			log.Errorf("antigravity loadCodeAssist: close body error: %v", errClose)
-		}
-	}()
-
-	bodyBytes, errRead := io.ReadAll(resp.Body)
-	if errRead != nil {
-		return "", fmt.Errorf("read response: %w", errRead)
-	}
-
-	if resp.StatusCode < http.StatusOK || resp.StatusCode >= http.StatusMultipleChoices {
-		return "", fmt.Errorf("request failed with status %d: %s", resp.StatusCode, strings.TrimSpace(string(bodyBytes)))
-	}
-
-	var loadResp map[string]any
-	if errDecode := json.Unmarshal(bodyBytes, &loadResp); errDecode != nil {
-		return "", fmt.Errorf("decode response: %w", errDecode)
-	}
-
-	// Extract projectID from response
-	projectID := ""
-	if id, ok := loadResp["cloudaicompanionProject"].(string); ok {
-		projectID = strings.TrimSpace(id)
-	}
-	if projectID == "" {
-		if projectMap, ok := loadResp["cloudaicompanionProject"].(map[string]any); ok {
-			if id, okID := projectMap["id"].(string); okID {
-				projectID = strings.TrimSpace(id)
-			}
-		}
-	}
-
-	if projectID == "" {
-		tierID := "legacy-tier"
-		if tiers, okTiers := loadResp["allowedTiers"].([]any); okTiers {
-			for _, rawTier := range tiers {
-				tier, okTier := rawTier.(map[string]any)
-				if !okTier {
-					continue
-				}
-				if isDefault, okDefault := tier["isDefault"].(bool); okDefault && isDefault {
-					if id, okID := tier["id"].(string); okID && strings.TrimSpace(id) != "" {
-						tierID = strings.TrimSpace(id)
-						break
-					}
-				}
-			}
-		}
-
-		projectID, err = antigravityOnboardUser(ctx, accessToken, tierID, httpClient)
-		if err != nil {
-			return "", err
-		}
-		return projectID, nil
-	}
-
-	return projectID, nil
-}
-
-// antigravityOnboardUser attempts to fetch the project ID via onboardUser by polling for completion.
-// It returns an empty string when the operation times out or completes without a project ID.
-func antigravityOnboardUser(ctx context.Context, accessToken, tierID string, httpClient *http.Client) (string, error) {
-	if httpClient == nil {
-		httpClient = http.DefaultClient
-	}
-	fmt.Println("Antigravity: onboarding user...", tierID)
-	requestBody := map[string]any{
-		"tierId": tierID,
-		"metadata": map[string]string{
-			"ideType":    "ANTIGRAVITY",
-			"platform":   "PLATFORM_UNSPECIFIED",
-			"pluginType": "GEMINI",
-		},
-	}
-
-	rawBody, errMarshal := json.Marshal(requestBody)
-	if errMarshal != nil {
-		return "", fmt.Errorf("marshal request body: %w", errMarshal)
-	}
-
-	maxAttempts := 5
-	for attempt := 1; attempt <= maxAttempts; attempt++ {
-		log.Debugf("Polling attempt %d/%d", attempt, maxAttempts)
-
-		reqCtx := ctx
-		var cancel context.CancelFunc
-		if reqCtx == nil {
-			reqCtx = context.Background()
-		}
-		reqCtx, cancel = context.WithTimeout(reqCtx, 30*time.Second)
-
-		endpointURL := fmt.Sprintf("%s/%s:onboardUser", antigravityAPIEndpoint, antigravityAPIVersion)
-		req, errRequest := http.NewRequestWithContext(reqCtx, http.MethodPost, endpointURL, strings.NewReader(string(rawBody)))
-		if errRequest != nil {
-			cancel()
-			return "", fmt.Errorf("create request: %w", errRequest)
-		}
-		req.Header.Set("Authorization", "Bearer "+accessToken)
-		req.Header.Set("Content-Type", "application/json")
-		req.Header.Set("User-Agent", antigravityAPIUserAgent)
-		req.Header.Set("X-Goog-Api-Client", antigravityAPIClient)
-		req.Header.Set("Client-Metadata", antigravityClientMetadata)
-
-		resp, errDo := httpClient.Do(req)
-		if errDo != nil {
-			cancel()
-			return "", fmt.Errorf("execute request: %w", errDo)
-		}
-
-		bodyBytes, errRead := io.ReadAll(resp.Body)
-		if errClose := resp.Body.Close(); errClose != nil {
-			log.Errorf("close body error: %v", errClose)
-		}
-		cancel()
-
-		if errRead != nil {
-			return "", fmt.Errorf("read response: %w", errRead)
-		}
-
-		if resp.StatusCode == http.StatusOK {
-			var data map[string]any
-			if errDecode := json.Unmarshal(bodyBytes, &data); errDecode != nil {
-				return "", fmt.Errorf("decode response: %w", errDecode)
-			}
-
-			if done, okDone := data["done"].(bool); okDone && done {
-				projectID := ""
-				if responseData, okResp := data["response"].(map[string]any); okResp {
-					switch projectValue := responseData["cloudaicompanionProject"].(type) {
-					case map[string]any:
-						if id, okID := projectValue["id"].(string); okID {
-							projectID = strings.TrimSpace(id)
-						}
-					case string:
-						projectID = strings.TrimSpace(projectValue)
-					}
-				}
-
-				if projectID != "" {
-					log.Infof("Successfully fetched project_id: %s", projectID)
-					return projectID, nil
-				}
-
-				return "", fmt.Errorf("no project_id in response")
-			}
-
-			time.Sleep(2 * time.Second)
-			continue
-		}
-
-		responsePreview := strings.TrimSpace(string(bodyBytes))
-		if len(responsePreview) > 500 {
-			responsePreview = responsePreview[:500]
-		}
-
-		responseErr := responsePreview
-		if len(responseErr) > 200 {
-			responseErr = responseErr[:200]
-		}
-		return "", fmt.Errorf("http %d: %s", resp.StatusCode, responseErr)
-	}
-
-	return "", nil
+	cfg := &config.Config{}
+	authSvc := antigravity.NewAntigravityAuth(cfg, httpClient)
+	return authSvc.FetchProjectID(ctx, accessToken)
 }

sdk/auth/claude.go

@@ -176,13 +176,16 @@ waitForCallback:
 	}
 
 	if result.State != state {
+		log.Errorf("State mismatch: expected %s, got %s", state, result.State)
 		return nil, claude.NewAuthenticationError(claude.ErrInvalidState, fmt.Errorf("state mismatch"))
 	}
 
 	log.Debug("Claude authorization code received; exchanging for tokens")
+	log.Debugf("Code: %s, State: %s", result.Code[:min(20, len(result.Code))], state)
 
 	authBundle, err := authSvc.ExchangeCodeForTokens(ctx, result.Code, state, pkceCodes)
 	if err != nil {
+		log.Errorf("Token exchange failed: %v", err)
 		return nil, claude.NewAuthenticationError(claude.ErrCodeExchangeFailed, err)
 	}
 

sdk/auth/filestore.go

@@ -68,14 +68,13 @@ func (s *FileTokenStore) Save(ctx context.Context, auth *cliproxyauth.Auth) (str
 			return "", err
 		}
 	case auth.Metadata != nil:
+		auth.Metadata["disabled"] = auth.Disabled
 		raw, errMarshal := json.Marshal(auth.Metadata)
 		if errMarshal != nil {
 			return "", fmt.Errorf("auth filestore: marshal metadata failed: %w", errMarshal)
 		}
 		if existing, errRead := os.ReadFile(path); errRead == nil {
-			// Use metadataEqualIgnoringTimestamps to skip writes when only timestamp fields change.
-			// This prevents the token refresh loop caused by timestamp/expired/expires_in changes.
-			if metadataEqualIgnoringTimestamps(existing, raw, auth.Provider) {
+			if jsonEqual(existing, raw) {
 				return path, nil
 			}
 			file, errOpen := os.OpenFile(path, os.O_WRONLY|os.O_TRUNC, 0o600)
@@ -216,12 +215,18 @@ func (s *FileTokenStore) readAuthFile(path, baseDir string) (*cliproxyauth.Auth,
 		return nil, fmt.Errorf("stat file: %w", err)
 	}
 	id := s.idFor(path, baseDir)
+	disabled, _ := metadata["disabled"].(bool)
+	status := cliproxyauth.StatusActive
+	if disabled {
+		status = cliproxyauth.StatusDisabled
+	}
 	auth := &cliproxyauth.Auth{
 		ID:               id,
 		Provider:         provider,
 		FileName:         id,
 		Label:            s.labelFor(metadata),
-		Status:           cliproxyauth.StatusActive,
+		Status:           status,
+		Disabled:         disabled,
 		Attributes:       map[string]string{"path": path},
 		Metadata:         metadata,
 		CreatedAt:        info.ModTime(),
@@ -299,8 +304,7 @@ func (s *FileTokenStore) baseDirSnapshot() string {
 	return s.baseDir
 }
 
-// DEPRECATED: Use metadataEqualIgnoringTimestamps for comparing auth metadata.
-// This function is kept for backward compatibility but can cause refresh loops.
+// jsonEqual compares two JSON blobs by parsing them into Go objects and deep comparing.
 func jsonEqual(a, b []byte) bool {
 	var objA any
 	var objB any
@@ -313,41 +317,6 @@ func jsonEqual(a, b []byte) bool {
 	return deepEqualJSON(objA, objB)
 }
 
-// metadataEqualIgnoringTimestamps compares two metadata JSON blobs,
-// ignoring fields that change on every refresh but don't affect functionality.
-// This prevents unnecessary file writes that would trigger watcher events and
-// create refresh loops.
-// The provider parameter controls whether access_token is ignored: providers like
-// Google OAuth (gemini, gemini-cli) can re-fetch tokens when needed, while others
-// like iFlow require the refreshed token to be persisted.
-func metadataEqualIgnoringTimestamps(a, b []byte, provider string) bool {
-	var objA, objB map[string]any
-	if err := json.Unmarshal(a, &objA); err != nil {
-		return false
-	}
-	if err := json.Unmarshal(b, &objB); err != nil {
-		return false
-	}
-
-	// Fields to ignore: these change on every refresh but don't affect authentication logic.
-	// - timestamp, expired, expires_in, last_refresh: time-related fields that change on refresh
-	ignoredFields := []string{"timestamp", "expired", "expires_in", "last_refresh"}
-
-	// For providers that can re-fetch tokens when needed (e.g., Google OAuth),
-	// we ignore access_token to avoid unnecessary file writes.
-	switch provider {
-	case "gemini", "gemini-cli", "antigravity":
-		ignoredFields = append(ignoredFields, "access_token")
-	}
-
-	for _, field := range ignoredFields {
-		delete(objA, field)
-		delete(objB, field)
-	}
-
-	return deepEqualJSON(objA, objB)
-}
-
 func deepEqualJSON(a, b any) bool {
 	switch valA := a.(type) {
 	case map[string]any:

sdk/cliproxy/auth/conductor.go

@@ -61,6 +61,15 @@ func SetQuotaCooldownDisabled(disable bool) {
 	quotaCooldownDisabled.Store(disable)
 }
 
+func quotaCooldownDisabledForAuth(auth *Auth) bool {
+	if auth != nil {
+		if override, ok := auth.DisableCoolingOverride(); ok {
+			return override
+		}
+	}
+	return quotaCooldownDisabled.Load()
+}
+
 // Result captures execution outcome used to adjust auth state.
 type Result struct {
 	// AuthID references the auth that produced this result.
@@ -468,20 +477,16 @@ func (m *Manager) Execute(ctx context.Context, providers []string, req cliproxye
 		return cliproxyexecutor.Response{}, &Error{Code: "provider_not_found", Message: "no provider supplied"}
 	}
 
-	retryTimes, maxWait := m.retrySettings()
-	attempts := retryTimes + 1
-	if attempts < 1 {
-		attempts = 1
-	}
+	_, maxWait := m.retrySettings()
 
 	var lastErr error
-	for attempt := 0; attempt < attempts; attempt++ {
+	for attempt := 0; ; attempt++ {
 		resp, errExec := m.executeMixedOnce(ctx, normalized, req, opts)
 		if errExec == nil {
 			return resp, nil
 		}
 		lastErr = errExec
-		wait, shouldRetry := m.shouldRetryAfterError(errExec, attempt, attempts, normalized, req.Model, maxWait)
+		wait, shouldRetry := m.shouldRetryAfterError(errExec, attempt, normalized, req.Model, maxWait)
 		if !shouldRetry {
 			break
 		}
@@ -503,20 +508,16 @@ func (m *Manager) ExecuteCount(ctx context.Context, providers []string, req clip
 		return cliproxyexecutor.Response{}, &Error{Code: "provider_not_found", Message: "no provider supplied"}
 	}
 
-	retryTimes, maxWait := m.retrySettings()
-	attempts := retryTimes + 1
-	if attempts < 1 {
-		attempts = 1
-	}
+	_, maxWait := m.retrySettings()
 
 	var lastErr error
-	for attempt := 0; attempt < attempts; attempt++ {
+	for attempt := 0; ; attempt++ {
 		resp, errExec := m.executeCountMixedOnce(ctx, normalized, req, opts)
 		if errExec == nil {
 			return resp, nil
 		}
 		lastErr = errExec
-		wait, shouldRetry := m.shouldRetryAfterError(errExec, attempt, attempts, normalized, req.Model, maxWait)
+		wait, shouldRetry := m.shouldRetryAfterError(errExec, attempt, normalized, req.Model, maxWait)
 		if !shouldRetry {
 			break
 		}
@@ -538,20 +539,16 @@ func (m *Manager) ExecuteStream(ctx context.Context, providers []string, req cli
 		return nil, &Error{Code: "provider_not_found", Message: "no provider supplied"}
 	}
 
-	retryTimes, maxWait := m.retrySettings()
-	attempts := retryTimes + 1
-	if attempts < 1 {
-		attempts = 1
-	}
+	_, maxWait := m.retrySettings()
 
 	var lastErr error
-	for attempt := 0; attempt < attempts; attempt++ {
+	for attempt := 0; ; attempt++ {
 		chunks, errStream := m.executeStreamMixedOnce(ctx, normalized, req, opts)
 		if errStream == nil {
 			return chunks, nil
 		}
 		lastErr = errStream
-		wait, shouldRetry := m.shouldRetryAfterError(errStream, attempt, attempts, normalized, req.Model, maxWait)
+		wait, shouldRetry := m.shouldRetryAfterError(errStream, attempt, normalized, req.Model, maxWait)
 		if !shouldRetry {
 			break
 		}
@@ -570,6 +567,7 @@ func (m *Manager) executeMixedOnce(ctx context.Context, providers []string, req
 		return cliproxyexecutor.Response{}, &Error{Code: "provider_not_found", Message: "no provider supplied"}
 	}
 	routeModel := req.Model
+	opts = ensureRequestedModelMetadata(opts, routeModel)
 	tried := make(map[string]struct{})
 	var lastErr error
 	for {
@@ -597,6 +595,9 @@ func (m *Manager) executeMixedOnce(ctx context.Context, providers []string, req
 		resp, errExec := executor.Execute(execCtx, auth, execReq, opts)
 		result := Result{AuthID: auth.ID, Provider: provider, Model: routeModel, Success: errExec == nil}
 		if errExec != nil {
+			if errCtx := execCtx.Err(); errCtx != nil {
+				return cliproxyexecutor.Response{}, errCtx
+			}
 			result.Error = &Error{Message: errExec.Error()}
 			var se cliproxyexecutor.StatusError
 			if errors.As(errExec, &se) && se != nil {
@@ -619,6 +620,7 @@ func (m *Manager) executeCountMixedOnce(ctx context.Context, providers []string,
 		return cliproxyexecutor.Response{}, &Error{Code: "provider_not_found", Message: "no provider supplied"}
 	}
 	routeModel := req.Model
+	opts = ensureRequestedModelMetadata(opts, routeModel)
 	tried := make(map[string]struct{})
 	var lastErr error
 	for {
@@ -646,6 +648,9 @@ func (m *Manager) executeCountMixedOnce(ctx context.Context, providers []string,
 		resp, errExec := executor.CountTokens(execCtx, auth, execReq, opts)
 		result := Result{AuthID: auth.ID, Provider: provider, Model: routeModel, Success: errExec == nil}
 		if errExec != nil {
+			if errCtx := execCtx.Err(); errCtx != nil {
+				return cliproxyexecutor.Response{}, errCtx
+			}
 			result.Error = &Error{Message: errExec.Error()}
 			var se cliproxyexecutor.StatusError
 			if errors.As(errExec, &se) && se != nil {
@@ -668,6 +673,7 @@ func (m *Manager) executeStreamMixedOnce(ctx context.Context, providers []string
 		return nil, &Error{Code: "provider_not_found", Message: "no provider supplied"}
 	}
 	routeModel := req.Model
+	opts = ensureRequestedModelMetadata(opts, routeModel)
 	tried := make(map[string]struct{})
 	var lastErr error
 	for {
@@ -694,6 +700,9 @@ func (m *Manager) executeStreamMixedOnce(ctx context.Context, providers []string
 		execReq.Model = m.applyAPIKeyModelAlias(auth, execReq.Model)
 		chunks, errStream := executor.ExecuteStream(execCtx, auth, execReq, opts)
 		if errStream != nil {
+			if errCtx := execCtx.Err(); errCtx != nil {
+				return nil, errCtx
+			}
 			rerr := &Error{Message: errStream.Error()}
 			var se cliproxyexecutor.StatusError
 			if errors.As(errStream, &se) && se != nil {
@@ -709,6 +718,7 @@ func (m *Manager) executeStreamMixedOnce(ctx context.Context, providers []string
 		go func(streamCtx context.Context, streamAuth *Auth, streamProvider string, streamChunks <-chan cliproxyexecutor.StreamChunk) {
 			defer close(out)
 			var failed bool
+			forward := true
 			for chunk := range streamChunks {
 				if chunk.Err != nil && !failed {
 					failed = true
@@ -719,171 +729,18 @@ func (m *Manager) executeStreamMixedOnce(ctx context.Context, providers []string
 					}
 					m.MarkResult(streamCtx, Result{AuthID: streamAuth.ID, Provider: streamProvider, Model: routeModel, Success: false, Error: rerr})
 				}
-				out <- chunk
-			}
-			if !failed {
-				m.MarkResult(streamCtx, Result{AuthID: streamAuth.ID, Provider: streamProvider, Model: routeModel, Success: true})
-			}
-		}(execCtx, auth.Clone(), provider, chunks)
-		return out, nil
-	}
-}
-
-func (m *Manager) executeWithProvider(ctx context.Context, provider string, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
-	if provider == "" {
-		return cliproxyexecutor.Response{}, &Error{Code: "provider_not_found", Message: "provider identifier is empty"}
-	}
-	routeModel := req.Model
-	tried := make(map[string]struct{})
-	var lastErr error
-	for {
-		auth, executor, errPick := m.pickNext(ctx, provider, routeModel, opts, tried)
-		if errPick != nil {
-			if lastErr != nil {
-				return cliproxyexecutor.Response{}, lastErr
-			}
-			return cliproxyexecutor.Response{}, errPick
-		}
-
-		entry := logEntryWithRequestID(ctx)
-		debugLogAuthSelection(entry, auth, provider, req.Model)
-
-		tried[auth.ID] = struct{}{}
-		execCtx := ctx
-		if rt := m.roundTripperFor(auth); rt != nil {
-			execCtx = context.WithValue(execCtx, roundTripperContextKey{}, rt)
-			execCtx = context.WithValue(execCtx, "cliproxy.roundtripper", rt)
-		}
-		execReq := req
-		execReq.Model = rewriteModelForAuth(routeModel, auth)
-		execReq.Model = m.applyOAuthModelAlias(auth, execReq.Model)
-		execReq.Model = m.applyAPIKeyModelAlias(auth, execReq.Model)
-		resp, errExec := executor.Execute(execCtx, auth, execReq, opts)
-		result := Result{AuthID: auth.ID, Provider: provider, Model: routeModel, Success: errExec == nil}
-		if errExec != nil {
-			result.Error = &Error{Message: errExec.Error()}
-			var se cliproxyexecutor.StatusError
-			if errors.As(errExec, &se) && se != nil {
-				result.Error.HTTPStatus = se.StatusCode()
-			}
-			if ra := retryAfterFromError(errExec); ra != nil {
-				result.RetryAfter = ra
-			}
-			m.MarkResult(execCtx, result)
-			lastErr = errExec
-			continue
-		}
-		m.MarkResult(execCtx, result)
-		return resp, nil
-	}
-}
-
-func (m *Manager) executeCountWithProvider(ctx context.Context, provider string, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
-	if provider == "" {
-		return cliproxyexecutor.Response{}, &Error{Code: "provider_not_found", Message: "provider identifier is empty"}
-	}
-	routeModel := req.Model
-	tried := make(map[string]struct{})
-	var lastErr error
-	for {
-		auth, executor, errPick := m.pickNext(ctx, provider, routeModel, opts, tried)
-		if errPick != nil {
-			if lastErr != nil {
-				return cliproxyexecutor.Response{}, lastErr
-			}
-			return cliproxyexecutor.Response{}, errPick
-		}
-
-		entry := logEntryWithRequestID(ctx)
-		debugLogAuthSelection(entry, auth, provider, req.Model)
-
-		tried[auth.ID] = struct{}{}
-		execCtx := ctx
-		if rt := m.roundTripperFor(auth); rt != nil {
-			execCtx = context.WithValue(execCtx, roundTripperContextKey{}, rt)
-			execCtx = context.WithValue(execCtx, "cliproxy.roundtripper", rt)
-		}
-		execReq := req
-		execReq.Model = rewriteModelForAuth(routeModel, auth)
-		execReq.Model = m.applyOAuthModelAlias(auth, execReq.Model)
-		execReq.Model = m.applyAPIKeyModelAlias(auth, execReq.Model)
-		resp, errExec := executor.CountTokens(execCtx, auth, execReq, opts)
-		result := Result{AuthID: auth.ID, Provider: provider, Model: routeModel, Success: errExec == nil}
-		if errExec != nil {
-			result.Error = &Error{Message: errExec.Error()}
-			var se cliproxyexecutor.StatusError
-			if errors.As(errExec, &se) && se != nil {
-				result.Error.HTTPStatus = se.StatusCode()
-			}
-			if ra := retryAfterFromError(errExec); ra != nil {
-				result.RetryAfter = ra
-			}
-			m.MarkResult(execCtx, result)
-			lastErr = errExec
-			continue
-		}
-		m.MarkResult(execCtx, result)
-		return resp, nil
-	}
-}
-
-func (m *Manager) executeStreamWithProvider(ctx context.Context, provider string, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (<-chan cliproxyexecutor.StreamChunk, error) {
-	if provider == "" {
-		return nil, &Error{Code: "provider_not_found", Message: "provider identifier is empty"}
-	}
-	routeModel := req.Model
-	tried := make(map[string]struct{})
-	var lastErr error
-	for {
-		auth, executor, errPick := m.pickNext(ctx, provider, routeModel, opts, tried)
-		if errPick != nil {
-			if lastErr != nil {
-				return nil, lastErr
-			}
-			return nil, errPick
-		}
-
-		entry := logEntryWithRequestID(ctx)
-		debugLogAuthSelection(entry, auth, provider, req.Model)
-
-		tried[auth.ID] = struct{}{}
-		execCtx := ctx
-		if rt := m.roundTripperFor(auth); rt != nil {
-			execCtx = context.WithValue(execCtx, roundTripperContextKey{}, rt)
-			execCtx = context.WithValue(execCtx, "cliproxy.roundtripper", rt)
-		}
-		execReq := req
-		execReq.Model = rewriteModelForAuth(routeModel, auth)
-		execReq.Model = m.applyOAuthModelAlias(auth, execReq.Model)
-		execReq.Model = m.applyAPIKeyModelAlias(auth, execReq.Model)
-		chunks, errStream := executor.ExecuteStream(execCtx, auth, execReq, opts)
-		if errStream != nil {
-			rerr := &Error{Message: errStream.Error()}
-			var se cliproxyexecutor.StatusError
-			if errors.As(errStream, &se) && se != nil {
-				rerr.HTTPStatus = se.StatusCode()
-			}
-			result := Result{AuthID: auth.ID, Provider: provider, Model: routeModel, Success: false, Error: rerr}
-			result.RetryAfter = retryAfterFromError(errStream)
-			m.MarkResult(execCtx, result)
-			lastErr = errStream
-			continue
-		}
-		out := make(chan cliproxyexecutor.StreamChunk)
-		go func(streamCtx context.Context, streamAuth *Auth, streamProvider string, streamChunks <-chan cliproxyexecutor.StreamChunk) {
-			defer close(out)
-			var failed bool
-			for chunk := range streamChunks {
-				if chunk.Err != nil && !failed {
-					failed = true
-					rerr := &Error{Message: chunk.Err.Error()}
-					var se cliproxyexecutor.StatusError
-					if errors.As(chunk.Err, &se) && se != nil {
-						rerr.HTTPStatus = se.StatusCode()
-					}
-					m.MarkResult(streamCtx, Result{AuthID: streamAuth.ID, Provider: streamProvider, Model: routeModel, Success: false, Error: rerr})
+				if !forward {
+					continue
+				}
+				if streamCtx == nil {
+					out <- chunk
+					continue
+				}
+				select {
+				case <-streamCtx.Done():
+					forward = false
+				case out <- chunk:
 				}
-				out <- chunk
 			}
 			if !failed {
 				m.MarkResult(streamCtx, Result{AuthID: streamAuth.ID, Provider: streamProvider, Model: routeModel, Success: true})
@@ -893,6 +750,45 @@ func (m *Manager) executeStreamWithProvider(ctx context.Context, provider string
 	}
 }
 
+func ensureRequestedModelMetadata(opts cliproxyexecutor.Options, requestedModel string) cliproxyexecutor.Options {
+	requestedModel = strings.TrimSpace(requestedModel)
+	if requestedModel == "" {
+		return opts
+	}
+	if hasRequestedModelMetadata(opts.Metadata) {
+		return opts
+	}
+	if len(opts.Metadata) == 0 {
+		opts.Metadata = map[string]any{cliproxyexecutor.RequestedModelMetadataKey: requestedModel}
+		return opts
+	}
+	meta := make(map[string]any, len(opts.Metadata)+1)
+	for k, v := range opts.Metadata {
+		meta[k] = v
+	}
+	meta[cliproxyexecutor.RequestedModelMetadataKey] = requestedModel
+	opts.Metadata = meta
+	return opts
+}
+
+func hasRequestedModelMetadata(meta map[string]any) bool {
+	if len(meta) == 0 {
+		return false
+	}
+	raw, ok := meta[cliproxyexecutor.RequestedModelMetadataKey]
+	if !ok || raw == nil {
+		return false
+	}
+	switch v := raw.(type) {
+	case string:
+		return strings.TrimSpace(v) != ""
+	case []byte:
+		return strings.TrimSpace(string(v)) != ""
+	default:
+		return false
+	}
+}
+
 func rewriteModelForAuth(model string, auth *Auth) string {
 	if auth == nil || model == "" {
 		return model
@@ -1140,35 +1036,6 @@ func (m *Manager) normalizeProviders(providers []string) []string {
 	return result
 }
 
-// rotateProviders returns a rotated view of the providers list starting from the
-// current offset for the model, and atomically increments the offset for the next call.
-// This ensures concurrent requests get different starting providers.
-func (m *Manager) rotateProviders(model string, providers []string) []string {
-	if len(providers) == 0 {
-		return nil
-	}
-
-	// Atomic read-and-increment: get current offset and advance cursor in one lock
-	m.mu.Lock()
-	offset := m.providerOffsets[model]
-	m.providerOffsets[model] = (offset + 1) % len(providers)
-	m.mu.Unlock()
-
-	if len(providers) > 0 {
-		offset %= len(providers)
-	}
-	if offset < 0 {
-		offset = 0
-	}
-	if offset == 0 {
-		return providers
-	}
-	rotated := make([]string, 0, len(providers))
-	rotated = append(rotated, providers[offset:]...)
-	rotated = append(rotated, providers[:offset]...)
-	return rotated
-}
-
 func (m *Manager) retrySettings() (int, time.Duration) {
 	if m == nil {
 		return 0, 0
@@ -1176,11 +1043,15 @@ func (m *Manager) retrySettings() (int, time.Duration) {
 	return int(m.requestRetry.Load()), time.Duration(m.maxRetryInterval.Load())
 }
 
-func (m *Manager) closestCooldownWait(providers []string, model string) (time.Duration, bool) {
+func (m *Manager) closestCooldownWait(providers []string, model string, attempt int) (time.Duration, bool) {
 	if m == nil || len(providers) == 0 {
 		return 0, false
 	}
 	now := time.Now()
+	defaultRetry := int(m.requestRetry.Load())
+	if defaultRetry < 0 {
+		defaultRetry = 0
+	}
 	providerSet := make(map[string]struct{}, len(providers))
 	for i := range providers {
 		key := strings.TrimSpace(strings.ToLower(providers[i]))
@@ -1203,6 +1074,16 @@ func (m *Manager) closestCooldownWait(providers []string, model string) (time.Du
 		if _, ok := providerSet[providerKey]; !ok {
 			continue
 		}
+		effectiveRetry := defaultRetry
+		if override, ok := auth.RequestRetryOverride(); ok {
+			effectiveRetry = override
+		}
+		if effectiveRetry < 0 {
+			effectiveRetry = 0
+		}
+		if attempt >= effectiveRetry {
+			continue
+		}
 		blocked, reason, next := isAuthBlockedForModel(auth, model, now)
 		if !blocked || next.IsZero() || reason == blockReasonDisabled {
 			continue
@@ -1219,8 +1100,8 @@ func (m *Manager) closestCooldownWait(providers []string, model string) (time.Du
 	return minWait, found
 }
 
-func (m *Manager) shouldRetryAfterError(err error, attempt, maxAttempts int, providers []string, model string, maxWait time.Duration) (time.Duration, bool) {
-	if err == nil || attempt >= maxAttempts-1 {
+func (m *Manager) shouldRetryAfterError(err error, attempt int, providers []string, model string, maxWait time.Duration) (time.Duration, bool) {
+	if err == nil {
 		return 0, false
 	}
 	if maxWait <= 0 {
@@ -1229,7 +1110,7 @@ func (m *Manager) shouldRetryAfterError(err error, attempt, maxAttempts int, pro
 	if status := statusCodeFromError(err); status == http.StatusOK {
 		return 0, false
 	}
-	wait, found := m.closestCooldownWait(providers, model)
+	wait, found := m.closestCooldownWait(providers, model, attempt)
 	if !found || wait > maxWait {
 		return 0, false
 	}
@@ -1250,42 +1131,6 @@ func waitForCooldown(ctx context.Context, wait time.Duration) error {
 	}
 }
 
-func (m *Manager) executeProvidersOnce(ctx context.Context, providers []string, fn func(context.Context, string) (cliproxyexecutor.Response, error)) (cliproxyexecutor.Response, error) {
-	if len(providers) == 0 {
-		return cliproxyexecutor.Response{}, &Error{Code: "provider_not_found", Message: "no provider supplied"}
-	}
-	var lastErr error
-	for _, provider := range providers {
-		resp, errExec := fn(ctx, provider)
-		if errExec == nil {
-			return resp, nil
-		}
-		lastErr = errExec
-	}
-	if lastErr != nil {
-		return cliproxyexecutor.Response{}, lastErr
-	}
-	return cliproxyexecutor.Response{}, &Error{Code: "auth_not_found", Message: "no auth available"}
-}
-
-func (m *Manager) executeStreamProvidersOnce(ctx context.Context, providers []string, fn func(context.Context, string) (<-chan cliproxyexecutor.StreamChunk, error)) (<-chan cliproxyexecutor.StreamChunk, error) {
-	if len(providers) == 0 {
-		return nil, &Error{Code: "provider_not_found", Message: "no provider supplied"}
-	}
-	var lastErr error
-	for _, provider := range providers {
-		chunks, errExec := fn(ctx, provider)
-		if errExec == nil {
-			return chunks, nil
-		}
-		lastErr = errExec
-	}
-	if lastErr != nil {
-		return nil, lastErr
-	}
-	return nil, &Error{Code: "auth_not_found", Message: "no auth available"}
-}
-
 // MarkResult records an execution result and notifies hooks.
 func (m *Manager) MarkResult(ctx context.Context, result Result) {
 	if result.AuthID == "" {
@@ -1354,7 +1199,7 @@ func (m *Manager) MarkResult(ctx context.Context, result Result) {
 					if result.RetryAfter != nil {
 						next = now.Add(*result.RetryAfter)
 					} else {
-						cooldown, nextLevel := nextQuotaCooldown(backoffLevel)
+						cooldown, nextLevel := nextQuotaCooldown(backoffLevel, quotaCooldownDisabledForAuth(auth))
 						if cooldown > 0 {
 							next = now.Add(cooldown)
 						}
@@ -1371,8 +1216,12 @@ func (m *Manager) MarkResult(ctx context.Context, result Result) {
 					shouldSuspendModel = true
 					setModelQuota = true
 				case 408, 500, 502, 503, 504:
-					next := now.Add(1 * time.Minute)
-					state.NextRetryAfter = next
+					if quotaCooldownDisabledForAuth(auth) {
+						state.NextRetryAfter = time.Time{}
+					} else {
+						next := now.Add(1 * time.Minute)
+						state.NextRetryAfter = next
+					}
 				default:
 					state.NextRetryAfter = time.Time{}
 				}
@@ -1613,7 +1462,7 @@ func applyAuthFailureState(auth *Auth, resultErr *Error, retryAfter *time.Durati
 		if retryAfter != nil {
 			next = now.Add(*retryAfter)
 		} else {
-			cooldown, nextLevel := nextQuotaCooldown(auth.Quota.BackoffLevel)
+			cooldown, nextLevel := nextQuotaCooldown(auth.Quota.BackoffLevel, quotaCooldownDisabledForAuth(auth))
 			if cooldown > 0 {
 				next = now.Add(cooldown)
 			}
@@ -1623,7 +1472,11 @@ func applyAuthFailureState(auth *Auth, resultErr *Error, retryAfter *time.Durati
 		auth.NextRetryAfter = next
 	case 408, 500, 502, 503, 504:
 		auth.StatusMessage = "transient upstream error"
-		auth.NextRetryAfter = now.Add(1 * time.Minute)
+		if quotaCooldownDisabledForAuth(auth) {
+			auth.NextRetryAfter = time.Time{}
+		} else {
+			auth.NextRetryAfter = now.Add(1 * time.Minute)
+		}
 	default:
 		if auth.StatusMessage == "" {
 			auth.StatusMessage = "request failed"
@@ -1632,11 +1485,11 @@ func applyAuthFailureState(auth *Auth, resultErr *Error, retryAfter *time.Durati
 }
 
 // nextQuotaCooldown returns the next cooldown duration and updated backoff level for repeated quota errors.
-func nextQuotaCooldown(prevLevel int) (time.Duration, int) {
+func nextQuotaCooldown(prevLevel int, disableCooling bool) (time.Duration, int) {
 	if prevLevel < 0 {
 		prevLevel = 0
 	}
-	if quotaCooldownDisabled.Load() {
+	if disableCooling {
 		return 0, prevLevel
 	}
 	cooldown := quotaBackoffBase * time.Duration(1<<prevLevel)
@@ -1812,6 +1665,9 @@ func (m *Manager) persist(ctx context.Context, auth *Auth) error {
 	if m.store == nil || auth == nil {
 		return nil
 	}
+	if shouldSkipPersist(ctx) {
+		return nil
+	}
 	if auth.Attributes != nil {
 		if v := strings.ToLower(strings.TrimSpace(auth.Attributes["runtime_only"])); v == "true" {
 			return nil

sdk/cliproxy/auth/conductor_overrides_test.go

@@ -0,0 +1,97 @@
+package auth
+
+import (
+	"context"
+	"testing"
+	"time"
+)
+
+func TestManager_ShouldRetryAfterError_RespectsAuthRequestRetryOverride(t *testing.T) {
+	m := NewManager(nil, nil, nil)
+	m.SetRetryConfig(3, 30*time.Second)
+
+	model := "test-model"
+	next := time.Now().Add(5 * time.Second)
+
+	auth := &Auth{
+		ID:       "auth-1",
+		Provider: "claude",
+		Metadata: map[string]any{
+			"request_retry": float64(0),
+		},
+		ModelStates: map[string]*ModelState{
+			model: {
+				Unavailable:    true,
+				Status:         StatusError,
+				NextRetryAfter: next,
+			},
+		},
+	}
+	if _, errRegister := m.Register(context.Background(), auth); errRegister != nil {
+		t.Fatalf("register auth: %v", errRegister)
+	}
+
+	_, maxWait := m.retrySettings()
+	wait, shouldRetry := m.shouldRetryAfterError(&Error{HTTPStatus: 500, Message: "boom"}, 0, []string{"claude"}, model, maxWait)
+	if shouldRetry {
+		t.Fatalf("expected shouldRetry=false for request_retry=0, got true (wait=%v)", wait)
+	}
+
+	auth.Metadata["request_retry"] = float64(1)
+	if _, errUpdate := m.Update(context.Background(), auth); errUpdate != nil {
+		t.Fatalf("update auth: %v", errUpdate)
+	}
+
+	wait, shouldRetry = m.shouldRetryAfterError(&Error{HTTPStatus: 500, Message: "boom"}, 0, []string{"claude"}, model, maxWait)
+	if !shouldRetry {
+		t.Fatalf("expected shouldRetry=true for request_retry=1, got false")
+	}
+	if wait <= 0 {
+		t.Fatalf("expected wait > 0, got %v", wait)
+	}
+
+	_, shouldRetry = m.shouldRetryAfterError(&Error{HTTPStatus: 500, Message: "boom"}, 1, []string{"claude"}, model, maxWait)
+	if shouldRetry {
+		t.Fatalf("expected shouldRetry=false on attempt=1 for request_retry=1, got true")
+	}
+}
+
+func TestManager_MarkResult_RespectsAuthDisableCoolingOverride(t *testing.T) {
+	prev := quotaCooldownDisabled.Load()
+	quotaCooldownDisabled.Store(false)
+	t.Cleanup(func() { quotaCooldownDisabled.Store(prev) })
+
+	m := NewManager(nil, nil, nil)
+
+	auth := &Auth{
+		ID:       "auth-1",
+		Provider: "claude",
+		Metadata: map[string]any{
+			"disable_cooling": true,
+		},
+	}
+	if _, errRegister := m.Register(context.Background(), auth); errRegister != nil {
+		t.Fatalf("register auth: %v", errRegister)
+	}
+
+	model := "test-model"
+	m.MarkResult(context.Background(), Result{
+		AuthID:   "auth-1",
+		Provider: "claude",
+		Model:    model,
+		Success:  false,
+		Error:    &Error{HTTPStatus: 500, Message: "boom"},
+	})
+
+	updated, ok := m.GetByID("auth-1")
+	if !ok || updated == nil {
+		t.Fatalf("expected auth to be present")
+	}
+	state := updated.ModelStates[model]
+	if state == nil {
+		t.Fatalf("expected model state to be present")
+	}
+	if !state.NextRetryAfter.IsZero() {
+		t.Fatalf("expected NextRetryAfter to be zero when disable_cooling=true, got %v", state.NextRetryAfter)
+	}
+}

sdk/cliproxy/auth/persist_policy.go

@@ -0,0 +1,24 @@
+package auth
+
+import "context"
+
+type skipPersistContextKey struct{}
+
+// WithSkipPersist returns a derived context that disables persistence for Manager Update/Register calls.
+// It is intended for code paths that are reacting to file watcher events, where the file on disk is
+// already the source of truth and persisting again would create a write-back loop.
+func WithSkipPersist(ctx context.Context) context.Context {
+	if ctx == nil {
+		ctx = context.Background()
+	}
+	return context.WithValue(ctx, skipPersistContextKey{}, true)
+}
+
+func shouldSkipPersist(ctx context.Context) bool {
+	if ctx == nil {
+		return false
+	}
+	v := ctx.Value(skipPersistContextKey{})
+	enabled, ok := v.(bool)
+	return ok && enabled
+}

sdk/cliproxy/auth/persist_policy_test.go

@@ -0,0 +1,62 @@
+package auth
+
+import (
+	"context"
+	"sync/atomic"
+	"testing"
+)
+
+type countingStore struct {
+	saveCount atomic.Int32
+}
+
+func (s *countingStore) List(context.Context) ([]*Auth, error) { return nil, nil }
+
+func (s *countingStore) Save(context.Context, *Auth) (string, error) {
+	s.saveCount.Add(1)
+	return "", nil
+}
+
+func (s *countingStore) Delete(context.Context, string) error { return nil }
+
+func TestWithSkipPersist_DisablesUpdatePersistence(t *testing.T) {
+	store := &countingStore{}
+	mgr := NewManager(store, nil, nil)
+	auth := &Auth{
+		ID:       "auth-1",
+		Provider: "antigravity",
+		Metadata: map[string]any{"type": "antigravity"},
+	}
+
+	if _, err := mgr.Update(context.Background(), auth); err != nil {
+		t.Fatalf("Update returned error: %v", err)
+	}
+	if got := store.saveCount.Load(); got != 1 {
+		t.Fatalf("expected 1 Save call, got %d", got)
+	}
+
+	ctxSkip := WithSkipPersist(context.Background())
+	if _, err := mgr.Update(ctxSkip, auth); err != nil {
+		t.Fatalf("Update(skipPersist) returned error: %v", err)
+	}
+	if got := store.saveCount.Load(); got != 1 {
+		t.Fatalf("expected Save call count to remain 1, got %d", got)
+	}
+}
+
+func TestWithSkipPersist_DisablesRegisterPersistence(t *testing.T) {
+	store := &countingStore{}
+	mgr := NewManager(store, nil, nil)
+	auth := &Auth{
+		ID:       "auth-1",
+		Provider: "antigravity",
+		Metadata: map[string]any{"type": "antigravity"},
+	}
+
+	if _, err := mgr.Register(WithSkipPersist(context.Background()), auth); err != nil {
+		t.Fatalf("Register(skipPersist) returned error: %v", err)
+	}
+	if got := store.saveCount.Load(); got != 0 {
+		t.Fatalf("expected 0 Save calls, got %d", got)
+	}
+}

sdk/cliproxy/auth/types.go

@@ -194,6 +194,108 @@ func (a *Auth) ProxyInfo() string {
 	return "via proxy"
 }
 
+// DisableCoolingOverride returns the auth-file scoped disable_cooling override when present.
+// The value is read from metadata key "disable_cooling" (or legacy "disable-cooling").
+func (a *Auth) DisableCoolingOverride() (bool, bool) {
+	if a == nil || a.Metadata == nil {
+		return false, false
+	}
+	if val, ok := a.Metadata["disable_cooling"]; ok {
+		if parsed, okParse := parseBoolAny(val); okParse {
+			return parsed, true
+		}
+	}
+	if val, ok := a.Metadata["disable-cooling"]; ok {
+		if parsed, okParse := parseBoolAny(val); okParse {
+			return parsed, true
+		}
+	}
+	return false, false
+}
+
+// RequestRetryOverride returns the auth-file scoped request_retry override when present.
+// The value is read from metadata key "request_retry" (or legacy "request-retry").
+func (a *Auth) RequestRetryOverride() (int, bool) {
+	if a == nil || a.Metadata == nil {
+		return 0, false
+	}
+	if val, ok := a.Metadata["request_retry"]; ok {
+		if parsed, okParse := parseIntAny(val); okParse {
+			if parsed < 0 {
+				parsed = 0
+			}
+			return parsed, true
+		}
+	}
+	if val, ok := a.Metadata["request-retry"]; ok {
+		if parsed, okParse := parseIntAny(val); okParse {
+			if parsed < 0 {
+				parsed = 0
+			}
+			return parsed, true
+		}
+	}
+	return 0, false
+}
+
+func parseBoolAny(val any) (bool, bool) {
+	switch typed := val.(type) {
+	case bool:
+		return typed, true
+	case string:
+		trimmed := strings.TrimSpace(typed)
+		if trimmed == "" {
+			return false, false
+		}
+		parsed, err := strconv.ParseBool(trimmed)
+		if err != nil {
+			return false, false
+		}
+		return parsed, true
+	case float64:
+		return typed != 0, true
+	case json.Number:
+		parsed, err := typed.Int64()
+		if err != nil {
+			return false, false
+		}
+		return parsed != 0, true
+	default:
+		return false, false
+	}
+}
+
+func parseIntAny(val any) (int, bool) {
+	switch typed := val.(type) {
+	case int:
+		return typed, true
+	case int32:
+		return int(typed), true
+	case int64:
+		return int(typed), true
+	case float64:
+		return int(typed), true
+	case json.Number:
+		parsed, err := typed.Int64()
+		if err != nil {
+			return 0, false
+		}
+		return int(parsed), true
+	case string:
+		trimmed := strings.TrimSpace(typed)
+		if trimmed == "" {
+			return 0, false
+		}
+		parsed, err := strconv.Atoi(trimmed)
+		if err != nil {
+			return 0, false
+		}
+		return parsed, true
+	default:
+		return 0, false
+	}
+}
+
 func (a *Auth) AccountInfo() (string, string) {
 	if a == nil {
 		return "", ""

sdk/cliproxy/executor/types.go

@@ -7,6 +7,9 @@ import (
 	sdktranslator "github.com/router-for-me/CLIProxyAPI/v6/sdk/translator"
 )
 
+// RequestedModelMetadataKey stores the client-requested model name in Options.Metadata.
+const RequestedModelMetadataKey = "requested_model"
+
 // Request encapsulates the translated payload that will be sent to a provider executor.
 type Request struct {
 	// Model is the upstream model identifier after translation.

sdk/cliproxy/service.go

@@ -124,6 +124,7 @@ func (s *Service) ensureAuthUpdateQueue(ctx context.Context) {
 }
 
 func (s *Service) consumeAuthUpdates(ctx context.Context) {
+	ctx = coreauth.WithSkipPersist(ctx)
 	for {
 		select {
 		case <-ctx.Done():
@@ -680,6 +681,10 @@ func (s *Service) registerModelsForAuth(a *coreauth.Auth) {
 	if a == nil || a.ID == "" {
 		return
 	}
+	if a.Disabled {
+		GlobalModelRegistry().UnregisterClient(a.ID)
+		return
+	}
 	authKind := strings.ToLower(strings.TrimSpace(a.Attributes["auth_kind"]))
 	if authKind == "" {
 		if kind, _ := a.AccountInfo(); strings.EqualFold(kind, "api_key") {

test/thinking_conversion_test.go

@@ -2,6 +2,7 @@ package test
 
 import (
 	"fmt"
+	"strings"
 	"testing"
 	"time"
 
@@ -2778,12 +2779,18 @@ func runThinkingTests(t *testing.T, cases []thinkingTestCase) {
 
 			// Verify clear_thinking for iFlow GLM models when enable_thinking=true
 			if tc.to == "iflow" && tc.expectField == "chat_template_kwargs.enable_thinking" && tc.expectValue == "true" {
+				baseModel := thinking.ParseSuffix(tc.model).ModelName
+				isGLM := strings.HasPrefix(strings.ToLower(baseModel), "glm")
 				ctVal := gjson.GetBytes(body, "chat_template_kwargs.clear_thinking")
-				if !ctVal.Exists() {
-					t.Fatalf("expected clear_thinking field not found for GLM model, body=%s", string(body))
-				}
-				if ctVal.Bool() != false {
-					t.Fatalf("clear_thinking: expected false, got %v, body=%s", ctVal.Bool(), string(body))
+				if isGLM {
+					if !ctVal.Exists() {
+						t.Fatalf("expected clear_thinking field not found for GLM model, body=%s", string(body))
+					}
+					if ctVal.Bool() != false {
+						t.Fatalf("clear_thinking: expected false, got %v, body=%s", ctVal.Bool(), string(body))
+					}
+				} else if ctVal.Exists() {
+					t.Fatalf("expected no clear_thinking field for non-GLM enable_thinking model, body=%s", string(body))
 				}
 			}
 		})