AIRouter

🌐 Language / 语言

🇨🇳 中文版 | 🇺🇸 English Version

</div>

🇨🇳 中文版

一个智能的AI路由器，为大型语言模型(LLM)提供高性能的统一API接口，支持多源负载均衡和智能故障转移。

✨ 主要特性

🔄 统一API接口: 通过统一的接口访问多种LLM提供商（OpenRouter、DeepInfra、DeerAPI、TogetherAI、Google、OpenAI、Anthropic等）
⚡ 智能负载均衡: 基于响应时间、成本和成功率的智能负载均衡与故障转移
📊 实时健康监控: 自动监控API健康状态和性能指标
🔑 高性能API密钥管理: 100倍性能提升的API密钥管理系统，支持智能失败避免
🎯 多模态支持: 支持文本生成、多模态输入（图像+文本）和函数调用
🚀 帕累托最优选择: 从多个模型中智能选择最优模型
💰 成本优化: 健康检查屏蔽功能，避免昂贵模型的不必要检查
🐳 容器化部署: 完整的Docker支持，开箱即用

🏗️ 系统架构

┌─────────────────┐    ┌──────────────────┐    ┌─────────────────┐
│   AIRouter      │    │  LoadBalancing   │    │ Health Monitor  │
│   (Core API)    │◄──►│   (智能路由)     │◄──►│   (健康检查)    │
└─────────────────┘    └──────────────────┘    └─────────────────┘
         │                        │                        │
         ▼                        ▼                        ▼
┌─────────────────┐    ┌──────────────────┐    ┌─────────────────┐
│ Multiple LLM    │    │  API Key Manager │    │   Performance   │
│ Providers       │    │    (Service)     │    │   Analytics     │
└─────────────────┘    └──────────────────┘    └─────────────────┘

📦 快速开始

安装

方式一：作为Python包安装（推荐）

git clone https://github.com/your-username/AIRouter.git
cd AIRouter
pip install -e .

方式二：直接安装依赖

git clone https://github.com/your-username/AIRouter.git
cd AIRouter  
pip install -r requirements.txt

配置

1. 配置API密钥

cp ew_config/api_keys.example.py ew_config/api_keys_local.py
# 编辑 ew_config/api_keys_local.py，填入您的真实API密钥

2. 数据库设置

CREATE DATABASE airouter;
CREATE TABLE api_key_usage (
    request_id VARCHAR(50) PRIMARY KEY,
    api_key VARCHAR(100) NOT NULL,
    model_name VARCHAR(50) NOT NULL,
    source_name VARCHAR(50) NOT NULL,
    prompt_tokens INT,
    completion_tokens INT,
    create_time DATETIME NOT NULL,
    finish_time DATETIME NOT NULL,
    execution_time FLOAT NOT NULL,
    status BOOLEAN NOT NULL
);

3. 环境变量

# 复制环境变量配置文件
cp env.example .env

# 编辑 .env 文件，填入您的真实数据库信息
# 注意：请确保设置 DB_PASSWORD，这是必需的环境变量

启动服务

Docker部署（推荐）

# 启动服务
docker-compose up -d

# 查看服务状态
docker-compose ps

手动启动

# 启动健康检查服务
python CheckHealthy.py

# 在另一个终端启动API密钥管理服务
python -m api_key_manager.main

🚀 使用示例

基本文本生成

from LLMwrapper import LLM_Wrapper

# 简单的文本生成
response = LLM_Wrapper.generate(
    model_name="gpt4o_mini",
    prompt="解释量子计算的基本原理"
)
print(response)

多模态输入

import base64

# 图像+文本输入
with open("image.jpg", "rb") as f:
    img_base64 = base64.b64encode(f.read()).decode()

response = LLM_Wrapper.generate_mm(
    model_name="gpt4o_mini",
    prompt="描述这张图片的内容",
    img_base64=img_base64
)
print(response)

函数调用

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "获取天气信息",
            "parameters": {
                "type": "object",
                "properties": {
                    "location": {"type": "string", "description": "城市名称"}
                },
                "required": ["location"]
            }
        }
    }
]

response = LLM_Wrapper.function_calling(
    model_name="gpt4o_mini",
    prompt="北京今天天气如何？",
    tools=tools
)
print(response)

🔧 高级配置

负载均衡策略

# 配置负载均衡模式
response = LLM_Wrapper.generate(
    model_name="gpt4o_mini",
    prompt="你好",
    mode="cost_first"  # 可选: fast_first, cost_first, balanced
)

帕累托最优选择

# 从多个模型中选择最优结果
response = LLM_Wrapper.generate_fromTHEbest(
    model_list=["gpt4o_mini", "claude35_sonnet", "gemini15_pro"],
    prompt="复杂推理任务"
)

🤝 贡献指南

我们欢迎所有形式的贡献！请查看 CONTRIBUTING.md 了解详细信息。

📄 许可证

这个项目使用 MIT 许可证 - 查看 LICENSE 文件了解详细信息。

🆘 支持

如果您在使用过程中遇到问题，请：

查看 GitHub Issues
创建新的 Issue 描述您的问题

🇺🇸 English Version

An intelligent AI router that provides high-performance unified API interfaces for Large Language Models (LLMs), supporting multi-source load balancing and intelligent failover.

✨ Key Features

🔄 Unified API Interface: Access multiple LLM providers through a single interface (OpenRouter, DeepInfra, DeerAPI, TogetherAI, Google, OpenAI, Anthropic, etc.)
⚡ Smart Load Balancing: Intelligent load balancing and failover based on response time, cost, and success rate
📊 Real-time Health Monitoring: Automatic monitoring of API health status and performance metrics
🔑 High-performance API Key Management: 100x performance improvement in API key management with intelligent failure avoidance
🎯 Multimodal Support: Support for text generation, multimodal input (image + text), and function calling
🚀 Pareto Optimal Selection: Intelligently select the optimal model from multiple options
💰 Cost Optimization: Health check blacklist feature to avoid unnecessary checks on expensive models
🐳 Containerized Deployment: Complete Docker support, ready to use out of the box

🏗️ System Architecture

┌─────────────────┐    ┌──────────────────┐    ┌─────────────────┐
│   AIRouter      │    │  LoadBalancing   │    │ Health Monitor  │
│   (Core API)    │◄──►│ (Smart Routing)  │◄──►│ (Health Check)  │
└─────────────────┘    └──────────────────┘    └─────────────────┘
         │                        │                        │
         ▼                        ▼                        ▼
┌─────────────────┐    ┌──────────────────┐    ┌─────────────────┐
│ Multiple LLM    │    │  API Key Manager │    │   Performance   │
│ Providers       │    │    (Service)     │    │   Analytics     │
└─────────────────┘    └──────────────────┘    └─────────────────┘

📦 Quick Start

Installation

Method 1: Install as Python Package (Recommended)

git clone https://github.com/your-username/AIRouter.git
cd AIRouter
pip install -e .

Method 2: Install Dependencies Directly

git clone https://github.com/your-username/AIRouter.git
cd AIRouter  
pip install -r requirements.txt

Configuration

1. Configure API Keys

cp ew_config/api_keys.example.py ew_config/api_keys_local.py
# Edit ew_config/api_keys_local.py and fill in your real API keys

2. Database Setup

CREATE DATABASE airouter;
CREATE TABLE api_key_usage (
    request_id VARCHAR(50) PRIMARY KEY,
    api_key VARCHAR(100) NOT NULL,
    model_name VARCHAR(50) NOT NULL,
    source_name VARCHAR(50) NOT NULL,
    prompt_tokens INT,
    completion_tokens INT,
    create_time DATETIME NOT NULL,
    finish_time DATETIME NOT NULL,
    execution_time FLOAT NOT NULL,
    status BOOLEAN NOT NULL
);

3. Environment Variables

# Copy environment variable configuration file
cp env.example .env

# Edit .env file and fill in your real database information
# Note: Make sure to set DB_PASSWORD, this is a required environment variable

Start Services

Docker Deployment (Recommended)

# Start services
docker-compose up -d

# Check service status
docker-compose ps

Manual Start

# Start health check service
python CheckHealthy.py

# Start API key manager service in another terminal
python -m api_key_manager.main

🚀 Usage Examples

Basic Text Generation

from LLMwrapper import LLM_Wrapper

# Simple text generation
response = LLM_Wrapper.generate(
    model_name="gpt4o_mini",
    prompt="Explain the basic principles of quantum computing"
)
print(response)

Multimodal Input

import base64

# Image + text input
with open("image.jpg", "rb") as f:
    img_base64 = base64.b64encode(f.read()).decode()

response = LLM_Wrapper.generate_mm(
    model_name="gpt4o_mini",
    prompt="Describe the content of this image",
    img_base64=img_base64
)
print(response)

Function Calling

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "Get weather information",
            "parameters": {
                "type": "object",
                "properties": {
                    "location": {"type": "string", "description": "City name"}
                },
                "required": ["location"]
            }
        }
    }
]

response = LLM_Wrapper.function_calling(
    model_name="gpt4o_mini",
    prompt="What's the weather like in Beijing today?",
    tools=tools
)
print(response)

🔧 Advanced Configuration

Load Balancing Strategies

# Configure load balancing mode
response = LLM_Wrapper.generate(
    model_name="gpt4o_mini",
    prompt="Hello",
    mode="cost_first"  # Options: fast_first, cost_first, balanced
)

Pareto Optimal Selection

# Select optimal results from multiple models
response = LLM_Wrapper.generate_fromTHEbest(
    model_list=["gpt4o_mini", "claude35_sonnet", "gemini15_pro"],
    prompt="Complex reasoning task"
)

🛠️ Development Guide

Project Structure

AIRouter/
├── LLMwrapper.py          # Core API interface
├── LoadBalancing.py       # Load balancing logic
├── CheckHealthy.py        # Health check service
├── api_key_manager/       # API key management service
├── ew_config/            # Configuration files
├── ew_api/               # API infrastructure
├── ew_decorator/         # Decorator utilities
└── unit_test/

AIRouter

Install / Use

README

AIRouter

🌐 Language / 语言

🇨🇳 中文版

✨ 主要特性

🏗️ 系统架构

📦 快速开始

安装

配置

启动服务

🚀 使用示例

基本文本生成

多模态输入

函数调用

🔧 高级配置

负载均衡策略

帕累托最优选择

🤝 贡献指南

📄 许可证

🆘 支持

🇺🇸 English Version

✨ Key Features

🏗️ System Architecture

📦 Quick Start

Installation

Configuration

Start Services

🚀 Usage Examples

Basic Text Generation

Multimodal Input

Function Calling

🔧 Advanced Configuration

Load Balancing Strategies

Pareto Optimal Selection

🛠️ Development Guide

Project Structure