Update workflows and build process for enhanced metadata injection

- Upgraded GitHub Actions (`actions/checkout` to v4, `actions/setup-go` to v4, `goreleaser-action` to v4).
- Added detailed build metadata (`VERSION`, `COMMIT`, `BUILD_DATE`) to workflows.
- Unified metadata injection into binaries and Docker images.
- Enhanced `.goreleaser.yml` with checksum, snapshot, and changelog configurations.

.dockerignore

@@ -0,0 +1,25 @@
+# Git and GitHub folders
+.git
+.github
+
+# Docker and CI/CD related files
+docker-compose.yml
+.dockerignore
+.gitignore
+.goreleaser.yml
+Dockerfile
+
+# Documentation and license
+README.md
+README_CN.md
+MANAGEMENT_API.md
+MANAGEMENT_API_CN.md
+LICENSE
+
+# Example configuration
+config.example.yaml
+
+# Runtime data folders (should be mounted as volumes)
+auths
+logs
+config.yaml

.github/workflows/docker-image.yml

@@ -24,8 +24,11 @@ jobs:
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_TOKEN }}
-      - name: Generate App Version
-        run: echo APP_VERSION=`git describe --tags --always` >> $GITHUB_ENV
+      - 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` >> $GITHUB_ENV
       - name: Build and push
         uses: docker/build-push-action@v6
         with:
@@ -35,8 +38,9 @@ jobs:
             linux/arm64
           push: true
           build-args: |
-            APP_NAME=${{ env.APP_NAME }}
-            APP_VERSION=${{ env.APP_VERSION }}
+            VERSION=${{ env.VERSION }}
+            COMMIT=${{ env.COMMIT }}
+            BUILD_DATE=${{ env.BUILD_DATE }}
           tags: |
             ${{ env.DOCKERHUB_REPO }}:latest
-            ${{ env.DOCKERHUB_REPO }}:${{ env.APP_VERSION }}
+            ${{ env.DOCKERHUB_REPO }}:${{ env.VERSION }}

.github/workflows/release.yaml

@@ -13,18 +13,26 @@ jobs:
   goreleaser:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v4
         with:
           fetch-depth: 0
       - run: git fetch --force --tags
-      - uses: actions/setup-go@v3
+      - uses: actions/setup-go@v4
         with:
           go-version: '>=1.24.0'
           cache: true
-      - uses: goreleaser/goreleaser-action@v3
+      - 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
+      - uses: goreleaser/goreleaser-action@v4
         with:
           distribution: goreleaser
           version: latest
           args: release --clean
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          VERSION: ${{ env.VERSION }}
+          COMMIT: ${{ env.COMMIT }}
+          BUILD_DATE: ${{ env.BUILD_DATE }}

.gitignore

@@ -1,3 +1,5 @@
 config.yaml
-docs/
-logs/
+docs/*
+logs/*
+auths/*
+!auths/.gitkeep

.goreleaser.yml

@@ -9,6 +9,8 @@ builds:
       - arm64
     main: ./cmd/server/
     binary: cli-proxy-api
+    ldflags:
+      - -s -w -X 'main.Version={{.Version}}' -X 'main.Commit={{.ShortCommit}}' -X 'main.BuildDate={{.Date}}'
 archives:
   - id: "cli-proxy-api"
     format: tar.gz
@@ -19,4 +21,17 @@ archives:
       - LICENSE
       - README.md
       - README_CN.md
-      - config.example.yaml
+      - config.example.yaml
+
+checksum:
+  name_template: 'checksums.txt'
+
+snapshot:
+  name_template: "{{ incpatch .Version }}-next"
+
+changelog:
+  sort: asc
+  filters:
+    exclude:
+      - '^docs:'
+      - '^test:'

Dockerfile

@@ -8,7 +8,11 @@ RUN go mod download
 
 COPY . .
 
-RUN CGO_ENABLED=0 GOOS=linux go build -o ./CLIProxyAPI ./cmd/server/
+ARG VERSION=dev
+ARG COMMIT=none
+ARG BUILD_DATE=unknown
+
+RUN CGO_ENABLED=0 GOOS=linux go build -ldflags="-s -w -X 'main.Version=${VERSION}' -X 'main.Commit=${COMMIT}' -X 'main.BuildDate=${BUILD_DATE}'" -o ./CLIProxyAPI ./cmd/server/
 
 FROM alpine:3.22.0
 

README.md

@@ -220,6 +220,7 @@ console.log(await claudeResponse.json());
 
 - gemini-2.5-pro
 - gemini-2.5-flash
+- gemini-2.5-flash-lite
 - gpt-5
 - claude-opus-4-1-20250805
 - claude-opus-4-20250514
@@ -254,6 +255,9 @@ The server uses a YAML configuration file (`config.yaml`) located in the project
 | `debug`                                 | boolean  | false              | Enable debug mode for verbose logging.                                                                                                                                                    |
 | `api-keys`                              | string[] | []                 | List of API keys that can be used to authenticate requests.                                                                                                                               |
 | `generative-language-api-key`           | string[] | []                 | List of Generative Language API keys.                                                                                                                                                     |
+| `codex-api-key`                         | object   | {}                 | List of Codex API keys.                                                                                                                                                                   |
+| `codex-api-key.api-key`                 | string   | ""                 | Codex API key.                                                                                                                                                                            |
+| `codex-api-key.base-url`                | string   | ""                 | Custom Codex API endpoint, if you use a third-party API endpoint.                                                                                                                         |
 | `claude-api-key`                        | object   | {}                 | List of Claude API keys.                                                                                                                                                                  |
 | `claude-api-key.api-key`                | string   | ""                 | Claude API key.                                                                                                                                                                           |
 | `claude-api-key.base-url`               | string   | ""                 | Custom Claude API endpoint, if you use a third-party API endpoint.                                                                                                                        |
@@ -310,6 +314,11 @@ generative-language-api-key:
   - "AIzaSy...02"
   - "AIzaSy...03"
   - "AIzaSy...04"
+
+# Codex API keys
+codex-api-key:
+  - api-key: "sk-atSM..."
+    base-url: "https://www.example.com" # use the custom codex API endpoint
   
 # Claude API keys
 claude-api-key:
@@ -486,6 +495,36 @@ Run the following command to start the server:
 docker run --rm -p 8317:8317 -v /path/to/your/config.yaml:/CLIProxyAPI/config.yaml -v /path/to/your/auth-dir:/root/.cli-proxy-api eceasy/cli-proxy-api:latest
 ```
 
+## Run with Docker Compose
+
+1.  Create a `config.yaml` from `config.example.yaml` and customize it.
+
+2.  Build and start the services using the build scripts:
+    - For Windows (PowerShell):
+      ```powershell
+      ./docker-build.ps1
+      ```
+    - For Linux/macOS:
+      ```bash
+      bash docker-build.sh
+      ```
+
+3.  To authenticate with providers, run the login command inside the container:
+    - **Gemini**: `docker compose exec cli-proxy-api /CLIProxyAPI/CLIProxyAPI -no-browser --login`
+    - **OpenAI (Codex)**: `docker compose exec cli-proxy-api /CLIProxyAPI/CLIProxyAPI -no-browser --codex-login`
+    - **Claude**: `docker compose exec cli-proxy-api /CLIProxyAPI/CLIProxyAPI -no-browser --claude-login`
+    - **Qwen**: `docker compose exec cli-proxy-api /CLIProxyAPI/CLIProxyAPI -no-browser --qwen-login`
+
+4.  To view the server logs:
+    ```bash
+    docker compose logs -f
+    ```
+
+5.  To stop the application:
+    ```bash
+    docker compose down
+    ```
+
 ## Management API
 
 see [MANAGEMENT_API.md](MANAGEMENT_API.md)

README_CN.md

@@ -237,6 +237,7 @@ console.log(await claudeResponse.json());
 
 - gemini-2.5-pro
 - gemini-2.5-flash
+- gemini-2.5-flash-lite
 - gpt-5
 - claude-opus-4-1-20250805
 - claude-opus-4-20250514
@@ -271,6 +272,9 @@ console.log(await claudeResponse.json());
 | `debug`                                 | boolean  | false              | 启用调试模式以获取详细日志。                                                      |
 | `api-keys`                              | string[] | []                 | 可用于验证请求的API密钥列表。                                                    |
 | `generative-language-api-key`           | string[] | []                 | 生成式语言API密钥列表。                                                       |
+| `codex-api-key`                         | object   | {}                 | Codex API密钥列表。                                                      |
+| `codex-api-key.api-key`                 | string   | ""                 | Codex API密钥。                                                        |
+| `codex-api-key.base-url`                | string   | ""                 | 自定义的Codex API端点                                                     |
 | `claude-api-key`                        | object   | {}                 | Claude API密钥列表。                                                     |
 | `claude-api-key.api-key`                | string   | ""                 | Claude API密钥。                                                       |
 | `claude-api-key.base-url`               | string   | ""                 | 自定义的Claude API端点，如果您使用第三方的API端点。                                    |
@@ -328,11 +332,16 @@ generative-language-api-key:
   - "AIzaSy...03"
   - "AIzaSy...04"
 
-# Claude API keys
-claude-api-key:
-  - api-key: "sk-atSM..." # use the official claude API key, no need to set the base url
+# Codex API 密钥
+codex-api-key:
   - api-key: "sk-atSM..."
-    base-url: "https://www.example.com" # use the custom claude API endpoint
+    base-url: "https://www.example.com" # 第三方 Codex API 中转服务端点
+
+# Claude API 密钥
+claude-api-key:
+  - api-key: "sk-atSM..." # 如果使用官方 Claude API，无需设置 base-url
+  - api-key: "sk-atSM..."
+    base-url: "https://www.example.com" # 第三方 Claude API 中转服务端点
 
 # OpenAI 兼容提供商
 openai-compatibility:
@@ -499,6 +508,36 @@ docker run -it -rm -v /path/to/your/config.yaml:/CLIProxyAPI/config.yaml -v /pat
 docker run --rm -p 8317:8317 -v /path/to/your/config.yaml:/CLIProxyAPI/config.yaml -v /path/to/your/auth-dir:/root/.cli-proxy-api eceasy/cli-proxy-api:latest
 ```
 
+## 使用 Docker Compose 运行
+
+1.  从 `config.example.yaml` 创建一个 `config.yaml` 文件并进行自定义。
+
+2.  使用构建脚本构建并启动服务：
+    - Windows (PowerShell):
+      ```powershell
+      ./docker-build.ps1
+      ```
+    - Linux/macOS:
+      ```bash
+      bash docker-build.sh
+      ```
+
+3.  要在容器内运行登录命令进行身份验证：
+    - **Gemini**: `docker compose exec cli-proxy-api /CLIProxyAPI/CLIProxyAPI -no-browser --login`
+    - **OpenAI (Codex)**: `docker compose exec cli-proxy-api /CLIProxyAPI/CLIProxyAPI -no-browser --codex-login`
+    - **Claude**: `docker compose exec cli-proxy-api /CLIProxyAPI/CLIProxyAPI -no-browser --claude-login`
+    - **Qwen**: `docker compose exec cli-proxy-api /CLIProxyAPI/CLIProxyAPI -no-browser --qwen-login`
+
+4.  查看服务器日志：
+    ```bash
+    docker compose logs -f
+    ```
+
+5.  停止应用程序：
+    ```bash
+    docker compose down
+    ```
+
 ## 管理 API 文档
 
 请参见 [MANAGEMENT_API_CN.md](MANAGEMENT_API_CN.md)

cmd/server/main.go

@@ -8,7 +8,7 @@ import (
 	"flag"
 	"fmt"
 	"os"
-	"path"
+	"path/filepath"
 	"strings"
 
 	"github.com/luispater/CLIProxyAPI/internal/cmd"
@@ -17,6 +17,12 @@ import (
 	log "github.com/sirupsen/logrus"
 )
 
+var (
+	Version   = "dev"
+	Commit    = "none"
+	BuildDate = "unknown"
+)
+
 // LogFormatter defines a custom log format for logrus.
 // This formatter adds timestamp, log level, and source location information
 // to each log entry for better debugging and monitoring.
@@ -36,7 +42,7 @@ func (m *LogFormatter) Format(entry *log.Entry) ([]byte, error) {
 	timestamp := entry.Time.Format("2006-01-02 15:04:05")
 	var newLog string
 	// Customize the log format to include timestamp, level, caller file/line, and message.
-	newLog = fmt.Sprintf("[%s] [%s] [%s:%d] %s\n", timestamp, entry.Level, path.Base(entry.Caller.File), entry.Caller.Line, entry.Message)
+	newLog = fmt.Sprintf("[%s] [%s] [%s:%d] %s\n", timestamp, entry.Level, filepath.Base(entry.Caller.File), entry.Caller.Line, entry.Message)
 
 	b.WriteString(newLog)
 	return b.Bytes(), nil
@@ -58,6 +64,8 @@ func init() {
 // It parses command-line flags, loads configuration, and starts the appropriate
 // service based on the provided flags (login, codex-login, or server mode).
 func main() {
+	log.Infof("CLIProxyAPI Version: %s, Commit: %s, BuiltAt: %s", Version, Commit, BuildDate)
+
 	// Command-line flags to control the application's behavior.
 	var login bool
 	var codexLogin bool
@@ -96,7 +104,7 @@ func main() {
 		if err != nil {
 			log.Fatalf("failed to get working directory: %v", err)
 		}
-		configFilePath = path.Join(wd, "config.yaml")
+		configFilePath = filepath.Join(wd, "config.yaml")
 		cfg, err = config.LoadConfig(configFilePath)
 	}
 	if err != nil {
@@ -120,7 +128,7 @@ func main() {
 		parts := strings.Split(cfg.AuthDir, string(os.PathSeparator))
 		if len(parts) > 1 {
 			parts[0] = home
-			cfg.AuthDir = path.Join(parts...)
+			cfg.AuthDir = filepath.Join(parts...)
 		} else {
 			// If the path is just "~", set it to the home directory.
 			cfg.AuthDir = home

config.example.yaml

@@ -41,6 +41,11 @@ generative-language-api-key:
   - "AIzaSy...03"
   - "AIzaSy...04"
 
+# Codex API keys
+codex-api-key:
+  - api-key: "sk-atSM..."
+    base-url: "https://www.example.com" # use the custom codex API endpoint
+
 # Claude API keys
 claude-api-key:
   - api-key: "sk-atSM..." # use the official claude API key, no need to set the base url

docker-build.ps1

@@ -0,0 +1,36 @@
+# build.ps1 - Windows PowerShell Build Script
+#
+# This script automates the process of building and running the Docker container
+# with version information dynamically injected at build time.
+
+# Stop script execution on any error
+$ErrorActionPreference = "Stop"
+
+# --- Step 1: Get Version Information ---
+# Get the latest git tag or commit hash as the version string.
+$VERSION = (git describe --tags --always --dirty)
+
+# Get the short commit hash.
+$COMMIT  = (git rev-parse --short HEAD)
+
+# Get the current UTC date and time in ISO 8601 format.
+$BUILD_DATE = (Get-Date).ToUniversalTime().ToString("yyyy-MM-ddTHH:mm:ssZ")
+
+Write-Host "--- Building with the following info ---"
+Write-Host "Version: $VERSION"
+Write-Host "Commit: $COMMIT"
+Write-Host "Build Date: $BUILD_DATE"
+Write-Host "----------------------------------------"
+
+# --- Step 2: Build the Docker Image ---
+# Pass the version information as build arguments to 'docker compose build'.
+# These arguments are then used by the Dockerfile to inject them into the Go binary.
+docker compose build --build-arg VERSION=$VERSION --build-arg COMMIT=$COMMIT --build-arg BUILD_DATE=$BUILD_DATE
+
+# --- Step 3: Start the Services ---
+# Start the services in detached mode using the newly built image.
+# '--remove-orphans' cleans up any containers for services that are no longer defined.
+docker compose up -d --remove-orphans
+
+Write-Host "Build complete. Services are starting."
+Write-Host "Run 'docker compose logs -f' to see the logs."

docker-build.sh

@@ -0,0 +1,41 @@
+#!/usr/bin/env bash
+#
+# build.sh - Linux/macOS Build Script
+#
+# This script automates the process of building and running the Docker container
+# with version information dynamically injected at build time.
+
+# Exit immediately if a command exits with a non-zero status.
+set -euo pipefail
+
+# --- Step 1: Get Version Information ---
+# Get the latest git tag or commit hash as the version string.
+VERSION="$(git describe --tags --always --dirty)"
+
+# Get the short commit hash.
+COMMIT="$(git rev-parse --short HEAD)"
+
+# Get the current UTC date and time in ISO 8601 format.
+BUILD_DATE="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
+
+echo "--- Building with the following info ---"
+echo "Version: ${VERSION}"
+echo "Commit: ${COMMIT}"
+echo "Build Date: ${BUILD_DATE}"
+echo "----------------------------------------"
+
+# --- Step 2: Build the Docker Image ---
+# Pass the version information as build arguments to 'docker compose build'.
+# These arguments are then used by the Dockerfile to inject them into the Go binary.
+docker compose build \
+  --build-arg VERSION="${VERSION}" \
+  --build-arg COMMIT="${COMMIT}" \
+  --build-arg BUILD_DATE="${BUILD_DATE}"
+
+# --- Step 3: Start the Services ---
+# Start the services in detached mode using the newly built image.
+# '--remove-orphans' cleans up any containers for services that are no longer defined.
+docker compose up -d --remove-orphans
+
+echo "Build complete. Services are starting."
+echo "Run 'docker compose logs -f' to see the logs."

docker-compose.yml

@@ -0,0 +1,21 @@
+services:
+  cli-proxy-api:
+    build:
+      context: .
+      dockerfile: Dockerfile
+      args:
+        VERSION: ${VERSION:-dev}
+        COMMIT: ${COMMIT:-none}
+        BUILD_DATE: ${BUILD_DATE:-unknown}
+    image: cli-proxy-api:latest
+    container_name: cli-proxy-api
+    ports:
+      - "8317:8317"
+      - "8085:8085"
+      - "1455:1455"
+      - "54545:54545"
+    volumes:
+      - ./config.yaml:/CLIProxyAPI/config.yaml
+      - ./auths:/root/.cli-proxy-api
+      - ./logs:/CLIProxyAPI/logs
+    restart: unless-stopped

internal/api/handlers/claude/code_handlers.go

@@ -139,12 +139,13 @@ func (h *ClaudeCodeAPIHandler) handleStreamingResponse(c *gin.Context, rawJSON [
 			}
 		}
 	}()
+
+	var errorResponse *interfaces.ErrorMessage
 	retryCount := 0
 	// Main client rotation loop with quota management
 	// This loop implements a sophisticated load balancing and failover mechanism
 outLoop:
 	for retryCount <= h.Cfg.RequestRetry {
-		var errorResponse *interfaces.ErrorMessage
 		cliClient, errorResponse = h.GetClient(modelName)
 		if errorResponse != nil {
 			c.Status(errorResponse.StatusCode)
@@ -185,6 +186,8 @@ outLoop:
 			// This manages various error conditions and implements retry logic
 			case errInfo, okError := <-errChan:
 				if okError {
+					errorResponse = errInfo
+					h.LoggingAPIResponseError(cliCtx, errInfo)
 					// Special handling for quota exceeded errors
 					// If configured, attempt to switch to a different project/client
 					switch errInfo.StatusCode {
@@ -221,4 +224,12 @@ outLoop:
 			}
 		}
 	}
+
+	if errorResponse != nil {
+		c.Status(errorResponse.StatusCode)
+		_, _ = fmt.Fprint(c.Writer, errorResponse.Error.Error())
+		flusher.Flush()
+		cliCancel(errorResponse.Error)
+		return
+	}
 }

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

@@ -169,10 +169,10 @@ func (h *GeminiCLIAPIHandler) handleInternalStreamGenerateContent(c *gin.Context
 		}
 	}()
 
+	var errorResponse *interfaces.ErrorMessage
 	retryCount := 0
 outLoop:
 	for retryCount <= h.Cfg.RequestRetry {
-		var errorResponse *interfaces.ErrorMessage
 		cliClient, errorResponse = h.GetClient(modelName)
 		if errorResponse != nil {
 			c.Status(errorResponse.StatusCode)
@@ -208,6 +208,9 @@ outLoop:
 			// Handle errors from the backend.
 			case err, okError := <-errChan:
 				if okError {
+					errorResponse = err
+					h.LoggingAPIResponseError(cliCtx, err)
+
 					switch err.StatusCode {
 					case 429:
 						if h.Cfg.QuotaExceeded.SwitchProject {
@@ -232,6 +235,13 @@ outLoop:
 			}
 		}
 	}
+	if errorResponse != nil {
+		c.Status(errorResponse.StatusCode)
+		_, _ = fmt.Fprint(c.Writer, errorResponse.Error.Error())
+		flusher.Flush()
+		cliCancel(errorResponse.Error)
+		return
+	}
 }
 
 // handleInternalGenerateContent handles non-streaming content generation requests.
@@ -252,9 +262,9 @@ func (h *GeminiCLIAPIHandler) handleInternalGenerateContent(c *gin.Context, rawJ
 		}
 	}()
 
+	var errorResponse *interfaces.ErrorMessage
 	retryCount := 0
 	for retryCount <= h.Cfg.RequestRetry {
-		var errorResponse *interfaces.ErrorMessage
 		cliClient, errorResponse = h.GetClient(modelName)
 		if errorResponse != nil {
 			c.Status(errorResponse.StatusCode)
@@ -265,6 +275,9 @@ func (h *GeminiCLIAPIHandler) handleInternalGenerateContent(c *gin.Context, rawJ
 
 		resp, err := cliClient.SendRawMessage(cliCtx, modelName, rawJSON, "")
 		if err != nil {
+			errorResponse = err
+			h.LoggingAPIResponseError(cliCtx, err)
+
 			switch err.StatusCode {
 			case 429:
 				if h.Cfg.QuotaExceeded.SwitchProject {
@@ -296,4 +309,11 @@ func (h *GeminiCLIAPIHandler) handleInternalGenerateContent(c *gin.Context, rawJ
 			break
 		}
 	}
+	if errorResponse != nil {
+		c.Status(errorResponse.StatusCode)
+		_, _ = c.Writer.Write([]byte(errorResponse.Error.Error()))
+		cliCancel(errorResponse.Error)
+		return
+	}
+
 }

internal/api/handlers/gemini/gemini_handlers.go

@@ -221,10 +221,10 @@ func (h *GeminiAPIHandler) handleStreamGenerateContent(c *gin.Context, modelName
 		}
 	}()
 
+	var errorResponse *interfaces.ErrorMessage
 	retryCount := 0
 outLoop:
 	for retryCount <= h.Cfg.RequestRetry {
-		var errorResponse *interfaces.ErrorMessage
 		cliClient, errorResponse = h.GetClient(modelName)
 		if errorResponse != nil {
 			c.Status(errorResponse.StatusCode)
@@ -263,6 +263,9 @@ outLoop:
 			// Handle errors from the backend.
 			case err, okError := <-errChan:
 				if okError {
+					errorResponse = err
+					h.LoggingAPIResponseError(cliCtx, err)
+
 					switch err.StatusCode {
 					case 429:
 						if h.Cfg.QuotaExceeded.SwitchProject {
@@ -287,6 +290,13 @@ outLoop:
 			}
 		}
 	}
+	if errorResponse != nil {
+		c.Status(errorResponse.StatusCode)
+		_, _ = fmt.Fprint(c.Writer, errorResponse.Error.Error())
+		flusher.Flush()
+		cliCancel(errorResponse.Error)
+		return
+	}
 }
 
 // handleCountTokens handles token counting requests for Gemini models.
@@ -365,9 +375,9 @@ func (h *GeminiAPIHandler) handleGenerateContent(c *gin.Context, modelName strin
 		}
 	}()
 
+	var errorResponse *interfaces.ErrorMessage
 	retryCount := 0
 	for retryCount <= h.Cfg.RequestRetry {
-		var errorResponse *interfaces.ErrorMessage
 		cliClient, errorResponse = h.GetClient(modelName)
 		if errorResponse != nil {
 			c.Status(errorResponse.StatusCode)
@@ -378,6 +388,9 @@ func (h *GeminiAPIHandler) handleGenerateContent(c *gin.Context, modelName strin
 
 		resp, err := cliClient.SendRawMessage(cliCtx, modelName, rawJSON, alt)
 		if err != nil {
+			errorResponse = err
+			h.LoggingAPIResponseError(cliCtx, err)
+
 			switch err.StatusCode {
 			case 429:
 				if h.Cfg.QuotaExceeded.SwitchProject {
@@ -409,4 +422,10 @@ func (h *GeminiAPIHandler) handleGenerateContent(c *gin.Context, modelName strin
 			break
 		}
 	}
+	if errorResponse != nil {
+		c.Status(errorResponse.StatusCode)
+		_, _ = c.Writer.Write([]byte(errorResponse.Error.Error()))
+		cliCancel(errorResponse.Error)
+		return
+	}
 }

internal/api/handlers/handlers.go

@@ -235,6 +235,22 @@ func (h *BaseAPIHandler) GetContextWithCancel(handler interfaces.APIHandler, c *
 	}
 }
 
+func (h *BaseAPIHandler) LoggingAPIResponseError(ctx context.Context, err *interfaces.ErrorMessage) {
+	if h.Cfg.RequestLog {
+		if ginContext, ok := ctx.Value("gin").(*gin.Context); ok {
+			if apiResponseErrors, isExist := ginContext.Get("API_RESPONSE_ERROR"); isExist {
+				if slicesAPIResponseError, isOk := apiResponseErrors.([]*interfaces.ErrorMessage); isOk {
+					slicesAPIResponseError = append(slicesAPIResponseError, err)
+					ginContext.Set("API_RESPONSE_ERROR", slicesAPIResponseError)
+				}
+			} else {
+				// Create new response data entry
+				ginContext.Set("API_RESPONSE_ERROR", []*interfaces.ErrorMessage{err})
+			}
+		}
+	}
+}
+
 // APIHandlerCancelFunc is a function type for canceling an API handler's context.
 // It can optionally accept parameters, which are used for logging the response.
 type APIHandlerCancelFunc func(params ...interface{})

internal/api/handlers/openai/openai_handlers.go

@@ -387,9 +387,9 @@ func (h *OpenAIAPIHandler) handleNonStreamingResponse(c *gin.Context, rawJSON []
 		}
 	}()
 
+	var errorResponse *interfaces.ErrorMessage
 	retryCount := 0
 	for retryCount <= h.Cfg.RequestRetry {
-		var errorResponse *interfaces.ErrorMessage
 		cliClient, errorResponse = h.GetClient(modelName)
 		if errorResponse != nil {
 			c.Status(errorResponse.StatusCode)
@@ -400,6 +400,9 @@ func (h *OpenAIAPIHandler) handleNonStreamingResponse(c *gin.Context, rawJSON []
 
 		resp, err := cliClient.SendRawMessage(cliCtx, modelName, rawJSON, "")
 		if err != nil {
+			errorResponse = err
+			h.LoggingAPIResponseError(cliCtx, err)
+
 			switch err.StatusCode {
 			case 429:
 				if h.Cfg.QuotaExceeded.SwitchProject {
@@ -431,6 +434,12 @@ func (h *OpenAIAPIHandler) handleNonStreamingResponse(c *gin.Context, rawJSON []
 			break
 		}
 	}
+	if errorResponse != nil {
+		c.Status(errorResponse.StatusCode)
+		_, _ = c.Writer.Write([]byte(errorResponse.Error.Error()))
+		cliCancel(errorResponse.Error)
+		return
+	}
 }
 
 // handleStreamingResponse handles streaming responses for Gemini models.
@@ -471,10 +480,10 @@ func (h *OpenAIAPIHandler) handleStreamingResponse(c *gin.Context, rawJSON []byt
 		}
 	}()
 
+	var errorResponse *interfaces.ErrorMessage
 	retryCount := 0
 outLoop:
 	for retryCount <= h.Cfg.RequestRetry {
-		var errorResponse *interfaces.ErrorMessage
 		cliClient, errorResponse = h.GetClient(modelName)
 		if errorResponse != nil {
 			c.Status(errorResponse.StatusCode)
@@ -511,6 +520,9 @@ outLoop:
 			// Handle errors from the backend.
 			case err, okError := <-errChan:
 				if okError {
+					errorResponse = err
+					h.LoggingAPIResponseError(cliCtx, err)
+
 					switch err.StatusCode {
 					case 429:
 						if h.Cfg.QuotaExceeded.SwitchProject {
@@ -535,6 +547,13 @@ outLoop:
 			}
 		}
 	}
+	if errorResponse != nil {
+		c.Status(errorResponse.StatusCode)
+		_, _ = fmt.Fprint(c.Writer, errorResponse.Error.Error())
+		flusher.Flush()
+		cliCancel(errorResponse.Error)
+		return
+	}
 }
 
 // handleCompletionsNonStreamingResponse handles non-streaming completions responses.
@@ -562,9 +581,9 @@ func (h *OpenAIAPIHandler) handleCompletionsNonStreamingResponse(c *gin.Context,
 		}
 	}()
 
+	var errorResponse *interfaces.ErrorMessage
 	retryCount := 0
 	for retryCount <= h.Cfg.RequestRetry {
-		var errorResponse *interfaces.ErrorMessage
 		cliClient, errorResponse = h.GetClient(modelName)
 		if errorResponse != nil {
 			c.Status(errorResponse.StatusCode)
@@ -576,6 +595,9 @@ func (h *OpenAIAPIHandler) handleCompletionsNonStreamingResponse(c *gin.Context,
 		// Send the converted chat completions request
 		resp, err := cliClient.SendRawMessage(cliCtx, modelName, chatCompletionsJSON, "")
 		if err != nil {
+			errorResponse = err
+			h.LoggingAPIResponseError(cliCtx, err)
+
 			switch err.StatusCode {
 			case 429:
 				if h.Cfg.QuotaExceeded.SwitchProject {
@@ -601,6 +623,13 @@ func (h *OpenAIAPIHandler) handleCompletionsNonStreamingResponse(c *gin.Context,
 			break
 		}
 	}
+	if errorResponse != nil {
+		c.Status(errorResponse.StatusCode)
+		_, _ = c.Writer.Write([]byte(errorResponse.Error.Error()))
+		cliCancel(errorResponse.Error)
+		return
+	}
+
 }
 
 // handleCompletionsStreamingResponse handles streaming completions responses.
@@ -644,10 +673,10 @@ func (h *OpenAIAPIHandler) handleCompletionsStreamingResponse(c *gin.Context, ra
 		}
 	}()
 
+	var errorResponse *interfaces.ErrorMessage
 	retryCount := 0
 outLoop:
 	for retryCount <= h.Cfg.RequestRetry {
-		var errorResponse *interfaces.ErrorMessage
 		cliClient, errorResponse = h.GetClient(modelName)
 		if errorResponse != nil {
 			c.Status(errorResponse.StatusCode)
@@ -689,6 +718,9 @@ outLoop:
 			// Handle errors from the backend.
 			case err, okError := <-errChan:
 				if okError {
+					errorResponse = err
+					h.LoggingAPIResponseError(cliCtx, err)
+
 					switch err.StatusCode {
 					case 429:
 						if h.Cfg.QuotaExceeded.SwitchProject {
@@ -713,4 +745,11 @@ outLoop:
 			}
 		}
 	}
+	if errorResponse != nil {
+		c.Status(errorResponse.StatusCode)
+		_, _ = fmt.Fprint(c.Writer, errorResponse.Error.Error())
+		flusher.Flush()
+		cliCancel(errorResponse.Error)
+		return
+	}
 }

internal/api/handlers/openai/openai_responses_handlers.go

@@ -115,9 +115,9 @@ func (h *OpenAIResponsesAPIHandler) handleNonStreamingResponse(c *gin.Context, r
 		}
 	}()
 
+	var errorResponse *interfaces.ErrorMessage
 	retryCount := 0
 	for retryCount <= h.Cfg.RequestRetry {
-		var errorResponse *interfaces.ErrorMessage
 		cliClient, errorResponse = h.GetClient(modelName)
 		if errorResponse != nil {
 			c.Status(errorResponse.StatusCode)
@@ -128,6 +128,9 @@ func (h *OpenAIResponsesAPIHandler) handleNonStreamingResponse(c *gin.Context, r
 
 		resp, err := cliClient.SendRawMessage(cliCtx, modelName, rawJSON, "")
 		if err != nil {
+			errorResponse = err
+			h.LoggingAPIResponseError(cliCtx, err)
+
 			switch err.StatusCode {
 			case 429:
 				if h.Cfg.QuotaExceeded.SwitchProject {
@@ -159,6 +162,13 @@ func (h *OpenAIResponsesAPIHandler) handleNonStreamingResponse(c *gin.Context, r
 			break
 		}
 	}
+	if errorResponse != nil {
+		c.Status(errorResponse.StatusCode)
+		_, _ = c.Writer.Write([]byte(errorResponse.Error.Error()))
+		cliCancel(errorResponse.Error)
+		return
+	}
+
 }
 
 // handleStreamingResponse handles streaming responses for Gemini models.
@@ -199,10 +209,10 @@ func (h *OpenAIResponsesAPIHandler) handleStreamingResponse(c *gin.Context, rawJ
 		}
 	}()
 
+	var errorResponse *interfaces.ErrorMessage
 	retryCount := 0
 outLoop:
 	for retryCount <= h.Cfg.RequestRetry {
-		var errorResponse *interfaces.ErrorMessage
 		cliClient, errorResponse = h.GetClient(modelName)
 		if errorResponse != nil {
 			c.Status(errorResponse.StatusCode)
@@ -238,6 +248,8 @@ outLoop:
 			// Handle errors from the backend.
 			case err, okError := <-errChan:
 				if okError {
+					errorResponse = err
+					h.LoggingAPIResponseError(cliCtx, err)
 					switch err.StatusCode {
 					case 429:
 						if h.Cfg.QuotaExceeded.SwitchProject {
@@ -262,4 +274,12 @@ outLoop:
 			}
 		}
 	}
+
+	if errorResponse != nil {
+		c.Status(errorResponse.StatusCode)
+		_, _ = fmt.Fprint(c.Writer, errorResponse.Error.Error())
+		flusher.Flush()
+		cliCancel(errorResponse.Error)
+		return
+	}
 }

internal/api/middleware/response_writer.go

@@ -8,6 +8,7 @@ import (
 	"strings"
 
 	"github.com/gin-gonic/gin"
+	"github.com/luispater/CLIProxyAPI/internal/interfaces"
 	"github.com/luispater/CLIProxyAPI/internal/logging"
 )
 
@@ -240,6 +241,16 @@ func (w *ResponseWriterWrapper) Finalize(c *gin.Context) error {
 			}
 		}
 
+		var slicesAPIResponseError []*interfaces.ErrorMessage
+		apiResponseError, isExist := c.Get("API_RESPONSE_ERROR")
+		if isExist {
+			var ok bool
+			slicesAPIResponseError, ok = apiResponseError.([]*interfaces.ErrorMessage)
+			if !ok {
+				slicesAPIResponseError = nil
+			}
+		}
+
 		// Log complete non-streaming response
 		return w.logger.LogRequest(
 			w.requestInfo.URL,
@@ -251,6 +262,7 @@ func (w *ResponseWriterWrapper) Finalize(c *gin.Context) error {
 			w.body.Bytes(),
 			apiRequestBody,
 			apiResponseBody,
+			slicesAPIResponseError,
 		)
 	}
 

internal/api/server.go

@@ -18,6 +18,7 @@ import (
 	managementHandlers "github.com/luispater/CLIProxyAPI/internal/api/handlers/management"
 	"github.com/luispater/CLIProxyAPI/internal/api/handlers/openai"
 	"github.com/luispater/CLIProxyAPI/internal/api/middleware"
+	"github.com/luispater/CLIProxyAPI/internal/client"
 	"github.com/luispater/CLIProxyAPI/internal/config"
 	"github.com/luispater/CLIProxyAPI/internal/interfaces"
 	"github.com/luispater/CLIProxyAPI/internal/logging"
@@ -315,7 +316,47 @@ func (s *Server) UpdateClients(clients map[string]interfaces.Client, cfg *config
 	if s.mgmt != nil {
 		s.mgmt.SetConfig(cfg)
 	}
-	log.Infof("server clients and configuration updated: %d clients", len(clientSlice))
+
+	// Count client types for detailed logging
+	authFiles := 0
+	glAPIKeyCount := 0
+	claudeAPIKeyCount := 0
+	codexAPIKeyCount := 0
+	openAICompatCount := 0
+
+	for _, c := range clientSlice {
+		switch cl := c.(type) {
+		case *client.GeminiCLIClient:
+			authFiles++
+		case *client.CodexClient:
+			if cl.GetAPIKey() == "" {
+				authFiles++
+			} else {
+				codexAPIKeyCount++
+			}
+		case *client.ClaudeClient:
+			if cl.GetAPIKey() == "" {
+				authFiles++
+			} else {
+				claudeAPIKeyCount++
+			}
+		case *client.QwenClient:
+			authFiles++
+		case *client.GeminiClient:
+			glAPIKeyCount++
+		case *client.OpenAICompatibilityClient:
+			openAICompatCount++
+		}
+	}
+
+	log.Infof("server clients and configuration updated: %d clients (%d auth files + %d GL API keys + %d Claude API keys + %d Codex keys + %d OpenAI-compat)",
+		len(clientSlice),
+		authFiles,
+		glAPIKeyCount,
+		claudeAPIKeyCount,
+		codexAPIKeyCount,
+		openAICompatCount,
+	)
 }
 
 // (management handlers moved to internal/api/handlers/management)

internal/auth/claude/token.go

@@ -7,7 +7,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"os"
-	"path"
+	"path/filepath"
 )
 
 // ClaudeTokenStorage stores OAuth2 token information for Anthropic Claude API authentication.
@@ -49,7 +49,7 @@ func (ts *ClaudeTokenStorage) SaveTokenToFile(authFilePath string) error {
 	ts.Type = "claude"
 
 	// Create directory structure if it doesn't exist
-	if err := os.MkdirAll(path.Dir(authFilePath), 0700); err != nil {
+	if err := os.MkdirAll(filepath.Dir(authFilePath), 0700); err != nil {
 		return fmt.Errorf("failed to create directory: %v", err)
 	}
 

internal/auth/codex/token.go

@@ -7,7 +7,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"os"
-	"path"
+	"path/filepath"
 )
 
 // CodexTokenStorage stores OAuth2 token information for OpenAI Codex API authentication.
@@ -43,7 +43,7 @@ type CodexTokenStorage struct {
 //   - error: An error if the operation fails, nil otherwise
 func (ts *CodexTokenStorage) SaveTokenToFile(authFilePath string) error {
 	ts.Type = "codex"
-	if err := os.MkdirAll(path.Dir(authFilePath), 0700); err != nil {
+	if err := os.MkdirAll(filepath.Dir(authFilePath), 0700); err != nil {
 		return fmt.Errorf("failed to create directory: %v", err)
 	}
 

internal/auth/gemini/gemini_token.go

@@ -7,7 +7,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"os"
-	"path"
+	"path/filepath"
 
 	log "github.com/sirupsen/logrus"
 )
@@ -46,7 +46,7 @@ type GeminiTokenStorage struct {
 //   - error: An error if the operation fails, nil otherwise
 func (ts *GeminiTokenStorage) SaveTokenToFile(authFilePath string) error {
 	ts.Type = "gemini"
-	if err := os.MkdirAll(path.Dir(authFilePath), 0700); err != nil {
+	if err := os.MkdirAll(filepath.Dir(authFilePath), 0700); err != nil {
 		return fmt.Errorf("failed to create directory: %v", err)
 	}
 

internal/auth/qwen/qwen_token.go

@@ -7,7 +7,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"os"
-	"path"
+	"path/filepath"
 )
 
 // QwenTokenStorage stores OAuth2 token information for Alibaba Qwen API authentication.
@@ -41,7 +41,7 @@ type QwenTokenStorage struct {
 //   - error: An error if the operation fails, nil otherwise
 func (ts *QwenTokenStorage) SaveTokenToFile(authFilePath string) error {
 	ts.Type = "qwen"
-	if err := os.MkdirAll(path.Dir(authFilePath), 0700); err != nil {
+	if err := os.MkdirAll(filepath.Dir(authFilePath), 0700); err != nil {
 		return fmt.Errorf("failed to create directory: %v", err)
 	}
 

internal/client/gemini-cli_client.go

@@ -38,8 +38,9 @@ const (
 
 var (
 	previewModels = map[string][]string{
-		"gemini-2.5-pro":   {"gemini-2.5-pro-preview-05-06", "gemini-2.5-pro-preview-06-05"},
-		"gemini-2.5-flash": {"gemini-2.5-flash-preview-04-17", "gemini-2.5-flash-preview-05-20"},
+		"gemini-2.5-pro":        {"gemini-2.5-pro-preview-05-06", "gemini-2.5-pro-preview-06-05"},
+		"gemini-2.5-flash":      {"gemini-2.5-flash-preview-04-17", "gemini-2.5-flash-preview-05-20"},
+		"gemini-2.5-flash-lite": {"gemini-2.5-flash-lite-preview-06-17"},
 	}
 )
 
@@ -99,6 +100,7 @@ func (c *GeminiCLIClient) CanProvideModel(modelName string) bool {
 	models := []string{
 		"gemini-2.5-pro",
 		"gemini-2.5-flash",
+		"gemini-2.5-flash-lite",
 	}
 	return util.InArray(models, modelName)
 }

internal/cmd/run.go

@@ -50,6 +50,7 @@ import (
 func StartService(cfg *config.Config, configPath string) {
 	// Create a pool of API clients, one for each token file found.
 	cliClients := make(map[string]interfaces.Client)
+	successfulAuthCount := 0
 	err := filepath.Walk(cfg.AuthDir, func(path string, info fs.FileInfo, err error) error {
 		if err != nil {
 			return err
@@ -89,6 +90,7 @@ func StartService(cfg *config.Config, configPath string) {
 					// Add the new client to the pool.
 					cliClient := client.NewGeminiCLIClient(httpClient, &ts, cfg)
 					cliClients[path] = cliClient
+					successfulAuthCount++
 				}
 			} else if tokenType == "codex" {
 				var ts codex.CodexTokenStorage
@@ -103,6 +105,7 @@ func StartService(cfg *config.Config, configPath string) {
 					}
 					log.Info("Authentication successful.")
 					cliClients[path] = codexClient
+					successfulAuthCount++
 				}
 			} else if tokenType == "claude" {
 				var ts claude.ClaudeTokenStorage
@@ -112,6 +115,7 @@ func StartService(cfg *config.Config, configPath string) {
 					claudeClient := client.NewClaudeClient(cfg, &ts)
 					log.Info("Authentication successful.")
 					cliClients[path] = claudeClient
+					successfulAuthCount++
 				}
 			} else if tokenType == "qwen" {
 				var ts qwen.QwenTokenStorage
@@ -121,6 +125,7 @@ func StartService(cfg *config.Config, configPath string) {
 					qwenClient := client.NewQwenClient(cfg, &ts)
 					log.Info("Authentication successful.")
 					cliClients[path] = qwenClient
+					successfulAuthCount++
 				}
 			}
 		}
@@ -130,51 +135,24 @@ func StartService(cfg *config.Config, configPath string) {
 		log.Fatalf("Error walking auth directory: %v", err)
 	}
 
-	clientSlice := clientsToSlice(cliClients)
+	apiKeyClients, glAPIKeyCount, claudeAPIKeyCount, codexAPIKeyCount, openAICompatCount := buildAPIKeyClients(cfg)
 
-	if len(cfg.GlAPIKey) > 0 {
-		// Initialize clients with Generative Language API Keys if provided in configuration.
-		for i := 0; i < len(cfg.GlAPIKey); i++ {
-			httpClient := util.SetProxy(cfg, &http.Client{})
+	totalNewClients := len(cliClients) + len(apiKeyClients)
+	log.Infof("full client load complete - %d clients (%d auth files + %d GL API keys + %d Claude API keys + %d Codex keys + %d OpenAI-compat)",
+		totalNewClients,
+		successfulAuthCount,
+		glAPIKeyCount,
+		claudeAPIKeyCount,
+		codexAPIKeyCount,
+		openAICompatCount,
+	)
 
-			log.Debug("Initializing with Generative Language API Key...")
-			cliClient := client.NewGeminiClient(httpClient, cfg, cfg.GlAPIKey[i])
-			clientSlice = append(clientSlice, cliClient)
-		}
-	}
-
-	if len(cfg.ClaudeKey) > 0 {
-		// Initialize clients with Claude API Keys if provided in configuration.
-		for i := 0; i < len(cfg.ClaudeKey); i++ {
-			log.Debug("Initializing with Claude API Key...")
-			cliClient := client.NewClaudeClientWithKey(cfg, i)
-			clientSlice = append(clientSlice, cliClient)
-		}
-	}
-
-	if len(cfg.CodexKey) > 0 {
-		// Initialize clients with Codex API Keys if provided in configuration.
-		for i := 0; i < len(cfg.CodexKey); i++ {
-			log.Debug("Initializing with Codex API Key...")
-			cliClient := client.NewCodexClientWithKey(cfg, i)
-			clientSlice = append(clientSlice, cliClient)
-		}
-	}
-
-	if len(cfg.OpenAICompatibility) > 0 {
-		// Initialize clients for OpenAI compatibility configurations
-		for _, compatConfig := range cfg.OpenAICompatibility {
-			log.Debugf("Initializing OpenAI compatibility client for provider: %s", compatConfig.Name)
-			compatClient, errClient := client.NewOpenAICompatibilityClient(cfg, &compatConfig)
-			if errClient != nil {
-				log.Fatalf("failed to create OpenAI compatibility client for %s: %v", compatConfig.Name, errClient)
-			}
-			clientSlice = append(clientSlice, compatClient)
-		}
-	}
+	// Combine file-based and API key-based clients for the initial server setup
+	allClients := clientsToSlice(cliClients)
+	allClients = append(allClients, clientsToSlice(apiKeyClients)...)
 
 	// Create and start the API server with the pool of clients in a separate goroutine.
-	apiServer := api.NewServer(cfg, clientSlice, configPath)
+	apiServer := api.NewServer(cfg, allClients, configPath)
 	log.Infof("Starting API server on port %d", cfg.Port)
 
 	// Start the API server in a goroutine so it doesn't block the main thread.
@@ -200,6 +178,7 @@ func StartService(cfg *config.Config, configPath string) {
 	// Set initial state for the watcher with current configuration and clients.
 	fileWatcher.SetConfig(cfg)
 	fileWatcher.SetClients(cliClients)
+	fileWatcher.SetAPIKeyClients(apiKeyClients)
 
 	// Start the file watcher in a separate context.
 	watcherCtx, watcherCancel := context.WithCancel(context.Background())
@@ -317,3 +296,55 @@ func clientsToSlice(clientMap map[string]interfaces.Client) []interfaces.Client
 	}
 	return s
 }
+
+// buildAPIKeyClients creates clients from API keys in the config
+func buildAPIKeyClients(cfg *config.Config) (map[string]interfaces.Client, int, int, int, int) {
+	apiKeyClients := make(map[string]interfaces.Client)
+	glAPIKeyCount := 0
+	claudeAPIKeyCount := 0
+	codexAPIKeyCount := 0
+	openAICompatCount := 0
+
+	if len(cfg.GlAPIKey) > 0 {
+		for _, key := range cfg.GlAPIKey {
+			httpClient := util.SetProxy(cfg, &http.Client{})
+			log.Debug("Initializing with Generative Language API Key...")
+			cliClient := client.NewGeminiClient(httpClient, cfg, key)
+			apiKeyClients[cliClient.GetClientID()] = cliClient
+			glAPIKeyCount++
+		}
+	}
+
+	if len(cfg.ClaudeKey) > 0 {
+		for i := range cfg.ClaudeKey {
+			log.Debug("Initializing with Claude API Key...")
+			cliClient := client.NewClaudeClientWithKey(cfg, i)
+			apiKeyClients[cliClient.GetClientID()] = cliClient
+			claudeAPIKeyCount++
+		}
+	}
+
+	if len(cfg.CodexKey) > 0 {
+		for i := range cfg.CodexKey {
+			log.Debug("Initializing with Codex API Key...")
+			cliClient := client.NewCodexClientWithKey(cfg, i)
+			apiKeyClients[cliClient.GetClientID()] = cliClient
+			codexAPIKeyCount++
+		}
+	}
+
+	if len(cfg.OpenAICompatibility) > 0 {
+		for _, compatConfig := range cfg.OpenAICompatibility {
+			log.Debugf("Initializing OpenAI compatibility client for provider: %s", compatConfig.Name)
+			compatClient, errClient := client.NewOpenAICompatibilityClient(cfg, &compatConfig)
+			if errClient != nil {
+				log.Errorf("failed to create OpenAI compatibility client for %s: %v", compatConfig.Name, errClient)
+				continue
+			}
+			apiKeyClients[compatClient.GetClientID()] = compatClient
+			openAICompatCount++
+		}
+	}
+
+	return apiKeyClients, glAPIKeyCount, claudeAPIKeyCount, codexAPIKeyCount, openAICompatCount
+}

internal/logging/request_logger.go

@@ -14,6 +14,8 @@ import (
 	"regexp"
 	"strings"
 	"time"
+
+	"github.com/luispater/CLIProxyAPI/internal/interfaces"
 )
 
 // RequestLogger defines the interface for logging HTTP requests and responses.
@@ -34,7 +36,7 @@ type RequestLogger interface {
 	//
 	// 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) error
+	LogRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte, apiResponseErrors []*interfaces.ErrorMessage) error
 
 	// LogStreamingRequest initiates logging for a streaming request and returns a writer for chunks.
 	//
@@ -139,7 +141,7 @@ func (l *FileRequestLogger) SetEnabled(enabled bool) {
 //
 // Returns:
 //   - error: An error if logging fails, nil otherwise
-func (l *FileRequestLogger) LogRequest(url, method string, requestHeaders map[string][]string, body []byte, statusCode int, responseHeaders map[string][]string, response, apiRequest, apiResponse []byte) 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) error {
 	if !l.enabled {
 		return nil
 	}
@@ -161,7 +163,7 @@ func (l *FileRequestLogger) LogRequest(url, method string, requestHeaders map[st
 	}
 
 	// Create log content
-	content := l.formatLogContent(url, method, requestHeaders, body, apiRequest, apiResponse, decompressedResponse, statusCode, responseHeaders)
+	content := l.formatLogContent(url, method, requestHeaders, body, apiRequest, apiResponse, decompressedResponse, statusCode, responseHeaders, apiResponseErrors)
 
 	// Write to file
 	if err = os.WriteFile(filePath, []byte(content), 0644); err != nil {
@@ -310,7 +312,7 @@ func (l *FileRequestLogger) sanitizeForFilename(path string) string {
 //
 // Returns:
 //   - string: The formatted log content
-func (l *FileRequestLogger) formatLogContent(url, method string, headers map[string][]string, body, apiRequest, apiResponse, response []byte, status int, responseHeaders map[string][]string) string {
+func (l *FileRequestLogger) formatLogContent(url, method string, headers map[string][]string, body, apiRequest, apiResponse, response []byte, status int, responseHeaders map[string][]string, apiResponseErrors []*interfaces.ErrorMessage) string {
 	var content strings.Builder
 
 	// Request info
@@ -320,6 +322,13 @@ func (l *FileRequestLogger) formatLogContent(url, method string, headers map[str
 	content.Write(apiRequest)
 	content.WriteString("\n\n")
 
+	for i := 0; i < len(apiResponseErrors); i++ {
+		content.WriteString("=== API ERROR RESPONSE ===\n")
+		content.WriteString(fmt.Sprintf("HTTP Status: %d\n", apiResponseErrors[i].StatusCode))
+		content.WriteString(apiResponseErrors[i].Error.Error())
+		content.WriteString("\n\n")
+	}
+
 	content.WriteString("=== API RESPONSE ===\n")
 	content.Write(apiResponse)
 	content.WriteString("\n\n")

internal/registry/model_definitions.go

@@ -130,6 +130,20 @@ func GetGeminiCLIModels() []*ModelInfo {
 			OutputTokenLimit:           65536,
 			SupportedGenerationMethods: []string{"generateContent", "countTokens", "createCachedContent", "batchGenerateContent"},
 		},
+		{
+			ID:                         "gemini-2.5-flash-lite",
+			Object:                     "model",
+			Created:                    time.Now().Unix(),
+			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"},
+		},
 	}
 }
 

internal/translator/claude/gemini/claude_gemini_request.go

@@ -89,6 +89,17 @@ func ConvertGeminiRequestToClaude(modelName string, inputRawJSON []byte, stream
 				out, _ = sjson.Set(out, "stop_sequences", stopSequences)
 			}
 		}
+		// Include thoughts configuration for reasoning process visibility
+		if thinkingConfig := genConfig.Get("thinkingConfig"); thinkingConfig.Exists() && thinkingConfig.IsObject() {
+			if includeThoughts := thinkingConfig.Get("include_thoughts"); includeThoughts.Exists() {
+				if includeThoughts.Type == gjson.True {
+					out, _ = sjson.Set(out, "thinking.type", "enabled")
+					if thinkingBudget := thinkingConfig.Get("thinkingBudget"); thinkingBudget.Exists() {
+						out, _ = sjson.Set(out, "thinking.budget_tokens", thinkingBudget.Int())
+					}
+				}
+			}
+		}
 	}
 
 	// System instruction conversion to Claude Code format

internal/translator/claude/gemini/claude_gemini_response.go

@@ -128,7 +128,7 @@ func ConvertClaudeResponseToGemini(_ context.Context, modelName string, original
 				}
 			case "thinking_delta":
 				// Thinking/reasoning content delta for models with reasoning capabilities
-				if text := delta.Get("text"); text.Exists() && text.String() != "" {
+				if text := delta.Get("thinking"); text.Exists() && text.String() != "" {
 					thinkingPart := `{"thought":true,"text":""}`
 					thinkingPart, _ = sjson.Set(thinkingPart, "text", text.String())
 					template, _ = sjson.SetRaw(template, "candidates.0.content.parts.-1", thinkingPart)
@@ -411,7 +411,7 @@ func ConvertClaudeResponseToGeminiNonStream(_ context.Context, modelName string,
 					}
 				case "thinking_delta":
 					// Process reasoning/thinking content
-					if text := delta.Get("text"); text.Exists() && text.String() != "" {
+					if text := delta.Get("thinking"); text.Exists() && text.String() != "" {
 						partJSON := `{"thought":true,"text":""}`
 						partJSON, _ = sjson.Set(partJSON, "text", text.String())
 						part := gjson.Parse(partJSON).Value().(map[string]interface{})

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

@@ -41,6 +41,21 @@ func ConvertOpenAIRequestToClaude(modelName string, inputRawJSON []byte, stream
 
 	root := gjson.ParseBytes(rawJSON)
 
+	if v := root.Get("reasoning_effort"); v.Exists() {
+		out, _ = sjson.Set(out, "thinking.type", "enabled")
+
+		switch v.String() {
+		case "none":
+			out, _ = sjson.Set(out, "thinking.type", "disabled")
+		case "low":
+			out, _ = sjson.Set(out, "thinking.budget_tokens", 1024)
+		case "medium":
+			out, _ = sjson.Set(out, "thinking.budget_tokens", 8192)
+		case "high":
+			out, _ = sjson.Set(out, "thinking.budget_tokens", 24576)
+		}
+	}
+
 	// Helper for generating tool call IDs in the form: toolu_<alphanum>
 	// This ensures unique identifiers for tool calls in the Claude Code format
 	genToolCallID := func() string {

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

@@ -128,10 +128,11 @@ func ConvertClaudeResponseToOpenAI(_ context.Context, modelName string, original
 				return []string{}
 			}
 		}
-		return []string{template}
+		return []string{}
 
 	case "content_block_delta":
 		// Handle content delta (text, tool use arguments, or reasoning content)
+		hasContent := false
 		if delta := root.Get("delta"); delta.Exists() {
 			deltaType := delta.Get("type").String()
 
@@ -140,8 +141,14 @@ func ConvertClaudeResponseToOpenAI(_ context.Context, modelName string, original
 				// Text content delta - send incremental text updates
 				if text := delta.Get("text"); text.Exists() {
 					template, _ = sjson.Set(template, "choices.0.delta.content", text.String())
+					hasContent = true
+				}
+			case "thinking_delta":
+				// Accumulate reasoning/thinking content
+				if thinking := delta.Get("thinking"); thinking.Exists() {
+					template, _ = sjson.Set(template, "choices.0.delta.reasoning_content", thinking.String())
+					hasContent = true
 				}
-
 			case "input_json_delta":
 				// Tool use input delta - accumulate arguments for tool calls
 				if partialJSON := delta.Get("partial_json"); partialJSON.Exists() {
@@ -156,7 +163,11 @@ func ConvertClaudeResponseToOpenAI(_ context.Context, modelName string, original
 				return []string{}
 			}
 		}
-		return []string{template}
+		if hasContent {
+			return []string{template}
+		} else {
+			return []string{}
+		}
 
 	case "content_block_stop":
 		// End of content block - output complete tool call if it's a tool_use block

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

@@ -28,6 +28,23 @@ func ConvertOpenAIResponsesRequestToClaude(modelName string, inputRawJSON []byte
 
 	root := gjson.ParseBytes(rawJSON)
 
+	if v := root.Get("reasoning.effort"); v.Exists() {
+		out, _ = sjson.Set(out, "thinking.type", "enabled")
+
+		switch v.String() {
+		case "none":
+			out, _ = sjson.Set(out, "thinking.type", "disabled")
+		case "minimal":
+			out, _ = sjson.Set(out, "thinking.budget_tokens", 1024)
+		case "low":
+			out, _ = sjson.Set(out, "thinking.budget_tokens", 4096)
+		case "medium":
+			out, _ = sjson.Set(out, "thinking.budget_tokens", 8192)
+		case "high":
+			out, _ = sjson.Set(out, "thinking.budget_tokens", 24576)
+		}
+	}
+
 	// Helper for generating tool call IDs when missing
 	genToolCallID := func() string {
 		const letters = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789"

internal/translator/codex/claude/codex_claude_request.go

@@ -8,6 +8,8 @@ package claude
 import (
 	"bytes"
 	"fmt"
+	"strconv"
+	"strings"
 
 	"github.com/luispater/CLIProxyAPI/internal/misc"
 	"github.com/tidwall/gjson"
@@ -94,7 +96,17 @@ func ConvertClaudeRequestToCodex(modelName string, inputRawJSON []byte, _ bool)
 						// Handle tool use content by creating function call message.
 						functionCallMessage := `{"type":"function_call"}`
 						functionCallMessage, _ = sjson.Set(functionCallMessage, "call_id", messageContentResult.Get("id").String())
-						functionCallMessage, _ = sjson.Set(functionCallMessage, "name", messageContentResult.Get("name").String())
+						{
+							// Shorten tool name if needed based on declared tools
+							name := messageContentResult.Get("name").String()
+							toolMap := buildReverseMapFromClaudeOriginalToShort(rawJSON)
+							if short, ok := toolMap[name]; ok {
+								name = short
+							} else {
+								name = shortenNameIfNeeded(name)
+							}
+							functionCallMessage, _ = sjson.Set(functionCallMessage, "name", name)
+						}
 						functionCallMessage, _ = sjson.Set(functionCallMessage, "arguments", messageContentResult.Get("input").Raw)
 						template, _ = sjson.SetRaw(template, "input.-1", functionCallMessage)
 					} else if contentType == "tool_result" {
@@ -130,10 +142,29 @@ func ConvertClaudeRequestToCodex(modelName string, inputRawJSON []byte, _ bool)
 		template, _ = sjson.SetRaw(template, "tools", `[]`)
 		template, _ = sjson.Set(template, "tool_choice", `auto`)
 		toolResults := toolsResult.Array()
+		// Build short name map from declared tools
+		var names []string
+		for i := 0; i < len(toolResults); i++ {
+			n := toolResults[i].Get("name").String()
+			if n != "" {
+				names = append(names, n)
+			}
+		}
+		shortMap := buildShortNameMap(names)
 		for i := 0; i < len(toolResults); i++ {
 			toolResult := toolResults[i]
 			tool := toolResult.Raw
 			tool, _ = sjson.Set(tool, "type", "function")
+			// Apply shortened name if needed
+			if v := toolResult.Get("name"); v.Exists() {
+				name := v.String()
+				if short, ok := shortMap[name]; ok {
+					name = short
+				} else {
+					name = shortenNameIfNeeded(name)
+				}
+				tool, _ = sjson.Set(tool, "name", name)
+			}
 			tool, _ = sjson.SetRaw(tool, "parameters", toolResult.Get("input_schema").Raw)
 			tool, _ = sjson.Delete(tool, "input_schema")
 			tool, _ = sjson.Delete(tool, "parameters.$schema")
@@ -170,3 +201,97 @@ func ConvertClaudeRequestToCodex(modelName string, inputRawJSON []byte, _ bool)
 
 	return []byte(template)
 }
+
+// shortenNameIfNeeded applies a simple shortening rule for a single name.
+func shortenNameIfNeeded(name string) string {
+	const limit = 64
+	if len(name) <= limit {
+		return name
+	}
+	if strings.HasPrefix(name, "mcp__") {
+		idx := strings.LastIndex(name, "__")
+		if idx > 0 {
+			cand := "mcp__" + name[idx+2:]
+			if len(cand) > limit {
+				return cand[:limit]
+			}
+			return cand
+		}
+	}
+	return name[:limit]
+}
+
+// buildShortNameMap ensures uniqueness of shortened names within a request.
+func buildShortNameMap(names []string) map[string]string {
+	const limit = 64
+	used := map[string]struct{}{}
+	m := map[string]string{}
+
+	baseCandidate := func(n string) string {
+		if len(n) <= limit {
+			return n
+		}
+		if strings.HasPrefix(n, "mcp__") {
+			idx := strings.LastIndex(n, "__")
+			if idx > 0 {
+				cand := "mcp__" + n[idx+2:]
+				if len(cand) > limit {
+					cand = cand[:limit]
+				}
+				return cand
+			}
+		}
+		return n[:limit]
+	}
+
+	makeUnique := func(cand string) string {
+		if _, ok := used[cand]; !ok {
+			return cand
+		}
+		base := cand
+		for i := 1; ; i++ {
+			suffix := "~" + strconv.Itoa(i)
+			allowed := limit - len(suffix)
+			if allowed < 0 {
+				allowed = 0
+			}
+			tmp := base
+			if len(tmp) > allowed {
+				tmp = tmp[:allowed]
+			}
+			tmp = tmp + suffix
+			if _, ok := used[tmp]; !ok {
+				return tmp
+			}
+		}
+	}
+
+	for _, n := range names {
+		cand := baseCandidate(n)
+		uniq := makeUnique(cand)
+		used[uniq] = struct{}{}
+		m[n] = uniq
+	}
+	return m
+}
+
+// buildReverseMapFromClaudeOriginalToShort builds original->short map, used to map tool_use names to short.
+func buildReverseMapFromClaudeOriginalToShort(original []byte) map[string]string {
+	tools := gjson.GetBytes(original, "tools")
+	m := map[string]string{}
+	if !tools.IsArray() {
+		return m
+	}
+	var names []string
+	arr := tools.Array()
+	for i := 0; i < len(arr); i++ {
+		n := arr[i].Get("name").String()
+		if n != "" {
+			names = append(names, n)
+		}
+	}
+	if len(names) > 0 {
+		m = buildShortNameMap(names)
+	}
+	return m
+}

internal/translator/codex/claude/codex_claude_response.go

@@ -122,7 +122,15 @@ func ConvertCodexResponseToClaude(_ context.Context, _ string, originalRequestRa
 			template = `{"type":"content_block_start","index":0,"content_block":{"type":"tool_use","id":"","name":"","input":{}}}`
 			template, _ = sjson.Set(template, "index", rootResult.Get("output_index").Int())
 			template, _ = sjson.Set(template, "content_block.id", itemResult.Get("call_id").String())
-			template, _ = sjson.Set(template, "content_block.name", itemResult.Get("name").String())
+			{
+				// Restore original tool name if shortened
+				name := itemResult.Get("name").String()
+				rev := buildReverseMapFromClaudeOriginalShortToOriginal(originalRequestRawJSON)
+				if orig, ok := rev[name]; ok {
+					name = orig
+				}
+				template, _ = sjson.Set(template, "content_block.name", name)
+			}
 
 			output = "event: content_block_start\n"
 			output += fmt.Sprintf("data: %s\n\n", template)
@@ -171,3 +179,27 @@ func ConvertCodexResponseToClaude(_ context.Context, _ string, originalRequestRa
 func ConvertCodexResponseToClaudeNonStream(_ context.Context, _ string, originalRequestRawJSON, requestRawJSON, _ []byte, _ *any) string {
 	return ""
 }
+
+// buildReverseMapFromClaudeOriginalShortToOriginal builds a map[short]original from original Claude request tools.
+func buildReverseMapFromClaudeOriginalShortToOriginal(original []byte) map[string]string {
+	tools := gjson.GetBytes(original, "tools")
+	rev := map[string]string{}
+	if !tools.IsArray() {
+		return rev
+	}
+	var names []string
+	arr := tools.Array()
+	for i := 0; i < len(arr); i++ {
+		n := arr[i].Get("name").String()
+		if n != "" {
+			names = append(names, n)
+		}
+	}
+	if len(names) > 0 {
+		m := buildShortNameMap(names)
+		for orig, short := range m {
+			rev[short] = orig
+		}
+	}
+	return rev
+}

internal/translator/codex/gemini/codex_gemini_request.go

@@ -10,6 +10,7 @@ import (
 	"crypto/rand"
 	"fmt"
 	"math/big"
+	"strconv"
 	"strings"
 
 	"github.com/luispater/CLIProxyAPI/internal/misc"
@@ -46,6 +47,27 @@ func ConvertGeminiRequestToCodex(modelName string, inputRawJSON []byte, _ bool)
 
 	root := gjson.ParseBytes(rawJSON)
 
+	// Pre-compute tool name shortening map from declared functionDeclarations
+	shortMap := map[string]string{}
+	if tools := root.Get("tools"); tools.IsArray() {
+		var names []string
+		tarr := tools.Array()
+		for i := 0; i < len(tarr); i++ {
+			fns := tarr[i].Get("functionDeclarations")
+			if !fns.IsArray() {
+				continue
+			}
+			for _, fn := range fns.Array() {
+				if v := fn.Get("name"); v.Exists() {
+					names = append(names, v.String())
+				}
+			}
+		}
+		if len(names) > 0 {
+			shortMap = buildShortNameMap(names)
+		}
+	}
+
 	// helper for generating paired call IDs in the form: call_<alphanum>
 	// Gemini uses sequential pairing across possibly multiple in-flight
 	// functionCalls, so we keep a FIFO queue of generated call IDs and
@@ -124,7 +146,13 @@ func ConvertGeminiRequestToCodex(modelName string, inputRawJSON []byte, _ bool)
 				if fc := p.Get("functionCall"); fc.Exists() {
 					fn := `{"type":"function_call"}`
 					if name := fc.Get("name"); name.Exists() {
-						fn, _ = sjson.Set(fn, "name", name.String())
+						n := name.String()
+						if short, ok := shortMap[n]; ok {
+							n = short
+						} else {
+							n = shortenNameIfNeeded(n)
+						}
+						fn, _ = sjson.Set(fn, "name", n)
 					}
 					if args := fc.Get("args"); args.Exists() {
 						fn, _ = sjson.Set(fn, "arguments", args.Raw)
@@ -185,7 +213,13 @@ func ConvertGeminiRequestToCodex(modelName string, inputRawJSON []byte, _ bool)
 				tool := `{}`
 				tool, _ = sjson.Set(tool, "type", "function")
 				if v := fn.Get("name"); v.Exists() {
-					tool, _ = sjson.Set(tool, "name", v.String())
+					name := v.String()
+					if short, ok := shortMap[name]; ok {
+						name = short
+					} else {
+						name = shortenNameIfNeeded(name)
+					}
+					tool, _ = sjson.Set(tool, "name", name)
 				}
 				if v := fn.Get("description"); v.Exists() {
 					tool, _ = sjson.Set(tool, "description", v.String())
@@ -227,3 +261,76 @@ func ConvertGeminiRequestToCodex(modelName string, inputRawJSON []byte, _ bool)
 
 	return []byte(out)
 }
+
+// shortenNameIfNeeded applies the simple shortening rule for a single name.
+func shortenNameIfNeeded(name string) string {
+	const limit = 64
+	if len(name) <= limit {
+		return name
+	}
+	if strings.HasPrefix(name, "mcp__") {
+		idx := strings.LastIndex(name, "__")
+		if idx > 0 {
+			cand := "mcp__" + name[idx+2:]
+			if len(cand) > limit {
+				return cand[:limit]
+			}
+			return cand
+		}
+	}
+	return name[:limit]
+}
+
+// buildShortNameMap ensures uniqueness of shortened names within a request.
+func buildShortNameMap(names []string) map[string]string {
+	const limit = 64
+	used := map[string]struct{}{}
+	m := map[string]string{}
+
+	baseCandidate := func(n string) string {
+		if len(n) <= limit {
+			return n
+		}
+		if strings.HasPrefix(n, "mcp__") {
+			idx := strings.LastIndex(n, "__")
+			if idx > 0 {
+				cand := "mcp__" + n[idx+2:]
+				if len(cand) > limit {
+					cand = cand[:limit]
+				}
+				return cand
+			}
+		}
+		return n[:limit]
+	}
+
+	makeUnique := func(cand string) string {
+		if _, ok := used[cand]; !ok {
+			return cand
+		}
+		base := cand
+		for i := 1; ; i++ {
+			suffix := "~" + strconv.Itoa(i)
+			allowed := limit - len(suffix)
+			if allowed < 0 {
+				allowed = 0
+			}
+			tmp := base
+			if len(tmp) > allowed {
+				tmp = tmp[:allowed]
+			}
+			tmp = tmp + suffix
+			if _, ok := used[tmp]; !ok {
+				return tmp
+			}
+		}
+	}
+
+	for _, n := range names {
+		cand := baseCandidate(n)
+		uniq := makeUnique(cand)
+		used[uniq] = struct{}{}
+		m[n] = uniq
+	}
+	return m
+}

internal/translator/codex/gemini/codex_gemini_response.go

@@ -80,7 +80,15 @@ func ConvertCodexResponseToGemini(_ context.Context, modelName string, originalR
 		if itemType == "function_call" {
 			// Create function call part
 			functionCall := `{"functionCall":{"name":"","args":{}}}`
-			functionCall, _ = sjson.Set(functionCall, "functionCall.name", itemResult.Get("name").String())
+			{
+				// Restore original tool name if shortened
+				n := itemResult.Get("name").String()
+				rev := buildReverseMapFromGeminiOriginal(originalRequestRawJSON)
+				if orig, ok := rev[n]; ok {
+					n = orig
+				}
+				functionCall, _ = sjson.Set(functionCall, "functionCall.name", n)
+			}
 
 			// Parse and set arguments
 			argsStr := itemResult.Get("arguments").String()
@@ -250,7 +258,14 @@ func ConvertCodexResponseToGeminiNonStream(_ context.Context, modelName string,
 						hasToolCall = true
 						functionCall := map[string]interface{}{
 							"functionCall": map[string]interface{}{
-								"name": value.Get("name").String(),
+								"name": func() string {
+									n := value.Get("name").String()
+									rev := buildReverseMapFromGeminiOriginal(originalRequestRawJSON)
+									if orig, ok := rev[n]; ok {
+										return orig
+									}
+									return n
+								}(),
 								"args": map[string]interface{}{},
 							},
 						}
@@ -292,6 +307,35 @@ func ConvertCodexResponseToGeminiNonStream(_ context.Context, modelName string,
 	return ""
 }
 
+// buildReverseMapFromGeminiOriginal builds a map[short]original from original Gemini request tools.
+func buildReverseMapFromGeminiOriginal(original []byte) map[string]string {
+	tools := gjson.GetBytes(original, "tools")
+	rev := map[string]string{}
+	if !tools.IsArray() {
+		return rev
+	}
+	var names []string
+	tarr := tools.Array()
+	for i := 0; i < len(tarr); i++ {
+		fns := tarr[i].Get("functionDeclarations")
+		if !fns.IsArray() {
+			continue
+		}
+		for _, fn := range fns.Array() {
+			if v := fn.Get("name"); v.Exists() {
+				names = append(names, v.String())
+			}
+		}
+	}
+	if len(names) > 0 {
+		m := buildShortNameMap(names)
+		for orig, short := range m {
+			rev[short] = orig
+		}
+	}
+	return rev
+}
+
 // mustMarshalJSON marshals a value to JSON, panicking on error.
 func mustMarshalJSON(v interface{}) string {
 	data, err := json.Marshal(v)

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

@@ -9,6 +9,9 @@ package chat_completions
 import (
 	"bytes"
 
+	"strconv"
+	"strings"
+
 	"github.com/luispater/CLIProxyAPI/internal/misc"
 	"github.com/tidwall/gjson"
 	"github.com/tidwall/sjson"
@@ -67,6 +70,31 @@ func ConvertOpenAIRequestToCodex(modelName string, inputRawJSON []byte, stream b
 	// Model
 	out, _ = sjson.Set(out, "model", modelName)
 
+	// Build tool name shortening map from original tools (if any)
+	originalToolNameMap := map[string]string{}
+	{
+		tools := gjson.GetBytes(rawJSON, "tools")
+		if tools.IsArray() && len(tools.Array()) > 0 {
+			// Collect original tool names
+			var names []string
+			arr := tools.Array()
+			for i := 0; i < len(arr); i++ {
+				t := arr[i]
+				if t.Get("type").String() == "function" {
+					fn := t.Get("function")
+					if fn.Exists() {
+						if v := fn.Get("name"); v.Exists() {
+							names = append(names, v.String())
+						}
+					}
+				}
+			}
+			if len(names) > 0 {
+				originalToolNameMap = buildShortNameMap(names)
+			}
+		}
+	}
+
 	// Extract system instructions from first system message (string or text object)
 	messages := gjson.GetBytes(rawJSON, "messages")
 	instructions := misc.CodexInstructions
@@ -177,7 +205,15 @@ func ConvertOpenAIRequestToCodex(modelName string, inputRawJSON []byte, stream b
 								funcCall := `{}`
 								funcCall, _ = sjson.Set(funcCall, "type", "function_call")
 								funcCall, _ = sjson.Set(funcCall, "call_id", tc.Get("id").String())
-								funcCall, _ = sjson.Set(funcCall, "name", tc.Get("function.name").String())
+								{
+									name := tc.Get("function.name").String()
+									if short, ok := originalToolNameMap[name]; ok {
+										name = short
+									} else {
+										name = shortenNameIfNeeded(name)
+									}
+									funcCall, _ = sjson.Set(funcCall, "name", name)
+								}
 								funcCall, _ = sjson.Set(funcCall, "arguments", tc.Get("function.arguments").String())
 								out, _ = sjson.SetRaw(out, "input.-1", funcCall)
 							}
@@ -249,7 +285,13 @@ func ConvertOpenAIRequestToCodex(modelName string, inputRawJSON []byte, stream b
 				fn := t.Get("function")
 				if fn.Exists() {
 					if v := fn.Get("name"); v.Exists() {
-						item, _ = sjson.Set(item, "name", v.Value())
+						name := v.String()
+						if short, ok := originalToolNameMap[name]; ok {
+							name = short
+						} else {
+							name = shortenNameIfNeeded(name)
+						}
+						item, _ = sjson.Set(item, "name", name)
 					}
 					if v := fn.Get("description"); v.Exists() {
 						item, _ = sjson.Set(item, "description", v.Value())
@@ -273,3 +315,81 @@ func ConvertOpenAIRequestToCodex(modelName string, inputRawJSON []byte, stream b
 	out, _ = sjson.Set(out, "store", store)
 	return []byte(out)
 }
+
+// shortenNameIfNeeded applies the simple shortening rule for a single name.
+// If the name length exceeds 64, it will try to preserve the "mcp__" prefix and last segment.
+// Otherwise it truncates to 64 characters.
+func shortenNameIfNeeded(name string) string {
+	const limit = 64
+	if len(name) <= limit {
+		return name
+	}
+	if strings.HasPrefix(name, "mcp__") {
+		// Keep prefix and last segment after '__'
+		idx := strings.LastIndex(name, "__")
+		if idx > 0 {
+			candidate := "mcp__" + name[idx+2:]
+			if len(candidate) > limit {
+				return candidate[:limit]
+			}
+			return candidate
+		}
+	}
+	return name[:limit]
+}
+
+// buildShortNameMap generates unique short names (<=64) for the given list of names.
+// It preserves the "mcp__" prefix with the last segment when possible and ensures uniqueness
+// by appending suffixes like "~1", "~2" if needed.
+func buildShortNameMap(names []string) map[string]string {
+	const limit = 64
+	used := map[string]struct{}{}
+	m := map[string]string{}
+
+	baseCandidate := func(n string) string {
+		if len(n) <= limit {
+			return n
+		}
+		if strings.HasPrefix(n, "mcp__") {
+			idx := strings.LastIndex(n, "__")
+			if idx > 0 {
+				cand := "mcp__" + n[idx+2:]
+				if len(cand) > limit {
+					cand = cand[:limit]
+				}
+				return cand
+			}
+		}
+		return n[:limit]
+	}
+
+	makeUnique := func(cand string) string {
+		if _, ok := used[cand]; !ok {
+			return cand
+		}
+		base := cand
+		for i := 1; ; i++ {
+			suffix := "~" + strconv.Itoa(i)
+			allowed := limit - len(suffix)
+			if allowed < 0 {
+				allowed = 0
+			}
+			tmp := base
+			if len(tmp) > allowed {
+				tmp = tmp[:allowed]
+			}
+			tmp = tmp + suffix
+			if _, ok := used[tmp]; !ok {
+				return tmp
+			}
+		}
+	}
+
+	for _, n := range names {
+		cand := baseCandidate(n)
+		uniq := makeUnique(cand)
+		used[uniq] = struct{}{}
+		m[n] = uniq
+	}
+	return m
+}

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

@@ -21,9 +21,10 @@ var (
 
 // ConvertCliToOpenAIParams holds parameters for response conversion.
 type ConvertCliToOpenAIParams struct {
-	ResponseID string
-	CreatedAt  int64
-	Model      string
+	ResponseID        string
+	CreatedAt         int64
+	Model             string
+	FunctionCallIndex int
 }
 
 // ConvertCodexResponseToOpenAI translates a single chunk of a streaming response from the
@@ -43,9 +44,10 @@ type ConvertCliToOpenAIParams struct {
 func ConvertCodexResponseToOpenAI(_ context.Context, modelName string, originalRequestRawJSON, requestRawJSON, rawJSON []byte, param *any) []string {
 	if *param == nil {
 		*param = &ConvertCliToOpenAIParams{
-			Model:      modelName,
-			CreatedAt:  0,
-			ResponseID: "",
+			Model:             modelName,
+			CreatedAt:         0,
+			ResponseID:        "",
+			FunctionCallIndex: -1,
 		}
 	}
 
@@ -108,18 +110,36 @@ func ConvertCodexResponseToOpenAI(_ context.Context, modelName string, originalR
 			template, _ = sjson.Set(template, "choices.0.delta.content", deltaResult.String())
 		}
 	} else if dataType == "response.completed" {
-		template, _ = sjson.Set(template, "choices.0.finish_reason", "stop")
-		template, _ = sjson.Set(template, "choices.0.native_finish_reason", "stop")
+		finishReason := "stop"
+		if (*param).(*ConvertCliToOpenAIParams).FunctionCallIndex != -1 {
+			finishReason = "tool_calls"
+		}
+		template, _ = sjson.Set(template, "choices.0.finish_reason", finishReason)
+		template, _ = sjson.Set(template, "choices.0.native_finish_reason", finishReason)
 	} else if dataType == "response.output_item.done" {
-		functionCallItemTemplate := `{"id": "","type": "function","function": {"name": "","arguments": ""}}`
+		functionCallItemTemplate := `{"index":0,"id":"","type":"function","function":{"name":"","arguments":""}}`
 		itemResult := rootResult.Get("item")
 		if itemResult.Exists() {
 			if itemResult.Get("type").String() != "function_call" {
 				return []string{}
 			}
+
+			// set the index
+			(*param).(*ConvertCliToOpenAIParams).FunctionCallIndex++
+			functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "index", (*param).(*ConvertCliToOpenAIParams).FunctionCallIndex)
+
 			template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls", `[]`)
 			functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "id", itemResult.Get("call_id").String())
-			functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.name", itemResult.Get("name").String())
+
+			// Restore original tool name if it was shortened
+			name := itemResult.Get("name").String()
+			// Build reverse map on demand from original request tools
+			rev := buildReverseMapFromOriginalOpenAI(originalRequestRawJSON)
+			if orig, ok := rev[name]; ok {
+				name = orig
+			}
+			functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.name", name)
+
 			functionCallItemTemplate, _ = sjson.Set(functionCallItemTemplate, "function.arguments", itemResult.Get("arguments").String())
 			template, _ = sjson.Set(template, "choices.0.delta.role", "assistant")
 			template, _ = sjson.SetRaw(template, "choices.0.delta.tool_calls.-1", functionCallItemTemplate)
@@ -244,7 +264,12 @@ func ConvertCodexResponseToOpenAINonStream(_ context.Context, _ string, original
 					}
 
 					if nameResult := outputItem.Get("name"); nameResult.Exists() {
-						functionCallTemplate, _ = sjson.Set(functionCallTemplate, "function.name", nameResult.String())
+						n := nameResult.String()
+						rev := buildReverseMapFromOriginalOpenAI(originalRequestRawJSON)
+						if orig, ok := rev[n]; ok {
+							n = orig
+						}
+						functionCallTemplate, _ = sjson.Set(functionCallTemplate, "function.name", n)
 					}
 
 					if argsResult := outputItem.Get("arguments"); argsResult.Exists() {
@@ -289,3 +314,34 @@ func ConvertCodexResponseToOpenAINonStream(_ context.Context, _ string, original
 	}
 	return ""
 }
+
+// buildReverseMapFromOriginalOpenAI builds a map of shortened tool name -> original tool name
+// from the original OpenAI-style request JSON using the same shortening logic.
+func buildReverseMapFromOriginalOpenAI(original []byte) map[string]string {
+	tools := gjson.GetBytes(original, "tools")
+	rev := map[string]string{}
+	if tools.IsArray() && len(tools.Array()) > 0 {
+		var names []string
+		arr := tools.Array()
+		for i := 0; i < len(arr); i++ {
+			t := arr[i]
+			if t.Get("type").String() != "function" {
+				continue
+			}
+			fn := t.Get("function")
+			if !fn.Exists() {
+				continue
+			}
+			if v := fn.Get("name"); v.Exists() {
+				names = append(names, v.String())
+			}
+		}
+		if len(names) > 0 {
+			m := buildShortNameMap(names)
+			for orig, short := range m {
+				rev[short] = orig
+			}
+		}
+	}
+	return rev
+}

internal/watcher/watcher.go

@@ -6,11 +6,12 @@ package watcher
 
 import (
 	"context"
+	"crypto/sha256"
+	"encoding/hex"
 	"encoding/json"
 	"io/fs"
 	"net/http"
 	"os"
-	"path"
 	"path/filepath"
 	"strings"
 	"sync"
@@ -35,9 +36,11 @@ type Watcher struct {
 	authDir        string
 	config         *config.Config
 	clients        map[string]interfaces.Client
+	apiKeyClients  map[string]interfaces.Client // New field for caching API key clients
 	clientsMutex   sync.RWMutex
 	reloadCallback func(map[string]interfaces.Client, *config.Config)
 	watcher        *fsnotify.Watcher
+	lastAuthHashes map[string]string
 }
 
 // NewWatcher creates a new file watcher instance
@@ -53,6 +56,8 @@ func NewWatcher(configPath, authDir string, reloadCallback func(map[string]inter
 		reloadCallback: reloadCallback,
 		watcher:        watcher,
 		clients:        make(map[string]interfaces.Client),
+		apiKeyClients:  make(map[string]interfaces.Client),
+		lastAuthHashes: make(map[string]string),
 	}, nil
 }
 
@@ -90,13 +95,20 @@ func (w *Watcher) SetConfig(cfg *config.Config) {
 	w.config = cfg
 }
 
-// SetClients updates the current client list
+// SetClients sets the file-based clients.
 func (w *Watcher) SetClients(clients map[string]interfaces.Client) {
 	w.clientsMutex.Lock()
 	defer w.clientsMutex.Unlock()
 	w.clients = clients
 }
 
+// SetAPIKeyClients sets the API key-based clients.
+func (w *Watcher) SetAPIKeyClients(apiKeyClients map[string]interfaces.Client) {
+	w.clientsMutex.Lock()
+	defer w.clientsMutex.Unlock()
+	w.apiKeyClients = apiKeyClients
+}
+
 // processEvents handles file system events
 func (w *Watcher) processEvents(ctx context.Context) {
 	for {
@@ -202,14 +214,14 @@ func (w *Watcher) reloadConfig() {
 	w.reloadClients()
 }
 
-// reloadClients performs a full scan of the auth directory and reloads all clients.
-// This is used for initial startup and for handling config file reloads.
+// reloadClients performs a full scan and reload of all clients.
 func (w *Watcher) reloadClients() {
 	log.Debugf("starting full client reload process")
 
 	w.clientsMutex.RLock()
 	cfg := w.config
-	oldClientCount := len(w.clients)
+	oldFileClientCount := len(w.clients)
+	oldAPIKeyClientCount := len(w.apiKeyClients)
 	w.clientsMutex.RUnlock()
 
 	if cfg == nil {
@@ -217,127 +229,50 @@ func (w *Watcher) reloadClients() {
 		return
 	}
 
-	log.Debugf("scanning auth directory for initial load or full reload: %s", cfg.AuthDir)
-
-	// Create new client map
-	newClients := make(map[string]interfaces.Client)
-	authFileCount := 0
-	successfulAuthCount := 0
-
-	// Handle tilde expansion for auth directory
-	if strings.HasPrefix(cfg.AuthDir, "~") {
-		home, errUserHomeDir := os.UserHomeDir()
-		if errUserHomeDir != nil {
-			log.Fatalf("failed to get home directory: %v", errUserHomeDir)
-		}
-		parts := strings.Split(cfg.AuthDir, string(os.PathSeparator))
-		if len(parts) > 1 {
-			parts[0] = home
-			cfg.AuthDir = path.Join(parts...)
-		} else {
-			cfg.AuthDir = home
+	// Unregister all old API key clients before creating new ones
+	log.Debugf("unregistering %d old API key clients", oldAPIKeyClientCount)
+	for _, oldClient := range w.apiKeyClients {
+		if u, ok := oldClient.(interface{ UnregisterClient() }); ok {
+			u.UnregisterClient()
 		}
 	}
 
-	// Load clients from auth directory
-	errWalk := filepath.Walk(cfg.AuthDir, func(path string, info fs.FileInfo, err error) error {
-		if err != nil {
-			log.Debugf("error accessing path %s: %v", path, err)
-			return err
-		}
-		if !info.IsDir() && strings.HasSuffix(info.Name(), ".json") {
-			authFileCount++
-			log.Debugf("processing auth file %d: %s", authFileCount, filepath.Base(path))
-			if client, err := w.createClientFromFile(path, cfg); err == nil {
-				newClients[path] = client
-				successfulAuthCount++
-			} else {
-				log.Errorf("failed to create client from file %s: %v", path, err)
-			}
-		}
-		return nil
-	})
-	if errWalk != nil {
-		log.Errorf("error walking auth directory: %v", errWalk)
-		return
-	}
-	log.Debugf("auth directory scan complete - found %d .json files, %d successful authentications", authFileCount, successfulAuthCount)
+	// Create new API key clients based on the new config
+	newAPIKeyClients, glAPIKeyCount, claudeAPIKeyCount, codexAPIKeyCount, openAICompatCount := buildAPIKeyClients(cfg)
+	log.Debugf("created %d new API key clients", len(newAPIKeyClients))
 
-	// Note: API key-based clients are not stored in the map as they don't correspond to a file.
-	// They are re-created each time, which is lightweight.
-	clientSlice := w.clientsToSlice(newClients)
+	// Load file-based clients
+	newFileClients, successfulAuthCount := w.loadFileClients(cfg)
+	log.Debugf("loaded %d new file-based clients", len(newFileClients))
 
-	// Add clients for Generative Language API keys if configured
-	glAPIKeyCount := 0
-	if len(cfg.GlAPIKey) > 0 {
-		log.Debugf("processing %d Generative Language API Keys", len(cfg.GlAPIKey))
-		for i := 0; i < len(cfg.GlAPIKey); i++ {
-			httpClient := util.SetProxy(cfg, &http.Client{})
-			log.Debugf("Initializing with Generative Language API Key %d...", i+1)
-			cliClient := client.NewGeminiClient(httpClient, cfg, cfg.GlAPIKey[i])
-			clientSlice = append(clientSlice, cliClient)
-			glAPIKeyCount++
-		}
-		log.Debugf("Successfully initialized %d Generative Language API Key clients", glAPIKeyCount)
-	}
-	// ... (Claude, Codex, OpenAI-compat clients are handled similarly) ...
-	claudeAPIKeyCount := 0
-	if len(cfg.ClaudeKey) > 0 {
-		log.Debugf("processing %d Claude API Keys", len(cfg.ClaudeKey))
-		for i := 0; i < len(cfg.ClaudeKey); i++ {
-			log.Debugf("Initializing with Claude API Key %d...", i+1)
-			cliClient := client.NewClaudeClientWithKey(cfg, i)
-			clientSlice = append(clientSlice, cliClient)
-			claudeAPIKeyCount++
-		}
-		log.Debugf("Successfully initialized %d Claude API Key clients", claudeAPIKeyCount)
-	}
-
-	codexAPIKeyCount := 0
-	if len(cfg.CodexKey) > 0 {
-		log.Debugf("processing %d Codex API Keys", len(cfg.CodexKey))
-		for i := 0; i < len(cfg.CodexKey); i++ {
-			log.Debugf("Initializing with Codex API Key %d...", i+1)
-			cliClient := client.NewCodexClientWithKey(cfg, i)
-			clientSlice = append(clientSlice, cliClient)
-			codexAPIKeyCount++
-		}
-		log.Debugf("Successfully initialized %d Codex API Key clients", codexAPIKeyCount)
-	}
-
-	openAICompatCount := 0
-	if len(cfg.OpenAICompatibility) > 0 {
-		log.Debugf("processing %d OpenAI-compatibility providers", len(cfg.OpenAICompatibility))
-		for i := 0; i < len(cfg.OpenAICompatibility); i++ {
-			compat := cfg.OpenAICompatibility[i]
-			compatClient, errClient := client.NewOpenAICompatibilityClient(cfg, &compat)
-			if errClient != nil {
-				log.Errorf("  failed to create OpenAI-compatibility client for %s: %v", compat.Name, errClient)
-				continue
-			}
-			clientSlice = append(clientSlice, compatClient)
-			openAICompatCount++
-		}
-		log.Debugf("Successfully initialized %d OpenAI-compatibility clients", openAICompatCount)
-	}
-
-	// Unregister all old clients
-	w.clientsMutex.RLock()
+	// Unregister all old file-based clients
+	log.Debugf("unregistering %d old file-based clients", oldFileClientCount)
 	for _, oldClient := range w.clients {
 		if u, ok := any(oldClient).(interface{ UnregisterClient() }); ok {
 			u.UnregisterClient()
 		}
 	}
-	w.clientsMutex.RUnlock()
 
-	// Update the client map
+	// Update client maps
 	w.clientsMutex.Lock()
-	w.clients = newClients
+	w.clients = newFileClients
+	w.apiKeyClients = newAPIKeyClients
+
+	// Rebuild auth file hash cache for current clients
+	w.lastAuthHashes = make(map[string]string, len(newFileClients))
+	for path := range newFileClients {
+		if data, err := os.ReadFile(path); err == nil && len(data) > 0 {
+			sum := sha256.Sum256(data)
+			w.lastAuthHashes[path] = hex.EncodeToString(sum[:])
+		}
+	}
 	w.clientsMutex.Unlock()
 
+	totalNewClients := len(newFileClients) + len(newAPIKeyClients)
+
 	log.Infof("full client reload complete - old: %d clients, new: %d clients (%d auth files + %d GL API keys + %d Claude API keys + %d Codex keys + %d OpenAI-compat)",
-		oldClientCount,
-		len(clientSlice),
+		oldFileClientCount+oldAPIKeyClientCount,
+		totalNewClients,
 		successfulAuthCount,
 		glAPIKeyCount,
 		claudeAPIKeyCount,
@@ -348,9 +283,8 @@ func (w *Watcher) reloadClients() {
 	// Trigger the callback to update the server
 	if w.reloadCallback != nil {
 		log.Debugf("triggering server update callback")
-		// Note: The callback signature expects a map now, but the API server internally works with a slice.
-		// We pass the map directly, and the server will handle converting it.
-		w.reloadCallback(w.clients, cfg)
+		combinedClients := w.buildCombinedClientMap()
+		w.reloadCallback(combinedClients, cfg)
 	}
 }
 
@@ -417,15 +351,38 @@ func (w *Watcher) clientsToSlice(clientMap map[string]interfaces.Client) []inter
 // addOrUpdateClient handles the addition or update of a single client.
 func (w *Watcher) addOrUpdateClient(path string) {
 	w.clientsMutex.Lock()
-	defer w.clientsMutex.Unlock()
 
 	cfg := w.config
 	if cfg == nil {
 		log.Error("config is nil, cannot add or update client")
+		w.clientsMutex.Unlock()
 		return
 	}
 
-	// Unregister old client if it exists
+	// Read file to check for emptiness and calculate hash
+	data, errRead := os.ReadFile(path)
+	if errRead != nil {
+		log.Errorf("failed to read auth file %s: %v", filepath.Base(path), errRead)
+		w.clientsMutex.Unlock()
+		return
+	}
+	if len(data) == 0 {
+		// Empty file: ignore (wait for a subsequent WRITE)
+		log.Debugf("ignoring empty auth file: %s", filepath.Base(path))
+		w.clientsMutex.Unlock()
+		return
+	}
+
+	// Calculate a hash of the current content and compare with the cache
+	sum := sha256.Sum256(data)
+	curHash := hex.EncodeToString(sum[:])
+	if prev, ok := w.lastAuthHashes[path]; ok && prev == curHash {
+		log.Debugf("auth file unchanged (hash match), skipping reload: %s", filepath.Base(path))
+		w.clientsMutex.Unlock()
+		return
+	}
+
+	// If an old client exists, unregister it first
 	if oldClient, ok := w.clients[path]; ok {
 		if u, canUnregister := any(oldClient).(interface{ UnregisterClient() }); canUnregister {
 			log.Debugf("unregistering old client for updated file: %s", filepath.Base(path))
@@ -433,32 +390,42 @@ func (w *Watcher) addOrUpdateClient(path string) {
 		}
 	}
 
+	// Create new client (reads the file again internally; this is acceptable as the files are small and it keeps the change minimal)
 	newClient, err := w.createClientFromFile(path, cfg)
 	if err != nil {
 		log.Errorf("failed to create/update client for %s: %v", filepath.Base(path), err)
-		// If creation fails, ensure the old client is removed from the map
+		// If creation fails, ensure the old client is removed from the map; don't update hash, let a subsequent change retry
 		delete(w.clients, path)
-	} else if newClient != nil { // Only update if a client was actually created
-		log.Debugf("successfully created/updated client for %s", filepath.Base(path))
-		w.clients[path] = newClient
-	} else {
-		// This case handles the empty file scenario gracefully
-		log.Debugf("ignoring empty auth file: %s", filepath.Base(path))
-		return // Do not trigger callback for an empty file
+		w.clientsMutex.Unlock()
+		return
 	}
+	if newClient == nil {
+		// This branch should not be reached normally (empty files are handled above); a fallback
+		log.Debugf("ignoring auth file with no client created: %s", filepath.Base(path))
+		w.clientsMutex.Unlock()
+		return
+	}
+
+	// Update client and hash cache
+	log.Debugf("successfully created/updated client for %s", filepath.Base(path))
+	w.clients[path] = newClient
+	w.lastAuthHashes[path] = curHash
+
+	w.clientsMutex.Unlock() // Unlock before the callback
 
 	if w.reloadCallback != nil {
 		log.Debugf("triggering server update callback after add/update")
-		w.reloadCallback(w.clients, cfg)
+		combinedClients := w.buildCombinedClientMap()
+		w.reloadCallback(combinedClients, cfg)
 	}
 }
 
 // removeClient handles the removal of a single client.
 func (w *Watcher) removeClient(path string) {
 	w.clientsMutex.Lock()
-	defer w.clientsMutex.Unlock()
 
 	cfg := w.config
+	var clientRemoved bool
 
 	// Unregister client if it exists
 	if oldClient, ok := w.clients[path]; ok {
@@ -467,11 +434,121 @@ func (w *Watcher) removeClient(path string) {
 			u.UnregisterClient()
 		}
 		delete(w.clients, path)
+		delete(w.lastAuthHashes, path)
 		log.Debugf("removed client for %s", filepath.Base(path))
+		clientRemoved = true
+	}
 
-		if w.reloadCallback != nil {
-			log.Debugf("triggering server update callback after removal")
-			w.reloadCallback(w.clients, cfg)
-		}
+	w.clientsMutex.Unlock() // Release the lock before the callback
+
+	if clientRemoved && w.reloadCallback != nil {
+		log.Debugf("triggering server update callback after removal")
+		combinedClients := w.buildCombinedClientMap()
+		w.reloadCallback(combinedClients, cfg)
 	}
 }
+
+// buildCombinedClientMap merges file-based clients with API key clients from the cache.
+func (w *Watcher) buildCombinedClientMap() map[string]interfaces.Client {
+	w.clientsMutex.RLock()
+	defer w.clientsMutex.RUnlock()
+
+	combined := make(map[string]interfaces.Client)
+
+	// Add file-based clients
+	for k, v := range w.clients {
+		combined[k] = v
+	}
+
+	// Add cached API key-based clients
+	for k, v := range w.apiKeyClients {
+		combined[k] = v
+	}
+
+	return combined
+}
+
+// loadFileClients scans the auth directory and creates clients from .json files.
+func (w *Watcher) loadFileClients(cfg *config.Config) (map[string]interfaces.Client, int) {
+	newClients := make(map[string]interfaces.Client)
+	authFileCount := 0
+	successfulAuthCount := 0
+
+	authDir := cfg.AuthDir
+	if strings.HasPrefix(authDir, "~") {
+		home, err := os.UserHomeDir()
+		if err != nil {
+			log.Errorf("failed to get home directory: %v", err)
+			return newClients, 0
+		}
+		authDir = filepath.Join(home, authDir[1:])
+	}
+
+	errWalk := filepath.Walk(authDir, func(path string, info fs.FileInfo, err error) error {
+		if err != nil {
+			log.Debugf("error accessing path %s: %v", path, err)
+			return err
+		}
+		if !info.IsDir() && strings.HasSuffix(info.Name(), ".json") {
+			authFileCount++
+			log.Debugf("processing auth file %d: %s", authFileCount, filepath.Base(path))
+			if cliClient, errCreate := w.createClientFromFile(path, cfg); errCreate == nil && cliClient != nil {
+				newClients[path] = cliClient
+				successfulAuthCount++
+			} else if errCreate != nil {
+				log.Errorf("failed to create client from file %s: %v", path, errCreate)
+			}
+		}
+		return nil
+	})
+
+	if errWalk != nil {
+		log.Errorf("error walking auth directory: %v", errWalk)
+	}
+	log.Debugf("auth directory scan complete - found %d .json files, %d successful authentications", authFileCount, successfulAuthCount)
+	return newClients, successfulAuthCount
+}
+
+// buildAPIKeyClients creates clients from API keys in the config.
+func buildAPIKeyClients(cfg *config.Config) (map[string]interfaces.Client, int, int, int, int) {
+	apiKeyClients := make(map[string]interfaces.Client)
+	glAPIKeyCount := 0
+	claudeAPIKeyCount := 0
+	codexAPIKeyCount := 0
+	openAICompatCount := 0
+
+	if len(cfg.GlAPIKey) > 0 {
+		for _, key := range cfg.GlAPIKey {
+			httpClient := util.SetProxy(cfg, &http.Client{})
+			cliClient := client.NewGeminiClient(httpClient, cfg, key)
+			apiKeyClients[cliClient.GetClientID()] = cliClient
+			glAPIKeyCount++
+		}
+	}
+	if len(cfg.ClaudeKey) > 0 {
+		for i := range cfg.ClaudeKey {
+			cliClient := client.NewClaudeClientWithKey(cfg, i)
+			apiKeyClients[cliClient.GetClientID()] = cliClient
+			claudeAPIKeyCount++
+		}
+	}
+	if len(cfg.CodexKey) > 0 {
+		for i := range cfg.CodexKey {
+			cliClient := client.NewCodexClientWithKey(cfg, i)
+			apiKeyClients[cliClient.GetClientID()] = cliClient
+			codexAPIKeyCount++
+		}
+	}
+	if len(cfg.OpenAICompatibility) > 0 {
+		for _, compatConfig := range cfg.OpenAICompatibility {
+			compatClient, errClient := client.NewOpenAICompatibilityClient(cfg, &compatConfig)
+			if errClient != nil {
+				log.Errorf("failed to create OpenAI-compatibility client for %s: %v", compatConfig.Name, errClient)
+				continue
+			}
+			apiKeyClients[compatClient.GetClientID()] = compatClient
+			openAICompatCount++
+		}
+	}
+	return apiKeyClients, glAPIKeyCount, claudeAPIKeyCount, codexAPIKeyCount, openAICompatCount
+}