feat(vertex): 集成 Vertex AI Express API 支持

本次更新引入了对 Google Vertex AI Express API 的支持，允许用户配置和使用 Vertex AI 模型。

主要变更包括：

后端：
- 新增 `VERTEX_API_KEYS` 和 `VERTEX_EXPRESS_BASE_URL` 至系统配置 ([`.env.example`](.env.example:13), [`app/config/config.py:62`](app/config/config.py:62), [`app/database/models.py`](app/database/models.py), [`app/database/services.py`](app/database/services.py))。
- 实现 `VertexExpressChatService` ([`app/service/chat/vertex_express_chat_service.py`](app/service/chat/vertex_express_chat_service.py)) 用于处理与 Vertex AI Express API 的交互。
- 添加 `vertex_express_routes` ([`app/router/vertex_express_routes.py`](app/router/vertex_express_routes.py)) 来暴露 Vertex AI 相关的 API 端点，并集成到主应用 ([`app/core/application.py:36`](app/core/application.py:36), [`app/router/routes.py:15`](app/router/routes.py:15))。
- 更新密钥管理器 ([`app/service/key/key_manager.py`](app/service/key/key_manager.py)) 以支持 Vertex API 密钥的获取、检查和轮换。

前端 (配置编辑器):
- 在配置页面 ([`app/templates/config_editor.html:463`](app/templates/config_editor.html:463)) 添加了 Vertex API 密钥列表和 Vertex Express API 基础 URL 的表单字段。
- 实现了批量添加和删除 Vertex API 密钥的功能，包括相应的模态框和操作逻辑 ([`app/static/js/config_editor.js:550`](app/static/js/config_editor.js:550), [`app/static/js/config_editor.js:1097`](app/static/js/config_editor.js:1097), [`app/templates/config_editor.html:1657`](app/templates/config_editor.html:1657))。
- 确保新的配置项在初始化 ([`app/static/js/config_editor.js:598`](app/static/js/config_editor.js:598)) 和表单填充 ([`app/static/js/config_editor.js:671`](app/static/js/config_editor.js:671)) 时得到正确处理。
- 更新了数组项添加逻辑以识别 `VERTEX_API_KEYS` 为敏感字段 ([`app/static/js/config_editor.js:1235`](app/static/js/config_editor.js:1235))。

此功能扩展了应用支持的 AI 服务范围，为用户提供了更多模型选择。

.env.example

@@ -10,6 +10,10 @@ MYSQL_DATABASE=default_db
 API_KEYS=["AIzaSyxxxxxxxxxxxxxxxxxxx","AIzaSyxxxxxxxxxxxxxxxxxxx"]
 ALLOWED_TOKENS=["sk-123456"]
 AUTH_TOKEN=sk-123456
+# For Vertex AI Platform API Keys
+VERTEX_API_KEYS=["AQ.Abxxxxxxxxxxxxxxxxxxx"]
+# For Vertex AI Platform Express API Base URL
+VERTEX_EXPRESS_BASE_URL=https://aiplatform.googleapis.com/v1beta1/publishers/google
 TEST_MODEL=gemini-1.5-flash
 THINKING_MODELS=["gemini-2.5-flash-preview-04-17"]
 THINKING_BUDGET_MAP={"gemini-2.5-flash-preview-04-17": 4000}

app/config/config.py

@@ -59,8 +59,10 @@ class Settings(BaseSettings):
     TEST_MODEL: str = DEFAULT_MODEL
     TIME_OUT: int = DEFAULT_TIMEOUT
     MAX_RETRIES: int = MAX_RETRIES
-    PROXIES: List[str] = []  # 新增：代理服务器列表
+    PROXIES: List[str] = []
-
+    VERTEX_API_KEYS: List[str] = []
+    VERTEX_EXPRESS_BASE_URL: str = "https://aiplatform.googleapis.com/v1beta1/publishers/google"
+ 
     # 模型相关配置
     SEARCH_MODELS: List[str] = ["gemini-2.0-flash-exp"]
     IMAGE_MODELS: List[str] = ["gemini-2.0-flash-exp"]
@@ -68,8 +70,8 @@ class Settings(BaseSettings):
     TOOLS_CODE_EXECUTION_ENABLED: bool = False
     SHOW_SEARCH_LINK: bool = True
     SHOW_THINKING_PROCESS: bool = True
-    THINKING_MODELS: List[str] = []  # 新增：用于思考过程的模型列表
+    THINKING_MODELS: List[str] = []
-    THINKING_BUDGET_MAP: Dict[str, float] = {}  # 新增：模型对应的预算映射
+    THINKING_BUDGET_MAP: Dict[str, float] = {}
 
     # 图像生成相关配置
     PAID_KEY: str = ""
@@ -101,12 +103,12 @@ class Settings(BaseSettings):
     GITHUB_REPO_NAME: str = "gemini-balance"
 
     # 日志配置
-    LOG_LEVEL: str = "INFO"  # 默认日志级别
+    LOG_LEVEL: str = "INFO"
-    AUTO_DELETE_ERROR_LOGS_ENABLED: bool = True  # 是否开启自动删除错误日志
+    AUTO_DELETE_ERROR_LOGS_ENABLED: bool = True
-    AUTO_DELETE_ERROR_LOGS_DAYS: int = 7  # 自动删除多少天前的错误日志 (1, 7, 30)
+    AUTO_DELETE_ERROR_LOGS_DAYS: int = 7
-    AUTO_DELETE_REQUEST_LOGS_ENABLED: bool = False  # 是否开启自动删除请求日志
+    AUTO_DELETE_REQUEST_LOGS_ENABLED: bool = False
-    AUTO_DELETE_REQUEST_LOGS_DAYS: int = 30  # 自动删除多少天前的请求日志 (1, 7, 30)
+    AUTO_DELETE_REQUEST_LOGS_DAYS: int = 30
-    SAFETY_SETTINGS: List[Dict[str, str]] = DEFAULT_SAFETY_SETTINGS  # 新增：安全设置
+    SAFETY_SETTINGS: List[Dict[str, str]] = DEFAULT_SAFETY_SETTINGS
 
     def __init__(self, **kwargs):
         super().__init__(**kwargs)
@@ -141,7 +143,6 @@ def _parse_db_value(key: str, db_value: str, target_type: Type) -> Any:
         elif target_type == Dict[str, float]:
             parsed_dict = {}
             try:
-                # First attempt: standard JSON parsing
                 parsed = json.loads(db_value)
                 if isinstance(parsed, dict):
                     parsed_dict = {str(k): float(v) for k, v in parsed.items()}
@@ -150,7 +151,6 @@ def _parse_db_value(key: str, db_value: str, target_type: Type) -> Any:
                         f"Parsed DB value for key '{key}' is not a dictionary type. Value: {db_value}"
                     )
             except (json.JSONDecodeError, ValueError, TypeError) as e1:
-                # Second attempt: try replacing single quotes if JSONDecodeError occurred
                 if isinstance(e1, json.JSONDecodeError) and "'" in db_value:
                     logger.warning(
                         f"Failed initial JSON parse for key '{key}'. Attempting to replace single quotes. Error: {e1}"
@@ -169,11 +169,10 @@ def _parse_db_value(key: str, db_value: str, target_type: Type) -> Any:
                             f"Could not parse '{db_value}' as Dict[str, float] for key '{key}' even after replacing quotes: {e2}. Returning empty dict."
                         )
                 else:
-                    # Log other errors (ValueError, TypeError) or JSON errors without single quotes
                     logger.error(
                         f"Could not parse '{db_value}' as Dict[str, float] for key '{key}': {e1}. Returning empty dict."
                     )
-            return parsed_dict  # Return the parsed dict or an empty one if all attempts fail
+            return parsed_dict
         # 处理 List[Dict[str, str]]
         elif target_type == List[Dict[str, str]]:
             try:
@@ -192,7 +191,7 @@ def _parse_db_value(key: str, db_value: str, target_type: Type) -> Any:
                         logger.warning(
                             f"Invalid structure in List[Dict[str, str]] for key '{key}'. Value: {db_value}"
                         )
-                        return []  # 或者返回默认值？这里返回空列表
+                        return []
                 else:
                     logger.warning(
                         f"Parsed DB value for key '{key}' is not a list type. Value: {db_value}"
@@ -374,7 +373,7 @@ async def sync_initial_settings():
             data = {
                 "key": key,
                 "value": db_value,
-                "description": f"{key} configuration setting",  # 默认描述
+                "description": f"{key} configuration setting",
                 "updated_at": now,
             }
 

app/core/application.py

@@ -15,7 +15,7 @@ from app.router.routes import setup_routers
 from app.scheduler.scheduled_tasks import start_scheduler, stop_scheduler
 from app.service.key.key_manager import get_key_manager_instance
 from app.service.update.update_service import check_for_updates
-from app.utils.helpers import get_current_version  # Import from helpers
+from app.utils.helpers import get_current_version
 
 logger = get_application_logger()
 
@@ -43,7 +43,7 @@ async def _setup_database_and_config(app_settings):
     logger.info("Database initialized successfully")
     await connect_to_db()
     await sync_initial_settings()
-    await get_key_manager_instance(app_settings.API_KEYS)
+    await get_key_manager_instance(app_settings.API_KEYS, app_settings.VERTEX_API_KEYS)
     logger.info("Database, config sync, and KeyManager initialized successfully")
 
 

app/database/connection.py

@@ -45,11 +45,8 @@ Base = declarative_base(metadata=metadata)
 if settings.DATABASE_TYPE == "sqlite":
     database = Database(DATABASE_URL)
 else:
-    database = Database(DATABASE_URL, min_size=5, max_size=20, pool_recycle=1800) # Reduced recycle time to 30 mins
+    database = Database(DATABASE_URL, min_size=5, max_size=20, pool_recycle=1800)
 
-# 移除了 SessionLocal 和 get_db 函数
-
-# --- Async connection functions for lifespan/async routes ---
 async def connect_to_db():
     """
     连接到数据库

app/database/models.py

@@ -2,7 +2,7 @@
 数据库模型模块
 """
 import datetime
-from sqlalchemy import Column, Integer, String, Text, DateTime, JSON, Boolean # 添加 Boolean
+from sqlalchemy import Column, Integer, String, Text, DateTime, JSON, Boolean
 
 from app.database.connection import Base
 
@@ -53,7 +53,7 @@ class RequestLog(Base):
     id = Column(Integer, primary_key=True, autoincrement=True)
     request_time = Column(DateTime, default=datetime.datetime.now, comment="请求时间")
     model_name = Column(String(100), nullable=True, comment="模型名称")
-    api_key = Column(String(100), nullable=True, comment="使用的API密钥") # 考虑安全性，后续可优化
+    api_key = Column(String(100), nullable=True, comment="使用的API密钥")
     is_success = Column(Boolean, nullable=False, comment="请求是否成功")
     status_code = Column(Integer, nullable=True, comment="API响应状态码")
     latency_ms = Column(Integer, nullable=True, comment="请求耗时(毫秒)")

app/database/services.py

@@ -189,7 +189,6 @@ async def get_error_logs(
             ErrorLog.request_time
         )
         
-        # Apply filters
         if key_search:
             query = query.where(ErrorLog.gemini_key.ilike(f"%{key_search}%"))
         if error_search:
@@ -219,14 +218,14 @@ async def get_error_logs(
         result = await database.fetch_all(query)
         return [dict(row) for row in result]
     except Exception as e:
-        logger.exception(f"Failed to get error logs with filters: {str(e)}") # Use exception for stack trace
+        logger.exception(f"Failed to get error logs with filters: {str(e)}")
         raise
 
 
 async def get_error_logs_count(
     key_search: Optional[str] = None,
     error_search: Optional[str] = None,
-    error_code_search: Optional[str] = None, # Added error code search
+    error_code_search: Optional[str] = None,
     start_date: Optional[datetime] = None,
     end_date: Optional[datetime] = None
 ) -> int:
@@ -294,7 +293,7 @@ async def get_error_log_details(log_id: int) -> Optional[Dict[str, Any]]:
                 try:
                     log_dict['request_msg'] = json.dumps(log_dict['request_msg'], ensure_ascii=False, indent=2)
                 except TypeError:
-                    log_dict['request_msg'] = str(log_dict['request_msg']) # Fallback to string
+                    log_dict['request_msg'] = str(log_dict['request_msg'])
             return log_dict
         else:
             return None
@@ -372,7 +371,6 @@ async def delete_all_error_logs() -> int:
     try:
         # 1. 获取删除前的总数
         count_query = select(func.count()).select_from(ErrorLog)
-        # fetch_val() is suitable here as we expect a single scalar value
         total_to_delete = await database.fetch_val(count_query)
  
         if total_to_delete == 0:
@@ -380,7 +378,6 @@ async def delete_all_error_logs() -> int:
             return 0
  
         # 2. 执行删除操作
-        # This creates a query like "DELETE FROM error_log"
         delete_query = delete(ErrorLog)
         await database.execute(delete_query)
         
@@ -388,7 +385,6 @@ async def delete_all_error_logs() -> int:
         return total_to_delete
     except Exception as e:
         logger.error(f"Failed to delete all error logs: {str(e)}", exc_info=True)
-        # Re-raise the exception so it can be caught by the service layer or route handler
         raise
  
  

app/log/logger.py

@@ -226,3 +226,8 @@ def get_error_log_logger():
 
 def get_request_log_logger():
     return Logger.setup_logger("request_log")
+
+
+def get_vertex_express_logger():
+    return Logger.setup_logger("vertex_express")
+

app/middleware/middleware.py

@@ -32,6 +32,7 @@ class AuthMiddleware(BaseHTTPMiddleware):
             and not request.url.path.startswith("/hf")
             and not request.url.path.startswith("/openai")
             and not request.url.path.startswith("/api/version/check")
+            and not request.url.path.startswith("/vertex-express")
         ):
 
             auth_token = request.cookies.get("auth_token")
@@ -60,7 +61,7 @@ def setup_middlewares(app: FastAPI) -> None:
     # 配置CORS中间件
     app.add_middleware(
         CORSMiddleware,
-        allow_origins=["*"],  # 生产环境建议配置具体的域名
+        allow_origins=["*"],
         allow_credentials=True,
         allow_methods=[
             "GET",
@@ -68,8 +69,8 @@ def setup_middlewares(app: FastAPI) -> None:
             "PUT",
             "DELETE",
             "OPTIONS",
-        ],  # 明确指定允许的HTTP方法
+        ],
-        allow_headers=["*"],  # 生产环境建议配置具体的请求头
+        allow_headers=["*"],
-        expose_headers=["*"],  # 允许前端访问的响应头
+        expose_headers=["*"],
-        max_age=600,  # 预检请求缓存时间(秒)
+        max_age=600,
     )

app/router/routes.py

@@ -8,7 +8,7 @@ from fastapi.templating import Jinja2Templates
 
 from app.core.security import verify_auth_token
 from app.log.logger import get_routes_logger
-from app.router import error_log_routes, gemini_routes, openai_routes, config_routes, scheduler_routes, stats_routes, version_routes, openai_compatiable_routes
+from app.router import error_log_routes, gemini_routes, openai_routes, config_routes, scheduler_routes, stats_routes, version_routes, openai_compatiable_routes, vertex_express_routes
 from app.service.key.key_manager import get_key_manager_instance
 from app.service.stats.stats_service import StatsService
 
@@ -33,6 +33,7 @@ def setup_routers(app: FastAPI) -> None:
     app.include_router(stats_routes.router)
     app.include_router(version_routes.router)
     app.include_router(openai_compatiable_routes.router)
+    app.include_router(vertex_express_routes.router)
 
     setup_page_routes(app)
 

app/router/vertex_express_routes.py

@@ -0,0 +1,146 @@
+from fastapi import APIRouter, Depends, HTTPException
+from fastapi.responses import StreamingResponse
+from copy import deepcopy
+from app.config.config import settings
+from app.log.logger import get_vertex_express_logger
+from app.core.security import SecurityService
+from app.domain.gemini_models import GeminiRequest
+from app.service.chat.vertex_express_chat_service import GeminiChatService
+from app.service.key.key_manager import KeyManager, get_key_manager_instance
+from app.service.model.model_service import ModelService
+from app.handler.retry_handler import RetryHandler
+from app.handler.error_handler import handle_route_errors
+from app.core.constants import API_VERSION
+
+router = APIRouter(prefix=f"/vertex-express/{API_VERSION}")
+logger = get_vertex_express_logger()
+
+security_service = SecurityService()
+model_service = ModelService()
+
+
+async def get_key_manager():
+    """获取密钥管理器实例"""
+    return await get_key_manager_instance()
+
+
+async def get_next_working_key(key_manager: KeyManager = Depends(get_key_manager)):
+    """获取下一个可用的API密钥"""
+    return await key_manager.get_next_working_vertex_key()
+
+
+async def get_chat_service(key_manager: KeyManager = Depends(get_key_manager)):
+    """获取Gemini聊天服务实例"""
+    return GeminiChatService(settings.VERTEX_EXPRESS_BASE_URL, key_manager)
+
+
+@router.get("/models")
+async def list_models(
+    _=Depends(security_service.verify_key_or_goog_api_key),
+    key_manager: KeyManager = Depends(get_key_manager)
+):
+    """获取可用的 Gemini 模型列表，并根据配置添加衍生模型（搜索、图像、非思考）。"""
+    operation_name = "list_gemini_models"
+    logger.info("-" * 50 + operation_name + "-" * 50)
+    logger.info("Handling Gemini models list request")
+
+    try:
+        api_key = await key_manager.get_first_valid_key()
+        if not api_key:
+            raise HTTPException(status_code=503, detail="No valid API keys available to fetch models.")
+        logger.info(f"Using API key: {api_key}")
+
+        models_data = await model_service.get_gemini_models(api_key)
+        if not models_data or "models" not in models_data:
+            raise HTTPException(status_code=500, detail="Failed to fetch base models list.")
+
+        models_json = deepcopy(models_data)
+        model_mapping = {x.get("name", "").split("/", maxsplit=1)[-1]: x for x in models_json.get("models", [])}
+
+        def add_derived_model(base_name, suffix, display_suffix):
+            model = model_mapping.get(base_name)
+            if not model:
+                logger.warning(f"Base model '{base_name}' not found for derived model '{suffix}'.")
+                return
+            item = deepcopy(model)
+            item["name"] = f"models/{base_name}{suffix}"
+            display_name = f'{item.get("displayName", base_name)}{display_suffix}'
+            item["displayName"] = display_name
+            item["description"] = display_name
+            models_json["models"].append(item)
+
+        if settings.SEARCH_MODELS:
+            for name in settings.SEARCH_MODELS:
+                add_derived_model(name, "-search", " For Search")
+        if settings.IMAGE_MODELS:
+            for name in settings.IMAGE_MODELS:
+                 add_derived_model(name, "-image", " For Image")
+        if settings.THINKING_MODELS:
+            for name in settings.THINKING_MODELS:
+                add_derived_model(name, "-non-thinking", " Non Thinking")
+
+        logger.info("Gemini models list request successful")
+        return models_json
+    except HTTPException as http_exc:
+        raise http_exc
+    except Exception as e:
+        logger.error(f"Error getting Gemini models list: {str(e)}")
+        raise HTTPException(
+            status_code=500, detail="Internal server error while fetching Gemini models list"
+        ) from e
+
+
+@router.post("/models/{model_name}:generateContent")
+@RetryHandler(key_arg="api_key")
+async def generate_content(
+    model_name: str,
+    request: GeminiRequest,
+    _=Depends(security_service.verify_key_or_goog_api_key),
+    api_key: str = Depends(get_next_working_key),
+    key_manager: KeyManager = Depends(get_key_manager),
+    chat_service: GeminiChatService = Depends(get_chat_service)
+):
+    """处理 Gemini 非流式内容生成请求。"""
+    operation_name = "gemini_generate_content"
+    async with handle_route_errors(logger, operation_name, failure_message="Content generation failed"):
+        logger.info(f"Handling Gemini content generation request for model: {model_name}")
+        logger.debug(f"Request: \n{request.model_dump_json(indent=2)}")
+        logger.info(f"Using API key: {api_key}")
+
+        if not await model_service.check_model_support(model_name):
+            raise HTTPException(status_code=400, detail=f"Model {model_name} is not supported")
+
+        response = await chat_service.generate_content(
+            model=model_name,
+            request=request,
+            api_key=api_key
+        )
+        return response
+
+
+@router.post("/models/{model_name}:streamGenerateContent")
+@RetryHandler(key_arg="api_key")
+async def stream_generate_content(
+    model_name: str,
+    request: GeminiRequest,
+    _=Depends(security_service.verify_key_or_goog_api_key),
+    api_key: str = Depends(get_next_working_key),
+    key_manager: KeyManager = Depends(get_key_manager),
+    chat_service: GeminiChatService = Depends(get_chat_service)
+):
+    """处理 Gemini 流式内容生成请求。"""
+    operation_name = "gemini_stream_generate_content"
+    async with handle_route_errors(logger, operation_name, failure_message="Streaming request initiation failed"):
+        logger.info(f"Handling Gemini streaming content generation for model: {model_name}")
+        logger.debug(f"Request: \n{request.model_dump_json(indent=2)}")
+        logger.info(f"Using API key: {api_key}")
+
+        if not await model_service.check_model_support(model_name):
+            raise HTTPException(status_code=400, detail=f"Model {model_name} is not supported")
+
+        response_stream = chat_service.stream_generate_content(
+            model=model_name,
+            request=request,
+            api_key=api_key
+        )
+        return StreamingResponse(response_stream, media_type="text/event-stream")

app/service/chat/gemini_chat_service.py

@@ -131,7 +131,7 @@ class GeminiChatService:
         self, original_response: Dict[str, Any], text: str
     ) -> Dict[str, Any]:
         """创建包含指定文本的响应"""
-        response_copy = json.loads(json.dumps(original_response))  # 深拷贝
+        response_copy = json.loads(json.dumps(original_response))
         if response_copy.get("candidates") and response_copy["candidates"][0].get(
             "content", {}
         ).get("parts"):
@@ -200,7 +200,7 @@ class GeminiChatService:
             request_datetime = datetime.datetime.now()
             start_time = time.perf_counter()
             current_attempt_key = api_key
-            final_api_key = current_attempt_key # Update final key used
+            final_api_key = current_attempt_key
             try:
                 async for line in self.api_client.stream_generate_content(
                     payload, model, current_attempt_key

app/service/chat/openai_chat_service.py

@@ -51,7 +51,7 @@ def _build_tools(
             or model.endswith("-image")
             or model.endswith("-image-generation")
         )
-        and not _has_media_parts(messages)  # Use the updated check
+        and not _has_media_parts(messages)
     ):
         tool["codeExecution"] = {}
         logger.debug("Code execution tool enabled.")

app/service/chat/vertex_express_chat_service.py

@@ -0,0 +1,277 @@
+# app/services/chat_service.py
+
+import json
+import re
+import datetime
+import time
+from typing import Any, AsyncGenerator, Dict, List
+from app.config.config import settings
+from app.core.constants import GEMINI_2_FLASH_EXP_SAFETY_SETTINGS
+from app.domain.gemini_models import GeminiRequest
+from app.handler.response_handler import GeminiResponseHandler
+from app.handler.stream_optimizer import gemini_optimizer
+from app.log.logger import get_gemini_logger
+from app.service.client.api_client import GeminiApiClient
+from app.service.key.key_manager import KeyManager
+from app.database.services import add_error_log, add_request_log
+
+logger = get_gemini_logger()
+
+
+def _has_image_parts(contents: List[Dict[str, Any]]) -> bool:
+    """判断消息是否包含图片部分"""
+    for content in contents:
+        if "parts" in content:
+            for part in content["parts"]:
+                if "image_url" in part or "inline_data" in part:
+                    return True
+    return False
+
+
+def _build_tools(model: str, payload: Dict[str, Any]) -> List[Dict[str, Any]]:
+    """构建工具"""
+    
+    def _merge_tools(tools: List[Dict[str, Any]]) -> Dict[str, Any]:
+        record = dict()
+        for item in tools:
+            if not item or not isinstance(item, dict):
+                continue
+
+            for k, v in item.items():
+                if k == "functionDeclarations" and v and isinstance(v, list):
+                    functions = record.get("functionDeclarations", [])
+                    functions.extend(v)
+                    record["functionDeclarations"] = functions
+                else:
+                    record[k] = v
+        return record
+
+    tool = dict()
+    if payload and isinstance(payload, dict) and "tools" in payload:
+        if payload.get("tools") and isinstance(payload.get("tools"), dict):
+            payload["tools"] = [payload.get("tools")]
+        items = payload.get("tools", [])
+        if items and isinstance(items, list):
+            tool.update(_merge_tools(items))
+
+    if (
+        settings.TOOLS_CODE_EXECUTION_ENABLED
+        and not (model.endswith("-search") or "-thinking" in model)
+        and not _has_image_parts(payload.get("contents", []))
+    ):
+        tool["codeExecution"] = {}
+    if model.endswith("-search"):
+        tool["googleSearch"] = {}
+
+    # 解决 "Tool use with function calling is unsupported" 问题
+    if tool.get("functionDeclarations"):
+        tool.pop("googleSearch", None)
+        tool.pop("codeExecution", None)
+
+    return [tool] if tool else []
+
+
+def _get_safety_settings(model: str) -> List[Dict[str, str]]:
+    """获取安全设置"""
+    if model == "gemini-2.0-flash-exp":
+        return GEMINI_2_FLASH_EXP_SAFETY_SETTINGS
+    return settings.SAFETY_SETTINGS
+
+
+def _build_payload(model: str, request: GeminiRequest) -> Dict[str, Any]:
+    """构建请求payload"""
+    request_dict = request.model_dump()
+    if request.generationConfig:
+        if request.generationConfig.maxOutputTokens is None:
+            # 如果未指定最大输出长度，则不传递该字段，解决截断的问题
+            request_dict["generationConfig"].pop("maxOutputTokens")
+    
+    payload = {
+        "contents": request_dict.get("contents", []),
+        "tools": _build_tools(model, request_dict),
+        "safetySettings": _get_safety_settings(model),
+        "generationConfig": request_dict.get("generationConfig"),
+        "systemInstruction": request_dict.get("systemInstruction"),
+    }
+
+    if model.endswith("-image") or model.endswith("-image-generation"):
+        payload.pop("systemInstruction")
+        payload["generationConfig"]["responseModalities"] = ["Text", "Image"]
+        
+    if model.endswith("-non-thinking"):
+        payload["generationConfig"]["thinkingConfig"] = {"thinkingBudget": 0} 
+    if model in settings.THINKING_BUDGET_MAP:
+        payload["generationConfig"]["thinkingConfig"] = {"thinkingBudget": settings.THINKING_BUDGET_MAP.get(model,1000)}
+
+    return payload
+
+
+class GeminiChatService:
+    """聊天服务"""
+
+    def __init__(self, base_url: str, key_manager: KeyManager):
+        self.api_client = GeminiApiClient(base_url, settings.TIME_OUT)
+        self.key_manager = key_manager
+        self.response_handler = GeminiResponseHandler()
+
+    def _extract_text_from_response(self, response: Dict[str, Any]) -> str:
+        """从响应中提取文本内容"""
+        if not response.get("candidates"):
+            return ""
+
+        candidate = response["candidates"][0]
+        content = candidate.get("content", {})
+        parts = content.get("parts", [])
+
+        if parts and "text" in parts[0]:
+            return parts[0].get("text", "")
+        return ""
+
+    def _create_char_response(
+        self, original_response: Dict[str, Any], text: str
+    ) -> Dict[str, Any]:
+        """创建包含指定文本的响应"""
+        response_copy = json.loads(json.dumps(original_response))  # 深拷贝
+        if response_copy.get("candidates") and response_copy["candidates"][0].get(
+            "content", {}
+        ).get("parts"):
+            response_copy["candidates"][0]["content"]["parts"][0]["text"] = text
+        return response_copy
+
+    async def generate_content(
+        self, model: str, request: GeminiRequest, api_key: str
+    ) -> Dict[str, Any]:
+        """生成内容"""
+        payload = _build_payload(model, request)
+        start_time = time.perf_counter()
+        request_datetime = datetime.datetime.now()
+        is_success = False
+        status_code = None
+        response = None
+
+        try:
+            response = await self.api_client.generate_content(payload, model, api_key)
+            is_success = True
+            status_code = 200
+            return self.response_handler.handle_response(response, model, stream=False)
+        except Exception as e:
+            is_success = False
+            error_log_msg = str(e)
+            logger.error(f"Normal API call failed with error: {error_log_msg}")
+            match = re.search(r"status code (\d+)", error_log_msg)
+            if match:
+                status_code = int(match.group(1))
+            else:
+                status_code = 500
+
+            await add_error_log(
+                gemini_key=api_key,
+                model_name=model,
+                error_type="gemini-chat-non-stream",
+                error_log=error_log_msg,
+                error_code=status_code,
+                request_msg=payload
+            )
+            raise e
+        finally:
+            end_time = time.perf_counter()
+            latency_ms = int((end_time - start_time) * 1000)
+            await add_request_log(
+                model_name=model,
+                api_key=api_key,
+                is_success=is_success,
+                status_code=status_code,
+                latency_ms=latency_ms,
+                request_time=request_datetime
+            )
+
+    async def stream_generate_content(
+        self, model: str, request: GeminiRequest, api_key: str
+    ) -> AsyncGenerator[str, None]:
+        """流式生成内容"""
+        retries = 0
+        max_retries = settings.MAX_RETRIES
+        payload = _build_payload(model, request)
+        is_success = False
+        status_code = None
+        final_api_key = api_key
+
+        while retries < max_retries:
+            request_datetime = datetime.datetime.now()
+            start_time = time.perf_counter()
+            current_attempt_key = api_key
+            final_api_key = current_attempt_key # Update final key used
+            try:
+                async for line in self.api_client.stream_generate_content(
+                    payload, model, current_attempt_key
+                ):
+                    # print(line)
+                    if line.startswith("data:"):
+                        line = line[6:]
+                        response_data = self.response_handler.handle_response(
+                            json.loads(line), model, stream=True
+                        )
+                        text = self._extract_text_from_response(response_data)
+                        # 如果有文本内容，且开启了流式输出优化器，则使用流式输出优化器处理
+                        if text and settings.STREAM_OPTIMIZER_ENABLED:
+                            # 使用流式输出优化器处理文本输出
+                            async for (
+                                optimized_chunk
+                            ) in gemini_optimizer.optimize_stream_output(
+                                text,
+                                lambda t: self._create_char_response(response_data, t),
+                                lambda c: "data: " + json.dumps(c) + "\n\n",
+                            ):
+                                yield optimized_chunk
+                        else:
+                            # 如果没有文本内容（如工具调用等），整块输出
+                            yield "data: " + json.dumps(response_data) + "\n\n"
+                logger.info("Streaming completed successfully")
+                is_success = True
+                status_code = 200
+                break
+            except Exception as e:
+                retries += 1
+                is_success = False
+                error_log_msg = str(e)
+                logger.warning(
+                    f"Streaming API call failed with error: {error_log_msg}. Attempt {retries} of {max_retries}"
+                )
+                match = re.search(r"status code (\d+)", error_log_msg)
+                if match:
+                    status_code = int(match.group(1))
+                else:
+                    status_code = 500
+
+                await add_error_log(
+                    gemini_key=current_attempt_key,
+                    model_name=model,
+                    error_type="gemini-chat-stream",
+                    error_log=error_log_msg,
+                    error_code=status_code,
+                    request_msg=payload
+                )
+
+                api_key = await self.key_manager.handle_api_failure(current_attempt_key, retries)
+                if api_key:
+                    logger.info(f"Switched to new API key: {api_key}")
+                else:
+                    logger.error(f"No valid API key available after {retries} retries.")
+                    break
+
+                if retries >= max_retries:
+                    logger.error(
+                        f"Max retries ({max_retries}) reached for streaming."
+                    )
+                    break
+            finally:
+                end_time = time.perf_counter()
+                latency_ms = int((end_time - start_time) * 1000)
+                await add_request_log(
+                    model_name=model,
+                    api_key=final_api_key,
+                    is_success=is_success,
+                    status_code=status_code,
+                    latency_ms=latency_ms,
+                    request_time=request_datetime
+                )

app/service/config/config_service.py

@@ -55,7 +55,7 @@ class ConfigService:
             # 处理不同类型的值
             if isinstance(value, list):
                 db_value = json.dumps(value)
-            elif isinstance(value, dict):  # 新增对 dict 类型的处理
+            elif isinstance(value, dict):
                 db_value = json.dumps(value)
             elif isinstance(value, bool):
                 db_value = str(value).lower()
@@ -115,7 +115,7 @@ class ConfigService:
         # 重置并重新初始化 KeyManager
         try:
             await reset_key_manager_instance()
-            await get_key_manager_instance(settings.API_KEYS)
+            await get_key_manager_instance(settings.API_KEYS, settings.VERTEX_API_KEYS)
             logger.info("KeyManager instance re-initialized with updated settings.")
         except Exception as e:
             logger.error(f"Failed to re-initialize KeyManager: {str(e)}")
@@ -154,7 +154,7 @@ class ConfigService:
         deleted_count = 0
         not_found_keys: List[str] = []
 
-        current_api_keys = list(settings.API_KEYS)  # 创建副本以进行修改
+        current_api_keys = list(settings.API_KEYS)
         keys_actually_removed: List[str] = []
 
         for key_to_del in keys_to_delete:
@@ -166,7 +166,7 @@ class ConfigService:
                 not_found_keys.append(key_to_del)
 
         if deleted_count > 0:
-            settings.API_KEYS = current_api_keys  # 更新内存中的 settings
+            settings.API_KEYS = current_api_keys
             await ConfigService.update_config({"API_KEYS": settings.API_KEYS})
             logger.info(
                 f"成功删除 {deleted_count} 个密钥。密钥: {keys_actually_removed}"
@@ -182,9 +182,9 @@ class ConfigService:
             }
         else:
             message = "没有密钥被删除。"
-            if not_found_keys:  # 如果提供了密钥但都未找到
+            if not_found_keys:
                 message = f"所有 {len(not_found_keys)} 个指定的密钥均未找到: {not_found_keys}。"
-            elif not keys_to_delete:  # 如果 keys_to_delete 列表为空
+            elif not keys_to_delete:
                 message = "未指定要删除的密钥。"
             logger.warning(message)
             return {

app/service/error_log/error_log_service.py

@@ -161,10 +161,9 @@ async def process_delete_all_error_logs() -> int:
     返回删除的日志数量。
     """
     try:
-        # 确保数据库已连接 (如果适用，类似于 delete_old_error_logs)
+        if not database.is_connected:
-        # if not database.is_connected:
+            await database.connect()
-        #     await database.connect()
+            logger.info("Database connection established for deleting all error logs.")
-        #     logger.info("Database connection established for deleting all error logs.")
 
         deleted_count = await db_services.delete_all_error_logs()
         logger.info(

app/service/key/key_manager.py

@@ -9,12 +9,19 @@ logger = get_key_manager_logger()
 
 
 class KeyManager:
-    def __init__(self, api_keys: list):
+    def __init__(self, api_keys: list, vertex_api_keys: list):
         self.api_keys = api_keys
+        self.vertex_api_keys = vertex_api_keys
         self.key_cycle = cycle(api_keys)
+        self.vertex_key_cycle = cycle(vertex_api_keys)
         self.key_cycle_lock = asyncio.Lock()
+        self.vertex_key_cycle_lock = asyncio.Lock()
         self.failure_count_lock = asyncio.Lock()
+        self.vertex_failure_count_lock = asyncio.Lock()
         self.key_failure_counts: Dict[str, int] = {key: 0 for key in api_keys}
+        self.vertex_key_failure_counts: Dict[str, int] = {
+            key: 0 for key in vertex_api_keys
+        }
         self.MAX_FAILURES = settings.MAX_FAILURES
         self.paid_key = settings.PAID_KEY
 
@@ -26,17 +33,33 @@ class KeyManager:
         async with self.key_cycle_lock:
             return next(self.key_cycle)
 
+    async def get_next_vertex_key(self) -> str:
+        """获取下一个 Vertex API key"""
+        async with self.vertex_key_cycle_lock:
+            return next(self.vertex_key_cycle)
+
     async def is_key_valid(self, key: str) -> bool:
         """检查key是否有效"""
         async with self.failure_count_lock:
             return self.key_failure_counts[key] < self.MAX_FAILURES
 
+    async def is_vertex_key_valid(self, key: str) -> bool:
+        """检查 Vertex key 是否有效"""
+        async with self.vertex_failure_count_lock:
+            return self.vertex_key_failure_counts[key] < self.MAX_FAILURES
+
     async def reset_failure_counts(self):
         """重置所有key的失败计数"""
         async with self.failure_count_lock:
             for key in self.key_failure_counts:
                 self.key_failure_counts[key] = 0
 
+    async def reset_vertex_failure_counts(self):
+        """重置所有 Vertex key 的失败计数"""
+        async with self.vertex_failure_count_lock:
+            for key in self.vertex_key_failure_counts:
+                self.vertex_key_failure_counts[key] = 0
+
     async def reset_key_failure_count(self, key: str) -> bool:
         """重置指定key的失败计数"""
         async with self.failure_count_lock:
@@ -49,6 +72,18 @@ class KeyManager:
             )
             return False
 
+    async def reset_vertex_key_failure_count(self, key: str) -> bool:
+        """重置指定 Vertex key 的失败计数"""
+        async with self.vertex_failure_count_lock:
+            if key in self.vertex_key_failure_counts:
+                self.vertex_key_failure_counts[key] = 0
+                logger.info(f"Reset failure count for Vertex key: {key}")
+                return True
+            logger.warning(
+                f"Attempt to reset failure count for non-existent Vertex key: {key}"
+            )
+            return False
+
     async def get_next_working_key(self) -> str:
         """获取下一可用的API key"""
         initial_key = await self.get_next_key()
@@ -62,6 +97,19 @@ class KeyManager:
             if current_key == initial_key:
                 return current_key
 
+    async def get_next_working_vertex_key(self) -> str:
+        """获取下一可用的 Vertex API key"""
+        initial_key = await self.get_next_vertex_key()
+        current_key = initial_key
+
+        while True:
+            if await self.is_vertex_key_valid(current_key):
+                return current_key
+
+            current_key = await self.get_next_vertex_key()
+            if current_key == initial_key:
+                return current_key
+
     async def handle_api_failure(self, api_key: str, retries: int) -> str:
         """处理API调用失败"""
         async with self.failure_count_lock:
@@ -75,10 +123,23 @@ class KeyManager:
         else:
             return ""
 
+    async def handle_vertex_api_failure(self, api_key: str, retries: int) -> str:
+        """处理 Vertex API 调用失败"""
+        async with self.vertex_failure_count_lock:
+            self.vertex_key_failure_counts[api_key] += 1
+            if self.vertex_key_failure_counts[api_key] >= self.MAX_FAILURES:
+                logger.warning(
+                    f"Vertex API key {api_key} has failed {self.MAX_FAILURES} times"
+                )
+
     def get_fail_count(self, key: str) -> int:
         """获取指定密钥的失败次数"""
         return self.key_failure_counts.get(key, 0)
 
+    def get_vertex_fail_count(self, key: str) -> int:
+        """获取指定 Vertex 密钥的失败次数"""
+        return self.vertex_key_failure_counts.get(key, 0)
+
     async def get_keys_by_status(self) -> dict:
         """获取分类后的API key列表，包括失败次数"""
         valid_keys = {}
@@ -94,6 +155,20 @@ class KeyManager:
 
         return {"valid_keys": valid_keys, "invalid_keys": invalid_keys}
 
+    async def get_vertex_keys_by_status(self) -> dict:
+        """获取分类后的 Vertex API key 列表，包括失败次数"""
+        valid_keys = {}
+        invalid_keys = {}
+
+        async with self.vertex_failure_count_lock:
+            for key in self.vertex_api_keys:
+                fail_count = self.vertex_key_failure_counts[key]
+                if fail_count < self.MAX_FAILURES:
+                    valid_keys[key] = fail_count
+                else:
+                    invalid_keys[key] = fail_count
+        return {"valid_keys": valid_keys, "invalid_keys": invalid_keys}
+
     async def get_first_valid_key(self) -> str:
         """获取第一个有效的API key"""
         async with self.failure_count_lock:
@@ -111,19 +186,24 @@ class KeyManager:
 _singleton_instance = None
 _singleton_lock = asyncio.Lock()
 _preserved_failure_counts: Dict[str, int] | None = None
+_preserved_vertex_failure_counts: Dict[str, int] | None = None
 _preserved_old_api_keys_for_reset: list | None = None
+_preserved_vertex_old_api_keys_for_reset: list | None = None
 _preserved_next_key_in_cycle: str | None = None
+_preserved_vertex_next_key_in_cycle: str | None = None
 
 
-async def get_key_manager_instance(api_keys: list = None) -> KeyManager:
+async def get_key_manager_instance(
+    api_keys: list = None, vertex_api_keys: list = None
+) -> KeyManager:
     """
     获取 KeyManager 单例实例。
 
-    如果尚未创建实例，将使用提供的 api_keys 初始化 KeyManager。
+    如果尚未创建实例，将使用提供的 api_keys,vertex_api_keys 初始化 KeyManager。
     如果已创建实例，则忽略 api_keys 参数，返回现有单例。
     如果在重置后调用，会尝试恢复之前的状态（失败计数、循环位置）。
     """
-    global _singleton_instance, _preserved_failure_counts, _preserved_old_api_keys_for_reset, _preserved_next_key_in_cycle
+    global _singleton_instance, _preserved_failure_counts, _preserved_vertex_failure_counts, _preserved_old_api_keys_for_reset, _preserved_vertex_old_api_keys_for_reset, _preserved_next_key_in_cycle, _preserved_vertex_next_key_in_cycle
 
     async with _singleton_lock:
         if _singleton_instance is None:
@@ -131,14 +211,23 @@ async def get_key_manager_instance(api_keys: list = None) -> KeyManager:
                 raise ValueError(
                     "API keys are required to initialize or re-initialize the KeyManager instance."
                 )
+            if vertex_api_keys is None:
+                raise ValueError(
+                    "Vertex API keys are required to initialize or re-initialize the KeyManager instance."
+                )
+
             if not api_keys:
                 logger.warning(
                     "Initializing KeyManager with an empty list of API keys."
                 )
+            if not vertex_api_keys:
+                logger.warning(
+                    "Initializing KeyManager with an empty list of Vertex API keys."
+                )
 
-            _singleton_instance = KeyManager(api_keys)
+            _singleton_instance = KeyManager(api_keys, vertex_api_keys)
             logger.info(
-                f"KeyManager instance created/re-created with {len(api_keys)} API keys."
+                f"KeyManager instance created/re-created with {len(api_keys)} API keys and {len(vertex_api_keys)} Vertex API keys."
             )
 
             # 1. 恢复失败计数
@@ -153,6 +242,19 @@ async def get_key_manager_instance(api_keys: list = None) -> KeyManager:
                 logger.info("Inherited failure counts for applicable keys.")
             _preserved_failure_counts = None
 
+            if _preserved_vertex_failure_counts:
+                current_vertex_failure_counts = {
+                    key: 0 for key in _singleton_instance.vertex_api_keys
+                }
+                for key, count in _preserved_vertex_failure_counts.items():
+                    if key in current_vertex_failure_counts:
+                        current_vertex_failure_counts[key] = count
+                _singleton_instance.vertex_key_failure_counts = (
+                    current_vertex_failure_counts
+                )
+                logger.info("Inherited failure counts for applicable Vertex keys.")
+            _preserved_vertex_failure_counts = None
+
             # 2. 调整 key_cycle 的起始点
             start_key_for_new_cycle = None
             if (
@@ -201,9 +303,7 @@ async def get_key_manager_instance(api_keys: list = None) -> KeyManager:
                         f"Determined start key '{start_key_for_new_cycle}' not found in new API keys during cycle advancement. "
                         "New cycle will start from the beginning."
                     )
-                except (
+                except StopIteration:
-                    StopIteration
-                ):
                     logger.error(
                         "StopIteration while advancing key cycle, implies empty new API key list previously missed."
                     )
@@ -225,6 +325,76 @@ async def get_key_manager_instance(api_keys: list = None) -> KeyManager:
             _preserved_old_api_keys_for_reset = None
             _preserved_next_key_in_cycle = None
 
+            # 3. 调整 vertex_key_cycle 的起始点
+            start_key_for_new_vertex_cycle = None
+            if (
+                _preserved_vertex_old_api_keys_for_reset
+                and _preserved_vertex_next_key_in_cycle
+                and _singleton_instance.vertex_api_keys
+            ):
+                try:
+                    start_idx_in_old = _preserved_vertex_old_api_keys_for_reset.index(
+                        _preserved_vertex_next_key_in_cycle
+                    )
+
+                    for i in range(len(_preserved_vertex_old_api_keys_for_reset)):
+                        current_old_key_idx = (start_idx_in_old + i) % len(
+                            _preserved_vertex_old_api_keys_for_reset
+                        )
+                        key_candidate = _preserved_vertex_old_api_keys_for_reset[
+                            current_old_key_idx
+                        ]
+                        if key_candidate in _singleton_instance.vertex_api_keys:
+                            start_key_for_new_vertex_cycle = key_candidate
+                            break
+                except ValueError:
+                    logger.warning(
+                        f"Preserved next key '{_preserved_vertex_next_key_in_cycle}' not found in preserved old Vertex API keys. "
+                        "New cycle will start from the beginning of the new list."
+                    )
+                except Exception as e:
+                    logger.error(
+                        f"Error determining start key for new Vertex key cycle from preserved state: {e}. "
+                        "New cycle will start from the beginning."
+                    )
+                
+            if start_key_for_new_vertex_cycle and _singleton_instance.vertex_api_keys:
+                try:
+                    target_idx = _singleton_instance.vertex_api_keys.index(
+                        start_key_for_new_vertex_cycle
+                    )
+                    for _ in range(target_idx):
+                        next(_singleton_instance.vertex_key_cycle)
+                    logger.info(
+                        f"Vertex key cycle in new instance advanced. Next call to get_next_vertex_key() will yield: {start_key_for_new_vertex_cycle}"
+                    )
+                except ValueError:
+                    logger.warning(
+                        f"Determined start key '{start_key_for_new_vertex_cycle}' not found in new Vertex API keys during cycle advancement. "
+                        "New cycle will start from the beginning."
+                    )
+                except StopIteration:
+                    logger.error(
+                        "StopIteration while advancing Vertex key cycle, implies empty new Vertex API key list previously missed."
+                    )
+                except Exception as e:
+                    logger.error(
+                        f"Error advancing new Vertex key cycle: {e}. Cycle will start from beginning."
+                    ) 
+            else:
+                if _singleton_instance.vertex_api_keys:
+                    logger.info(
+                        "New Vertex key cycle will start from the beginning of the new Vertex API key list (no specific start key determined or needed)."
+                    )
+                else:
+                    logger.info(
+                        "New Vertex key cycle not applicable as the new Vertex API key list is empty."
+                    )
+
+            # 清理所有保存的状态
+            _preserved_vertex_old_api_keys_for_reset = None
+            _preserved_vertex_next_key_in_cycle = None
+
         return _singleton_instance
 
 
@@ -234,34 +404,52 @@ async def reset_key_manager_instance():
     将保存当前实例的状态（失败计数、旧 API keys、下一个 key 提示）
     以供下一次 get_key_manager_instance 调用时恢复。
     """
-    global _singleton_instance, _preserved_failure_counts, _preserved_old_api_keys_for_reset, _preserved_next_key_in_cycle
+    global _singleton_instance, _preserved_failure_counts, _preserved_vertex_failure_counts, _preserved_old_api_keys_for_reset, _preserved_vertex_old_api_keys_for_reset, _preserved_next_key_in_cycle, _preserved_vertex_next_key_in_cycle
     async with _singleton_lock:
         if _singleton_instance:
             # 1. 保存失败计数
             _preserved_failure_counts = _singleton_instance.key_failure_counts.copy()
+            _preserved_vertex_failure_counts = _singleton_instance.vertex_key_failure_counts.copy()
 
             # 2. 保存旧的 API keys 列表
             _preserved_old_api_keys_for_reset = _singleton_instance.api_keys.copy()
+            _preserved_vertex_old_api_keys_for_reset = _singleton_instance.vertex_api_keys.copy()
 
             # 3. 保存 key_cycle 的下一个 key 提示
             try:
                 if _singleton_instance.api_keys:
-                    _preserved_next_key_in_cycle = (
+                    _preserved_next_key_in_cycle = (    
                         await _singleton_instance.get_next_key()
                     )
                 else:
                     _preserved_next_key_in_cycle = None
-            except (
+            except StopIteration:
-                StopIteration
-            ):
                 logger.warning(
                     "Could not preserve next key hint: key cycle was empty or exhausted in old instance."
                 )
-                _preserved_next_key_in_cycle = None
+                _preserved_next_key_in_cycle = None 
             except Exception as e:
                 logger.error(f"Error preserving next key hint during reset: {e}")
                 _preserved_next_key_in_cycle = None
 
+            # 4. 保存 vertex_key_cycle 的下一个 key 提示
+            try:
+                if _singleton_instance.vertex_api_keys:
+                    _preserved_vertex_next_key_in_cycle = (
+                        await _singleton_instance.get_next_vertex_key()
+                    )
+                else:
+                    _preserved_vertex_next_key_in_cycle = None
+            except StopIteration:
+                logger.warning(
+                    "Could not preserve next key hint: Vertex key cycle was empty or exhausted in old instance."
+                )   
+                _preserved_vertex_next_key_in_cycle = None
+            except Exception as e:
+                logger.error(f"Error preserving next key hint during reset: {e}")
+                _preserved_vertex_next_key_in_cycle = None
+                
+
             _singleton_instance = None
             logger.info(
                 "KeyManager instance has been reset. State (failure counts, old keys, next key hint) preserved for next instantiation."

app/service/openai_compatiable/openai_compatiable_service.py

@@ -131,7 +131,7 @@ class OpenAICompatiableService:
                 logger.info("Streaming completed successfully")
                 is_success = True
                 status_code = 200
-                break  # 成功后退出循环
+                break
             except Exception as e:
                 retries += 1
                 is_success = False

app/static/js/config_editor.js

@@ -10,8 +10,9 @@ const SHOW_CLASS = "show"; // For modals
 const API_KEY_REGEX = /AIzaSy\S{33}/g;
 const PROXY_REGEX =
   /(?:https?|socks5):\/\/(?:[^:@\/]+(?::[^@\/]+)?@)?(?:[^:\/\s]+)(?::\d+)?/g;
+const VERTEX_API_KEY_REGEX = /AQ\.[a-zA-Z0-9_]{50}/g; // 新增 Vertex API Key 正则
 const MASKED_VALUE = "••••••••";
-
+ 
 // DOM Elements - Global Scope for frequently accessed elements
 const safetySettingsContainer = document.getElementById(
   "SAFETY_SETTINGS_container"
@@ -30,7 +31,17 @@ const bulkDeleteProxyModal = document.getElementById("bulkDeleteProxyModal");
 const bulkDeleteProxyInput = document.getElementById("bulkDeleteProxyInput");
 const resetConfirmModal = document.getElementById("resetConfirmModal");
 const configForm = document.getElementById("configForm"); // Added for frequent use
-
+ 
+// Vertex API Key Modal Elements
+const vertexApiKeyModal = document.getElementById("vertexApiKeyModal");
+const vertexApiKeyBulkInput = document.getElementById("vertexApiKeyBulkInput");
+const bulkDeleteVertexApiKeyModal = document.getElementById(
+  "bulkDeleteVertexApiKeyModal"
+);
+const bulkDeleteVertexApiKeyInput = document.getElementById(
+  "bulkDeleteVertexApiKeyInput"
+);
+ 
 // Model Helper Modal Elements
 const modelHelperModal = document.getElementById("modelHelperModal");
 const modelHelperTitleElement = document.getElementById("modelHelperTitle");
@@ -244,6 +255,8 @@ document.addEventListener("DOMContentLoaded", function () {
       bulkDeleteApiKeyModal,
       proxyModal,
       bulkDeleteProxyModal,
+      vertexApiKeyModal, // 新增
+      bulkDeleteVertexApiKeyModal, // 新增
       modelHelperModal,
     ];
     modals.forEach((modal) => {
@@ -371,7 +384,71 @@ document.addEventListener("DOMContentLoaded", function () {
   }
 
   initializeSensitiveFields(); // Initialize sensitive field handling
-
+ 
+  // Vertex API Key Modal Elements and Events
+  const addVertexApiKeyBtn = document.getElementById("addVertexApiKeyBtn");
+  const closeVertexApiKeyModalBtn = document.getElementById(
+    "closeVertexApiKeyModalBtn"
+  );
+  const cancelAddVertexApiKeyBtn = document.getElementById(
+    "cancelAddVertexApiKeyBtn"
+  );
+  const confirmAddVertexApiKeyBtn = document.getElementById(
+    "confirmAddVertexApiKeyBtn"
+  );
+  const bulkDeleteVertexApiKeyBtn = document.getElementById(
+    "bulkDeleteVertexApiKeyBtn"
+  );
+  const closeBulkDeleteVertexModalBtn = document.getElementById(
+    "closeBulkDeleteVertexModalBtn"
+  );
+  const cancelBulkDeleteVertexApiKeyBtn = document.getElementById(
+    "cancelBulkDeleteVertexApiKeyBtn"
+  );
+  const confirmBulkDeleteVertexApiKeyBtn = document.getElementById(
+    "confirmBulkDeleteVertexApiKeyBtn"
+  );
+ 
+  if (addVertexApiKeyBtn) {
+    addVertexApiKeyBtn.addEventListener("click", () => {
+      openModal(vertexApiKeyModal);
+      if (vertexApiKeyBulkInput) vertexApiKeyBulkInput.value = "";
+    });
+  }
+  if (closeVertexApiKeyModalBtn)
+    closeVertexApiKeyModalBtn.addEventListener("click", () =>
+      closeModal(vertexApiKeyModal)
+    );
+  if (cancelAddVertexApiKeyBtn)
+    cancelAddVertexApiKeyBtn.addEventListener("click", () =>
+      closeModal(vertexApiKeyModal)
+    );
+  if (confirmAddVertexApiKeyBtn)
+    confirmAddVertexApiKeyBtn.addEventListener(
+      "click",
+      handleBulkAddVertexApiKeys
+    );
+ 
+  if (bulkDeleteVertexApiKeyBtn) {
+    bulkDeleteVertexApiKeyBtn.addEventListener("click", () => {
+      openModal(bulkDeleteVertexApiKeyModal);
+      if (bulkDeleteVertexApiKeyInput) bulkDeleteVertexApiKeyInput.value = "";
+    });
+  }
+  if (closeBulkDeleteVertexModalBtn)
+    closeBulkDeleteVertexModalBtn.addEventListener("click", () =>
+      closeModal(bulkDeleteVertexApiKeyModal)
+    );
+  if (cancelBulkDeleteVertexApiKeyBtn)
+    cancelBulkDeleteVertexApiKeyBtn.addEventListener("click", () =>
+      closeModal(bulkDeleteVertexApiKeyModal)
+    );
+  if (confirmBulkDeleteVertexApiKeyBtn)
+    confirmBulkDeleteVertexApiKeyBtn.addEventListener(
+      "click",
+      handleBulkDeleteVertexApiKeys
+    );
+ 
   // Model Helper Modal Event Listeners
   if (closeModelHelperModalBtn) {
     closeModelHelperModalBtn.addEventListener("click", () =>
@@ -591,6 +668,14 @@ async function initConfig() {
     ) {
       config.FILTERED_MODELS = ["gemini-1.0-pro-latest"];
     }
+    // --- 新增：处理 VERTEX_API_KEYS 默认值 ---
+    if (!config.VERTEX_API_KEYS || !Array.isArray(config.VERTEX_API_KEYS)) {
+      config.VERTEX_API_KEYS = [];
+    }
+    // --- 新增：处理 VERTEX_EXPRESS_BASE_URL 默认值 ---
+    if (typeof config.VERTEX_EXPRESS_BASE_URL === "undefined") {
+      config.VERTEX_EXPRESS_BASE_URL = "";
+    }
     // --- 新增：处理 PROXIES 默认值 ---
     if (!config.PROXIES || !Array.isArray(config.PROXIES)) {
       config.PROXIES = []; // 默认为空数组
@@ -666,10 +751,12 @@ async function initConfig() {
       SEARCH_MODELS: ["gemini-1.5-flash-latest"],
       FILTERED_MODELS: ["gemini-1.0-pro-latest"],
       UPLOAD_PROVIDER: "smms",
-      PROXIES: [], // 添加默认值
+      PROXIES: [],
+      VERTEX_API_KEYS: [], // 确保默认值存在
+      VERTEX_EXPRESS_BASE_URL: "", // 确保默认值存在
       THINKING_MODELS: [],
       THINKING_BUDGET_MAP: {},
-      AUTO_DELETE_ERROR_LOGS_ENABLED: false, // 新增默认值
+      AUTO_DELETE_ERROR_LOGS_ENABLED: false,
       AUTO_DELETE_ERROR_LOGS_DAYS: 7, // 新增默认值
       AUTO_DELETE_REQUEST_LOGS_ENABLED: false, // 新增默认值
       AUTO_DELETE_REQUEST_LOGS_DAYS: 30, // 新增默认值
@@ -678,7 +765,7 @@ async function initConfig() {
       FAKE_STREAM_EMPTY_DATA_INTERVAL_SECONDS: 5,
       // --- 结束：处理假流式配置的默认值 ---
     };
-
+ 
     populateForm(defaultConfig);
     if (configForm) {
       // Ensure form exists
@@ -1090,7 +1177,126 @@ function handleBulkDeleteProxies() {
   }
   bulkDeleteProxyInput.value = "";
 }
-
+ 
+/**
+ * Handles the bulk addition of Vertex API keys from the modal input.
+ */
+function handleBulkAddVertexApiKeys() {
+  const vertexApiKeyContainer = document.getElementById(
+    "VERTEX_API_KEYS_container"
+  );
+  if (
+    !vertexApiKeyBulkInput ||
+    !vertexApiKeyContainer ||
+    !vertexApiKeyModal
+  ) {
+    return;
+  }
+ 
+  const bulkText = vertexApiKeyBulkInput.value;
+  const extractedKeys = bulkText.match(VERTEX_API_KEY_REGEX) || [];
+ 
+  const currentKeyInputs = vertexApiKeyContainer.querySelectorAll(
+    `.${ARRAY_INPUT_CLASS}.${SENSITIVE_INPUT_CLASS}`
+  );
+  let currentKeys = Array.from(currentKeyInputs)
+    .map((input) => {
+      return input.hasAttribute("data-real-value")
+        ? input.getAttribute("data-real-value")
+        : input.value;
+    })
+    .filter((key) => key && key.trim() !== "" && key !== MASKED_VALUE);
+ 
+  const combinedKeys = new Set([...currentKeys, ...extractedKeys]);
+  const uniqueKeys = Array.from(combinedKeys);
+ 
+  vertexApiKeyContainer.innerHTML = ""; // Clear existing items
+ 
+  uniqueKeys.forEach((key) => {
+    addArrayItemWithValue("VERTEX_API_KEYS", key); // VERTEX_API_KEYS are sensitive
+  });
+ 
+  // Ensure new sensitive inputs are masked
+  const newKeyInputs = vertexApiKeyContainer.querySelectorAll(
+    `.${ARRAY_INPUT_CLASS}.${SENSITIVE_INPUT_CLASS}`
+  );
+  newKeyInputs.forEach((input) => {
+    if (configForm && typeof initializeSensitiveFields === "function") {
+      const focusoutEvent = new Event("focusout", {
+        bubbles: true,
+        cancelable: true,
+      });
+      input.dispatchEvent(focusoutEvent);
+    }
+  });
+ 
+  closeModal(vertexApiKeyModal);
+  showNotification(
+    `添加/更新了 ${uniqueKeys.length} 个唯一 Vertex 密钥`,
+    "success"
+  );
+  vertexApiKeyBulkInput.value = "";
+}
+ 
+/**
+ * Handles the bulk deletion of Vertex API keys based on input from the modal.
+ */
+function handleBulkDeleteVertexApiKeys() {
+  const vertexApiKeyContainer = document.getElementById(
+    "VERTEX_API_KEYS_container"
+  );
+  if (
+    !bulkDeleteVertexApiKeyInput ||
+    !vertexApiKeyContainer ||
+    !bulkDeleteVertexApiKeyModal
+  ) {
+    return;
+  }
+ 
+  const bulkText = bulkDeleteVertexApiKeyInput.value;
+  if (!bulkText.trim()) {
+    showNotification("请粘贴需要删除的 Vertex API 密钥", "warning");
+    return;
+  }
+ 
+  const keysToDelete = new Set(bulkText.match(VERTEX_API_KEY_REGEX) || []);
+ 
+  if (keysToDelete.size === 0) {
+    showNotification(
+      "未在输入内容中提取到有效的 Vertex API 密钥格式",
+      "warning"
+    );
+    return;
+  }
+ 
+  const keyItems = vertexApiKeyContainer.querySelectorAll(`.${ARRAY_ITEM_CLASS}`);
+  let deleteCount = 0;
+ 
+  keyItems.forEach((item) => {
+    const input = item.querySelector(
+      `.${ARRAY_INPUT_CLASS}.${SENSITIVE_INPUT_CLASS}`
+    );
+    const realValue =
+      input &&
+      (input.hasAttribute("data-real-value")
+        ? input.getAttribute("data-real-value")
+        : input.value);
+    if (realValue && keysToDelete.has(realValue)) {
+      item.remove();
+      deleteCount++;
+    }
+  });
+ 
+  closeModal(bulkDeleteVertexApiKeyModal);
+ 
+  if (deleteCount > 0) {
+    showNotification(`成功删除了 ${deleteCount} 个匹配的 Vertex 密钥`, "success");
+  } else {
+    showNotification("列表中未找到您输入的任何 Vertex 密钥进行删除", "info");
+  }
+  bulkDeleteVertexApiKeyInput.value = "";
+}
+ 
 /**
  * Switches the active configuration tab.
  * @param {string} tabId - The ID of the tab to switch to.
@@ -1231,9 +1437,11 @@ function addArrayItemWithValue(key, value) {
 
   const isThinkingModel = key === "THINKING_MODELS";
   const isAllowedToken = key === "ALLOWED_TOKENS";
-  const isSensitive = key === "API_KEYS" || isAllowedToken;
+  const isVertexApiKey = key === "VERTEX_API_KEYS"; // 新增判断
+  const isSensitive =
+    key === "API_KEYS" || isAllowedToken || isVertexApiKey; // 更新敏感判断
   const modelId = isThinkingModel ? generateUUID() : null;
-
+ 
   const arrayItem = document.createElement("div");
   arrayItem.className = `${ARRAY_ITEM_CLASS} flex items-center mb-2 gap-2`;
   if (isThinkingModel) {

app/templates/config_editor.html

@@ -456,7 +456,51 @@ endblock %} {% block head_extra_styles %}
           />
           <small class="text-gray-500 mt-1 block">Gemini API的基础URL</small>
         </div>
-
+ 
+        <!-- Vertex API密钥列表 -->
+        <div class="mb-6">
+          <label for="VERTEX_API_KEYS" class="block font-semibold mb-2 text-gray-700"
+            >Vertex API密钥列表</label
+          >
+          <div class="array-container" id="VERTEX_API_KEYS_container">
+            <!-- 数组项将在这里动态添加 -->
+          </div>
+          <div class="flex justify-end gap-2">
+            <button
+              type="button"
+              class="bg-red-600 hover:bg-red-700 text-white px-4 py-2 rounded-lg font-medium transition-all duration-200 flex items-center gap-2"
+              id="bulkDeleteVertexApiKeyBtn"
+            >
+              <i class="fas fa-trash-alt"></i> 删除Vertex密钥
+            </button>
+            <button
+              type="button"
+              class="bg-violet-600 hover:bg-violet-700 text-white px-4 py-2 rounded-lg font-medium transition-all duration-200 flex items-center gap-2"
+              id="addVertexApiKeyBtn"
+            >
+              <i class="fas fa-plus"></i> 添加Vertex密钥
+            </button>
+          </div>
+          <small class="text-gray-500 mt-1 block"
+            >Vertex AI Platform API密钥列表。点击按钮可批量添加或删除。</small
+          >
+        </div>
+ 
+        <!-- Vertex Express API基础URL -->
+        <div class="mb-6">
+          <label for="VERTEX_EXPRESS_BASE_URL" class="block font-semibold mb-2 text-gray-700"
+            >Vertex Express API基础URL</label
+          >
+          <input
+            type="text"
+            id="VERTEX_EXPRESS_BASE_URL"
+            name="VERTEX_EXPRESS_BASE_URL"
+            placeholder="https://aiplatform.googleapis.com/v1beta1/publishers/google/models"
+            class="w-full px-4 py-3 rounded-lg border border-gray-300 focus:border-primary-500 focus:ring focus:ring-primary-200 focus:ring-opacity-50 form-input-themed"
+          />
+          <small class="text-gray-500 mt-1 block">Vertex Express API的基础URL</small>
+        </div>
+ 
         <!-- 最大失败次数 -->
         <div class="mb-6">
           <label
@@ -1610,7 +1654,105 @@ endblock %} {% block head_extra_styles %}
     </div>
   </div>
 </div>
-
+ 
+<!-- Vertex API Key Add Modal -->
+<div id="vertexApiKeyModal" class="modal">
+  <div
+    class="w-full max-w-lg mx-auto rounded-2xl shadow-2xl overflow-hidden animate-fade-in"
+    style="
+      background-color: rgba(70, 50, 150, 0.95);
+      color: #ffffff;
+      border: 1px solid rgba(120, 100, 200, 0.4);
+    "
+  >
+    <div class="p-6">
+      <div class="flex justify-between items-center mb-4">
+        <h2 class="text-xl font-bold text-gray-100">批量添加 Vertex API 密钥</h2>
+        <button
+          id="closeVertexApiKeyModalBtn"
+          class="text-gray-300 hover:text-gray-100 text-xl"
+        >
+          &times;
+        </button>
+      </div>
+      <p class="text-gray-300 mb-4">
+        每行粘贴一个或多个密钥，将自动提取有效密钥 (格式: AQ.开头，共53位) 并去重。
+      </p>
+      <textarea
+        id="vertexApiKeyBulkInput"
+        rows="10"
+        placeholder="在此处粘贴 Vertex API 密钥..."
+        class="w-full px-4 py-3 rounded-lg border font-mono text-sm form-input-themed"
+      ></textarea>
+      <div class="flex justify-end gap-3 mt-6">
+        <button
+          type="button"
+          id="confirmAddVertexApiKeyBtn"
+          class="bg-violet-600 hover:bg-violet-700 text-white px-6 py-2 rounded-lg font-medium transition"
+        >
+          确认添加
+        </button>
+        <button
+          type="button"
+          id="cancelAddVertexApiKeyBtn"
+          class="bg-gray-500 bg-opacity-50 hover:bg-opacity-70 text-gray-200 px-6 py-2 rounded-lg font-medium transition"
+        >
+          取消
+        </button>
+      </div>
+    </div>
+  </div>
+</div>
+ 
+<!-- Bulk Delete Vertex API Key Modal -->
+<div id="bulkDeleteVertexApiKeyModal" class="modal">
+  <div
+    class="w-full max-w-lg mx-auto rounded-2xl shadow-2xl overflow-hidden animate-fade-in"
+    style="
+      background-color: rgba(70, 50, 150, 0.95);
+      color: #ffffff;
+      border: 1px solid rgba(120, 100, 200, 0.4);
+    "
+  >
+    <div class="p-6">
+      <div class="flex justify-between items-center mb-4">
+        <h2 class="text-xl font-bold text-gray-100">批量删除 Vertex API 密钥</h2>
+        <button
+          id="closeBulkDeleteVertexModalBtn"
+          class="text-gray-300 hover:text-gray-100 text-xl"
+        >
+          &times;
+        </button>
+      </div>
+      <p class="text-gray-300 mb-4">
+        每行粘贴一个或多个密钥，将自动提取有效密钥并从列表中删除。
+      </p>
+      <textarea
+        id="bulkDeleteVertexApiKeyInput"
+        rows="10"
+        placeholder="在此处粘贴要删除的 Vertex API 密钥..."
+        class="w-full px-4 py-3 rounded-lg border font-mono text-sm form-input-themed focus:border-red-500 focus:ring-red-500"
+      ></textarea>
+      <div class="flex justify-end gap-3 mt-6">
+        <button
+          type="button"
+          id="confirmBulkDeleteVertexApiKeyBtn"
+          class="bg-red-600 hover:bg-red-700 text-white px-6 py-2 rounded-lg font-medium transition"
+        >
+          确认删除
+        </button>
+        <button
+          type="button"
+          id="cancelBulkDeleteVertexApiKeyBtn"
+          class="bg-gray-500 bg-opacity-50 hover:bg-opacity-70 text-gray-200 px-6 py-2 rounded-lg font-medium transition"
+        >
+          取消
+        </button>
+      </div>
+    </div>
+  </div>
+</div>
+ 
 <!-- Model Helper Modal -->
 <div id="modelHelperModal" class="modal">
   <div