videos1.0/AIProviderConfig.md

# AIProviderConfig.md - AI 厂商动态配置架构设计

| 文档类型 | **Technical Design (技术设计文档)** |
| --- | --- |
| **项目名称** | SmartAudit (AI 营销内容合规审核平台) |
| **版本号** | V1.0 |
| **日期** | 2026-02-02 |
| **侧重** | AI 厂商动态配置、多租户隔离、运行时热更新 |

---

## 版本历史 (Version History)

| 版本 | 日期 | 作者 | 变更说明 |
| --- | --- | --- | --- |
| V1.0 | 2026-02-02 | Claude | 初稿：AI 厂商动态配置架构设计 |

---

## 1. 设计背景与目标

### 1.1 问题陈述

传统方案将 AI 模型的 API Key 和 Base URL 写死在环境变量中，存在以下问题：

1. **灵活性差：** 切换 AI 厂商需要修改环境变量并重启服务
2. **多租户困难：** 无法支持不同品牌方使用不同的 AI 厂商
3. **安全隐患：** 环境变量容易泄露，难以细粒度管理
4. **运维成本高：** 密钥轮换需要重新部署

### 1.2 设计目标

实现**商业 SaaS 级别的 AI 厂商动态配置系统**：

| 目标 | 描述 |
| --- | --- |
| **动态配置** | 管理员在后台配置 AI 厂商，无需修改代码或重启服务 |
| **多厂商支持** | 支持 DeepSeek、OpenAI、阿里云、OneAPI 中转等多种厂商 |
| **多租户隔离** | 不同品牌方可配置独立的 AI 厂商和配额 |
| **热更新** | 配置变更即时生效，无需重启服务 |
| **安全存储** | API Key 加密存储，支持密钥轮换 |
| **故障转移** | 主厂商不可用时自动切换到备用厂商 |

---

## 2. 系统架构

### 2.1 架构概览

```
┌─────────────────────────────────────────────────────────────────────────┐
│                          管理后台 (Admin Portal)                          │
│   ┌──────────────────────────────────────────────────────────────────┐  │
│   │  AI 厂商配置页面：添加/编辑/删除/测试连通性                          │  │
│   └──────────────────────────────────────────────────────────────────┘  │
└─────────────────────────────────────────────────────────────────────────┘
                                    │
                                    ▼
┌─────────────────────────────────────────────────────────────────────────┐
│                           API 层 (FastAPI)                               │
│   ┌──────────────────────────────────────────────────────────────────┐  │
│   │  POST /admin/ai-providers       - 创建 AI 厂商配置                  │  │
│   │  GET  /admin/ai-providers       - 获取厂商列表                      │  │
│   │  PUT  /admin/ai-providers/{id}  - 更新配置                         │  │
│   │  POST /admin/ai-providers/{id}/test - 测试连通性                   │  │
│   └──────────────────────────────────────────────────────────────────┘  │
└─────────────────────────────────────────────────────────────────────────┘
                                    │
                                    ▼
┌─────────────────────────────────────────────────────────────────────────┐
│                       AI 客户端工厂 (AIClientFactory)                     │
│   ┌──────────────────────────────────────────────────────────────────┐  │
│   │  • 根据配置动态创建 AI 客户端实例                                    │  │
│   │  • 支持连接池和客户端复用                                           │  │
│   │  • 配置变更时自动刷新客户端                                         │  │
│   └──────────────────────────────────────────────────────────────────┘  │
└─────────────────────────────────────────────────────────────────────────┘
                                    │
                    ┌───────────────┼───────────────┐
                    ▼               ▼               ▼
            ┌─────────────┐ ┌─────────────┐ ┌─────────────┐
            │  DeepSeek   │ │   OpenAI    │ │  OneAPI     │
            │   Client    │ │   Client    │ │  (中转)     │
            └─────────────┘ └─────────────┘ └─────────────┘
```

### 2.2 核心组件

| 组件 | 职责 |
| --- | --- |
| **AIProviderConfig** | 数据模型，存储厂商配置 |
| **AIClientFactory** | 工厂类，根据配置创建客户端 |
| **AIClientRegistry** | 注册表，缓存和管理客户端实例 |
| **ConfigWatcher** | 监听配置变更，触发客户端刷新 |
| **SecretsManager** | 加密存储和解密 API Key |

---

## 3. 数据模型设计

### 3.1 AI 厂商配置表 (ai_provider_configs)

```sql
CREATE TABLE ai_provider_configs (
    id              UUID PRIMARY KEY DEFAULT gen_random_uuid(),

    -- 基础信息
    name            VARCHAR(100) NOT NULL,          -- 配置名称，如 "生产环境 DeepSeek"
    provider_type   VARCHAR(50) NOT NULL,           -- 厂商类型：deepseek/openai/oneapi/aliyun/...
    description     TEXT,                           -- 配置说明

    -- 连接配置
    base_url        VARCHAR(500) NOT NULL,          -- API Base URL
    api_key_encrypted BYTEA NOT NULL,               -- 加密后的 API Key

    -- 模型配置
    default_model   VARCHAR(100),                   -- 默认模型，如 "deepseek-chat"
    available_models JSONB DEFAULT '[]',            -- 可用模型列表

    -- 能力标签
    capabilities    JSONB DEFAULT '[]',             -- 支持的能力：["chat", "vision", "embedding"]

    -- 使用场景
    use_cases       JSONB DEFAULT '[]',             -- 适用场景：["brief_parsing", "script_review", "video_audit"]

    -- 租户隔离
    tenant_id       UUID,                           -- 所属租户（品牌方），NULL 表示全局配置

    -- 优先级与状态
    priority        INT DEFAULT 100,                -- 优先级，数字越小优先级越高
    is_enabled      BOOLEAN DEFAULT true,           -- 是否启用
    is_default      BOOLEAN DEFAULT false,          -- 是否为默认配置

    -- 限流配置
    rate_limit_rpm  INT DEFAULT 60,                 -- 每分钟请求限制
    rate_limit_tpm  INT DEFAULT 100000,             -- 每分钟 Token 限制

    -- 故障转移
    fallback_provider_id UUID,                      -- 备用厂商配置 ID

    -- 扩展配置
    extra_config    JSONB DEFAULT '{}',             -- 厂商特定配置

    -- 元数据
    created_at      TIMESTAMPTZ DEFAULT NOW(),
    updated_at      TIMESTAMPTZ DEFAULT NOW(),
    created_by      UUID,

    -- 约束
    CONSTRAINT uk_tenant_default UNIQUE (tenant_id, is_default)
        WHERE is_default = true
);

-- 索引
CREATE INDEX idx_provider_tenant ON ai_provider_configs(tenant_id);
CREATE INDEX idx_provider_type ON ai_provider_configs(provider_type);
CREATE INDEX idx_provider_enabled ON ai_provider_configs(is_enabled);
CREATE INDEX idx_provider_use_cases ON ai_provider_configs USING GIN(use_cases);
```

### 3.2 厂商类型枚举

```python
from enum import Enum

class AIProviderType(str, Enum):
    """支持的 AI 厂商类型"""

    # 国内厂商
    DEEPSEEK = "deepseek"           # DeepSeek
    QWEN = "qwen"                   # 阿里云通义千问
    DOUBAO = "doubao"               # 字节豆包
    ZHIPU = "zhipu"                 # 智谱 GLM
    BAICHUAN = "baichuan"           # 百川
    MOONSHOT = "moonshot"           # Moonshot (Kimi)

    # 海外厂商（需注意合规）
    OPENAI = "openai"               # OpenAI
    ANTHROPIC = "anthropic"         # Anthropic Claude

    # 中转服务
    ONEAPI = "oneapi"               # OneAPI 中转
    OPENROUTER = "openrouter"       # OpenRouter

    # 本地部署
    OLLAMA = "ollama"               # Ollama 本地
    VLLM = "vllm"                   # vLLM 部署

    # ASR/OCR 专用
    ALIYUN_ASR = "aliyun_asr"       # 阿里云 ASR
    ALIYUN_OCR = "aliyun_ocr"       # 阿里云 OCR
    PADDLEOCR = "paddleocr"         # PaddleOCR 本地
    WHISPER = "whisper"             # OpenAI Whisper


class AICapability(str, Enum):
    """AI 能力标签"""
    CHAT = "chat"                   # 对话/文本生成
    VISION = "vision"              # 图像理解
    EMBEDDING = "embedding"        # 向量嵌入
    ASR = "asr"                    # 语音识别
    OCR = "ocr"                    # 文字识别
    TTS = "tts"                    # 语音合成


class AIUseCase(str, Enum):
    """AI 使用场景"""
    BRIEF_PARSING = "brief_parsing"           # Brief 解析
    SCRIPT_REVIEW = "script_review"           # 脚本预审
    VIDEO_AUDIT = "video_audit"               # 视频审核
    CONTEXT_CLASSIFICATION = "context_classification"  # 语境分类
    SENTIMENT_ANALYSIS = "sentiment_analysis" # 情感分析
    LOGO_DETECTION = "logo_detection"         # Logo 检测
    ASR_TRANSCRIPTION = "asr_transcription"   # 语音转写
    OCR_EXTRACTION = "ocr_extraction"         # 文字提取
```

### 3.3 使用日志表 (ai_usage_logs)

```sql
CREATE TABLE ai_usage_logs (
    id              UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    provider_id     UUID NOT NULL REFERENCES ai_provider_configs(id),
    tenant_id       UUID,

    -- 请求信息
    use_case        VARCHAR(50) NOT NULL,
    model           VARCHAR(100),

    -- 用量统计
    prompt_tokens   INT DEFAULT 0,
    completion_tokens INT DEFAULT 0,
    total_tokens    INT DEFAULT 0,

    -- 性能指标
    latency_ms      INT,
    status          VARCHAR(20),                -- success/error/timeout
    error_message   TEXT,

    -- 时间
    created_at      TIMESTAMPTZ DEFAULT NOW(),

    -- 分区键
    created_date    DATE DEFAULT CURRENT_DATE
) PARTITION BY RANGE (created_date);

-- 按月分区
CREATE TABLE ai_usage_logs_2026_02 PARTITION OF ai_usage_logs
    FOR VALUES FROM ('2026-02-01') TO ('2026-03-01');
```

---

## 4. 核心代码设计

### 4.1 配置模型 (Pydantic)

```python
# app/models/ai_provider.py

from pydantic import BaseModel, Field, SecretStr
from typing import Optional, List
from uuid import UUID
from datetime import datetime
from enum import Enum


class AIProviderCreate(BaseModel):
    """创建 AI 厂商配置请求"""
    name: str = Field(..., max_length=100)
    provider_type: AIProviderType
    description: Optional[str] = None
    base_url: str
    api_key: SecretStr  # 接收时为明文，存储时加密
    default_model: Optional[str] = None
    available_models: List[str] = []
    capabilities: List[AICapability] = []
    use_cases: List[AIUseCase] = []
    tenant_id: Optional[UUID] = None
    priority: int = 100
    is_enabled: bool = True
    is_default: bool = False
    rate_limit_rpm: int = 60
    rate_limit_tpm: int = 100000
    fallback_provider_id: Optional[UUID] = None
    extra_config: dict = {}


class AIProviderResponse(BaseModel):
    """AI 厂商配置响应"""
    id: UUID
    name: str
    provider_type: AIProviderType
    description: Optional[str]
    base_url: str
    # 注意：不返回 api_key
    default_model: Optional[str]
    available_models: List[str]
    capabilities: List[AICapability]
    use_cases: List[AIUseCase]
    tenant_id: Optional[UUID]
    priority: int
    is_enabled: bool
    is_default: bool
    rate_limit_rpm: int
    rate_limit_tpm: int
    fallback_provider_id: Optional[UUID]
    extra_config: dict
    created_at: datetime
    updated_at: datetime
```

### 4.2 AI 客户端工厂

```python
# app/services/ai/client_factory.py

from abc import ABC, abstractmethod
from typing import Dict, Optional, Type
from functools import lru_cache
import asyncio
from openai import AsyncOpenAI

from app.models.ai_provider import AIProviderType
from app.services.secrets_manager import SecretsManager


class BaseAIClient(ABC):
    """AI 客户端基类"""

    def __init__(self, config: dict):
        self.config = config
        self.base_url = config["base_url"]
        self.api_key = config["api_key"]
        self.default_model = config.get("default_model")

    @abstractmethod
    async def chat(self, messages: list, model: str = None, **kwargs) -> dict:
        """对话接口"""
        pass

    @abstractmethod
    async def health_check(self) -> bool:
        """健康检查"""
        pass


class OpenAICompatibleClient(BaseAIClient):
    """OpenAI 兼容客户端 (适用于 DeepSeek, OneAPI, Moonshot 等)"""

    def __init__(self, config: dict):
        super().__init__(config)
        self.client = AsyncOpenAI(
            api_key=self.api_key,
            base_url=self.base_url,
        )

    async def chat(self, messages: list, model: str = None, **kwargs) -> dict:
        model = model or self.default_model
        response = await self.client.chat.completions.create(
            model=model,
            messages=messages,
            **kwargs
        )
        return {
            "content": response.choices[0].message.content,
            "usage": {
                "prompt_tokens": response.usage.prompt_tokens,
                "completion_tokens": response.usage.completion_tokens,
                "total_tokens": response.usage.total_tokens,
            },
            "model": response.model,
        }

    async def health_check(self) -> bool:
        try:
            await self.client.models.list()
            return True
        except Exception:
            return False


class AIClientFactory:
    """AI 客户端工厂"""

    # 厂商类型到客户端类的映射
    _client_classes: Dict[AIProviderType, Type[BaseAIClient]] = {
        AIProviderType.DEEPSEEK: OpenAICompatibleClient,
        AIProviderType.OPENAI: OpenAICompatibleClient,
        AIProviderType.ONEAPI: OpenAICompatibleClient,
        AIProviderType.QWEN: OpenAICompatibleClient,
        AIProviderType.MOONSHOT: OpenAICompatibleClient,
        AIProviderType.ZHIPU: OpenAICompatibleClient,
        # 可扩展更多厂商...
    }

    def __init__(self, secrets_manager: SecretsManager):
        self.secrets_manager = secrets_manager
        self._client_cache: Dict[str, BaseAIClient] = {}
        self._cache_lock = asyncio.Lock()

    async def get_client(self, provider_config: dict) -> BaseAIClient:
        """获取或创建 AI 客户端"""
        cache_key = f"{provider_config['id']}:{provider_config['updated_at']}"

        if cache_key in self._client_cache:
            return self._client_cache[cache_key]

        async with self._cache_lock:
            # 双重检查
            if cache_key in self._client_cache:
                return self._client_cache[cache_key]

            # 解密 API Key
            api_key = await self.secrets_manager.decrypt(
                provider_config["api_key_encrypted"]
            )

            config = {
                **provider_config,
                "api_key": api_key,
            }

            # 创建客户端
            provider_type = AIProviderType(provider_config["provider_type"])
            client_class = self._client_classes.get(provider_type)

            if not client_class:
                raise ValueError(f"Unsupported provider type: {provider_type}")

            client = client_class(config)

            # 缓存客户端
            self._client_cache[cache_key] = client

            # 清理旧缓存
            self._cleanup_old_cache(provider_config['id'])

            return client

    def _cleanup_old_cache(self, provider_id: str):
        """清理同一 provider 的旧缓存"""
        keys_to_remove = [
            k for k in self._client_cache.keys()
            if k.startswith(f"{provider_id}:")
        ]
        # 保留最新的一个
        for key in keys_to_remove[:-1]:
            del self._client_cache[key]

    def invalidate_cache(self, provider_id: str = None):
        """使缓存失效"""
        if provider_id:
            keys_to_remove = [
                k for k in self._client_cache.keys()
                if k.startswith(f"{provider_id}:")
            ]
            for key in keys_to_remove:
                del self._client_cache[key]
        else:
            self._client_cache.clear()
```

### 4.3 AI 服务路由器

```python
# app/services/ai/router.py

from typing import Optional, List
from uuid import UUID

from app.models.ai_provider import AIUseCase, AICapability
from app.repositories.ai_provider_repo import AIProviderRepository
from app.services.ai.client_factory import AIClientFactory, BaseAIClient


class AIServiceRouter:
    """AI 服务路由器 - 根据场景选择合适的 AI 厂商"""

    def __init__(
        self,
        provider_repo: AIProviderRepository,
        client_factory: AIClientFactory,
    ):
        self.provider_repo = provider_repo
        self.client_factory = client_factory

    async def get_client_for_use_case(
        self,
        use_case: AIUseCase,
        tenant_id: Optional[UUID] = None,
        required_capabilities: List[AICapability] = None,
    ) -> BaseAIClient:
        """
        根据使用场景获取合适的 AI 客户端

        优先级：
        1. 租户专属配置 (tenant_id 匹配)
        2. 全局默认配置 (tenant_id = NULL)
        3. 按 priority 排序
        """
        # 查询符合条件的配置
        configs = await self.provider_repo.find_by_use_case(
            use_case=use_case,
            tenant_id=tenant_id,
            capabilities=required_capabilities,
            enabled_only=True,
        )

        if not configs:
            raise ValueError(
                f"No AI provider configured for use case: {use_case}"
            )

        # 选择优先级最高的配置
        selected_config = configs[0]

        # 创建并返回客户端
        client = await self.client_factory.get_client(selected_config)

        # 健康检查，失败则尝试备用
        if not await client.health_check():
            if selected_config.get("fallback_provider_id"):
                fallback_config = await self.provider_repo.get_by_id(
                    selected_config["fallback_provider_id"]
                )
                if fallback_config:
                    client = await self.client_factory.get_client(fallback_config)

        return client

    async def chat(
        self,
        messages: list,
        use_case: AIUseCase,
        tenant_id: Optional[UUID] = None,
        model: str = None,
        **kwargs
    ) -> dict:
        """统一的对话接口"""
        client = await self.get_client_for_use_case(
            use_case=use_case,
            tenant_id=tenant_id,
            required_capabilities=[AICapability.CHAT],
        )

        return await client.chat(messages, model=model, **kwargs)
```

### 4.4 管理后台 API

```python
# app/api/v1/endpoints/admin/ai_providers.py

from fastapi import APIRouter, Depends, HTTPException, status
from typing import List, Optional
from uuid import UUID

from app.models.ai_provider import (
    AIProviderCreate,
    AIProviderUpdate,
    AIProviderResponse,
)
from app.services.ai_provider_service import AIProviderService
from app.api.deps import get_current_admin_user

router = APIRouter()


@router.post("", response_model=AIProviderResponse, status_code=status.HTTP_201_CREATED)
async def create_ai_provider(
    request: AIProviderCreate,
    service: AIProviderService = Depends(),
    current_user = Depends(get_current_admin_user),
):
    """创建 AI 厂商配置（仅管理员）"""
    return await service.create(request, created_by=current_user.id)


@router.get("", response_model=List[AIProviderResponse])
async def list_ai_providers(
    tenant_id: Optional[UUID] = None,
    provider_type: Optional[str] = None,
    service: AIProviderService = Depends(),
    current_user = Depends(get_current_admin_user),
):
    """获取 AI 厂商配置列表"""
    return await service.list(tenant_id=tenant_id, provider_type=provider_type)


@router.get("/{provider_id}", response_model=AIProviderResponse)
async def get_ai_provider(
    provider_id: UUID,
    service: AIProviderService = Depends(),
    current_user = Depends(get_current_admin_user),
):
    """获取单个 AI 厂商配置"""
    provider = await service.get_by_id(provider_id)
    if not provider:
        raise HTTPException(status_code=404, detail="Provider not found")
    return provider


@router.put("/{provider_id}", response_model=AIProviderResponse)
async def update_ai_provider(
    provider_id: UUID,
    request: AIProviderUpdate,
    service: AIProviderService = Depends(),
    current_user = Depends(get_current_admin_user),
):
    """更新 AI 厂商配置"""
    provider = await service.update(provider_id, request)
    if not provider:
        raise HTTPException(status_code=404, detail="Provider not found")
    return provider


@router.delete("/{provider_id}", status_code=status.HTTP_204_NO_CONTENT)
async def delete_ai_provider(
    provider_id: UUID,
    service: AIProviderService = Depends(),
    current_user = Depends(get_current_admin_user),
):
    """删除 AI 厂商配置"""
    success = await service.delete(provider_id)
    if not success:
        raise HTTPException(status_code=404, detail="Provider not found")


@router.post("/{provider_id}/test")
async def test_ai_provider(
    provider_id: UUID,
    service: AIProviderService = Depends(),
    current_user = Depends(get_current_admin_user),
):
    """测试 AI 厂商连通性"""
    result = await service.test_connection(provider_id)
    return {
        "success": result.success,
        "latency_ms": result.latency_ms,
        "error": result.error,
    }


@router.post("/{provider_id}/rotate-key", response_model=AIProviderResponse)
async def rotate_api_key(
    provider_id: UUID,
    new_api_key: str,
    service: AIProviderService = Depends(),
    current_user = Depends(get_current_admin_user),
):
    """轮换 API Key"""
    return await service.rotate_api_key(provider_id, new_api_key)
```

---

## 5. 安全设计

### 5.1 API Key 加密存储

```python
# app/services/secrets_manager.py

from cryptography.fernet import Fernet
from cryptography.hazmat.primitives import hashes
from cryptography.hazmat.primitives.kdf.pbkdf2 import PBKDF2HMAC
import base64
import os


class SecretsManager:
    """密钥管理器 - 负责加密/解密敏感信息"""

    def __init__(self, master_key: str):
        """
        初始化密钥管理器

        Args:
            master_key: 主密钥，从安全存储（如 Vault、KMS）获取
        """
        # 从主密钥派生加密密钥
        salt = os.environ.get("ENCRYPTION_SALT", "smartaudit").encode()
        kdf = PBKDF2HMAC(
            algorithm=hashes.SHA256(),
            length=32,
            salt=salt,
            iterations=100000,
        )
        key = base64.urlsafe_b64encode(kdf.derive(master_key.encode()))
        self.fernet = Fernet(key)

    async def encrypt(self, plaintext: str) -> bytes:
        """加密明文"""
        return self.fernet.encrypt(plaintext.encode())

    async def decrypt(self, ciphertext: bytes) -> str:
        """解密密文"""
        return self.fernet.decrypt(ciphertext).decode()
```

### 5.2 权限控制

| 操作 | 系统管理员 | 品牌方管理员 | 代理商 | 达人 |
| --- | --- | --- | --- | --- |
| 创建全局配置 | ✅ | ❌ | ❌ | ❌ |
| 创建租户配置 | ✅ | ✅ (仅自己租户) | ❌ | ❌ |
| 查看配置列表 | ✅ (全部) | ✅ (仅自己租户) | ❌ | ❌ |
| 修改配置 | ✅ | ✅ (仅自己租户) | ❌ | ❌ |
| 删除配置 | ✅ | ✅ (仅自己租户) | ❌ | ❌ |
| 查看 API Key | ❌ | ❌ | ❌ | ❌ |
| 轮换 API Key | ✅ | ✅ (仅自己租户) | ❌ | ❌ |

---

## 6. 配置热更新

### 6.1 更新机制

```python
# app/services/ai/config_watcher.py

import asyncio
from datetime import datetime
from typing import Callable, List

from app.repositories.ai_provider_repo import AIProviderRepository
from app.services.ai.client_factory import AIClientFactory


class ConfigWatcher:
    """配置变更监听器"""

    def __init__(
        self,
        provider_repo: AIProviderRepository,
        client_factory: AIClientFactory,
        poll_interval: int = 30,  # 秒
    ):
        self.provider_repo = provider_repo
        self.client_factory = client_factory
        self.poll_interval = poll_interval
        self._last_check = datetime.min
        self._running = False
        self._callbacks: List[Callable] = []

    def on_config_change(self, callback: Callable):
        """注册配置变更回调"""
        self._callbacks.append(callback)

    async def start(self):
        """启动监听"""
        self._running = True
        while self._running:
            await self._check_for_changes()
            await asyncio.sleep(self.poll_interval)

    async def stop(self):
        """停止监听"""
        self._running = False

    async def _check_for_changes(self):
        """检查配置变更"""
        changed_configs = await self.provider_repo.find_updated_since(
            self._last_check
        )

        if changed_configs:
            self._last_check = datetime.utcnow()

            # 使相关缓存失效
            for config in changed_configs:
                self.client_factory.invalidate_cache(config["id"])

            # 触发回调
            for callback in self._callbacks:
                await callback(changed_configs)
```

### 6.2 应用启动集成

```python
# app/main.py

from contextlib import asynccontextmanager
from fastapi import FastAPI

from app.services.ai.config_watcher import ConfigWatcher


@asynccontextmanager
async def lifespan(app: FastAPI):
    # 启动时
    config_watcher = ConfigWatcher(
        provider_repo=app.state.provider_repo,
        client_factory=app.state.client_factory,
    )
    asyncio.create_task(config_watcher.start())

    yield

    # 关闭时
    await config_watcher.stop()


app = FastAPI(lifespan=lifespan)
```

---

## 7. 使用示例

### 7.1 在业务代码中使用

```python
# app/services/brief_parser.py

from app.services.ai.router import AIServiceRouter
from app.models.ai_provider import AIUseCase


class BriefParserService:
    """Brief 解析服务"""

    def __init__(self, ai_router: AIServiceRouter):
        self.ai_router = ai_router

    async def parse_brief(self, content: str, tenant_id: UUID = None) -> dict:
        """解析 Brief 文档"""

        messages = [
            {"role": "system", "content": "你是一个专业的 Brief 解析助手..."},
            {"role": "user", "content": f"请解析以下 Brief 内容：\n{content}"},
        ]

        # 自动选择合适的 AI 厂商
        result = await self.ai_router.chat(
            messages=messages,
            use_case=AIUseCase.BRIEF_PARSING,
            tenant_id=tenant_id,
        )

        return self._parse_response(result["content"])
```

### 7.2 管理员配置流程

```
1. 管理员登录后台
2. 进入「系统设置 → AI 厂商管理」
3. 点击「添加厂商」
4. 填写配置：
   - 名称：生产环境 DeepSeek
   - 厂商类型：DeepSeek
   - Base URL：https://api.deepseek.com/v1
   - API Key：sk-xxx
   - 默认模型：deepseek-chat
   - 适用场景：Brief 解析、脚本预审
   - 优先级：10
5. 点击「测试连通性」
6. 保存配置
7. 配置立即生效，无需重启服务
```

---

## 8. 监控与告警

### 8.1 监控指标

| 指标 | 说明 | 告警阈值 |
| --- | --- | --- |
| `ai_request_total` | AI 请求总数 | - |
| `ai_request_latency_p99` | P99 延迟 | > 10s |
| `ai_request_error_rate` | 错误率 | > 5% |
| `ai_token_usage_total` | Token 使用量 | 接近配额 80% |
| `ai_provider_health` | 厂商健康状态 | 连续失败 > 3 次 |

### 8.2 告警规则

```yaml
# prometheus/alerts/ai_provider.yml
groups:
  - name: ai_provider
    rules:
      - alert: AIProviderHighErrorRate
        expr: rate(ai_request_errors_total[5m]) / rate(ai_request_total[5m]) > 0.05
        for: 2m
        labels:
          severity: warning
        annotations:
          summary: "AI 厂商 {{ $labels.provider }} 错误率过高"

      - alert: AIProviderDown
        expr: ai_provider_health == 0
        for: 1m
        labels:
          severity: critical
        annotations:
          summary: "AI 厂商 {{ $labels.provider }} 不可用"
```

---

## 9. 相关文档

| 文档 | 说明 |
| --- | --- |
| DevelopmentPlan.md | 开发计划（已更新 AI 配置章节） |
| RequirementsDoc.md | 需求文档 |
| FeatureSummary.md | 功能清单 |
| API 接口规范 | 待编写 |