🤖 Gemini CLI Router (GCR)

Route your Gemini CLI requests to third-party AI providers seamlessly

GCR is a comprehensive solution that enables Google Gemini CLI to work seamlessly with third-party AI providers like SHUAIHONG, DeepSeek, OpenAI, Claude, and others. Due to Google SDK's technical limitations (hardcoded endpoints, certificate binding, authentication mechanisms), we provide both a proxy service and necessary CLI modifications to achieve complete third-party integration.

✨ Features

• 🔄 Complete Integration: Modified CLI with enhanced UI showing proxy status and third-party models
• 🌐 Multiple Providers: Support for SHUAIHONG, DeepSeek, OpenAI, Claude
• 🔑 Flexible Auth: Support both API key and OAuth authentication
• 🎯 Model Override: Use -m parameter to override models on-the-fly
• ⚡ Fast Setup: One-command global installation
• 🛡️ Privacy First: Your API keys stay local in ~/.gemini-cli-router/.env

🚀 Two Installation Options

Option 1: 🔧 Modified CLI (Recommended - Full Features)

For complete third-party integration with enhanced UI, install our modified version:

# Clone the repository
git clone https://github.com/Jasonzhangf/gemini-cli-router.git
cd gemini-cli-router

# One-click build and install
./build-and-install.sh

Features:

• ✅ UI shows proxy status: 🔄 Proxy:3458 | SHUAIHONG | gpt-4o
• ✅ Real-time proxy detection
• ✅ Third-party model display in footer
• ✅ Complete integration with all CLI features

Usage:

gemini-local -m gpt-4o -p "Hello GPT-4o!"
gemini-proxy -m claude-3.5-sonnet -p "Hi Claude!"

Option 2: 📦 Proxy-Only (Original CLI + External Proxy)

Use with unmodified official Gemini CLI:

Step 1: Install Official Gemini CLI

npm install -g @google/gemini-cli

Step 2: Install GCR Proxy

npm install -g gemini-cli-router

Limitations:

• ❌ UI doesn't show proxy information
• ❌ Requires manual proxy management
• ❌ Less seamless integration

Configuration

Step 3: Configure GCR

1. Edit your configuration file:

nano ~/.gemini-cli-router/.env

2. Add your provider settings:

# Gemini API Key (optional, uses OAuth if not set)
GCR_API_KEY=your_gemini_api_key_here

# Provider Configuration
GCR_PROVIDER=shuaihong
GCR_TARGET_API_KEY=your_provider_api_key_here
GCR_BASE_URL=https://ai.shuaihong.fun/v1
GCR_MODEL=gemini-2.5-pro

# Server Configuration
GCR_PORT=3458
GCR_HOST=localhost
GCR_DEBUG=false

Usage

Step 4: Start Using GCR

Once installed and configured, use gcr instead of gemini:

# Interactive chat
gcr chat "Hello, how are you?"

# Override model
gcr -m gpt-4o chat "Hello"

# Any gemini command works
gcr config
gcr --help

🔧 Supported Providers

| Provider | Base URL | Models | |----------|----------|---------| | SHUAIHONG | https://ai.shuaihong.fun/v1 | gemini-2.5-pro, gpt-4o, claude-3.5-sonnet | | DeepSeek | https://api.deepseek.com/v1 | deepseek-chat, deepseek-coder | | OpenAI | https://api.openai.com/v1 | gpt-4o, gpt-4, gpt-3.5-turbo | | Claude | https://api.anthropic.com/v1 | claude-3.5-sonnet, claude-3-opus |

📋 Environment Variables

| Variable | Description | Default | |----------|-------------|---------| | GCR_API_KEY | Gemini API key (optional) | OAuth | | GCR_PROVIDER | Target provider | shuaihong | | GCR_TARGET_API_KEY | Provider API key | (required) | | GCR_BASE_URL | Provider base URL | Provider default | | GCR_MODEL | Default model | gpt-4o | | GCR_PORT | Proxy port | 3458 | | GCR_HOST | Proxy host | localhost | | GCR_DEBUG | Debug logging | false |

🛠️ How It Works

1. Proxy Interception: GCR starts a local proxy server on port 3458
2. Environment Override: Sets GEMINI_API_BASE_URL to point to the proxy
3. API Translation: Converts Gemini API calls to target provider format
4. Response Translation: Converts provider responses back to Gemini format
5. Seamless Experience: Your Gemini CLI works exactly as before

Gemini CLI → GCR Proxy → Third-Party Provider → Response Translation → Gemini CLI

🔐 Authentication

GCR supports both authentication methods:

API Key (Recommended)

Set GCR_API_KEY in your config file to use API key authentication and avoid OAuth prompts.

OAuth

Leave GCR_API_KEY empty to use OAuth authentication (you'll be prompted to authenticate).

📦 Complete Installation Guide

Step-by-Step Installation

1. Install Official Gemini CLI (Required)

   npm install -g @google/gemini-cli

2. Install GCR via NPM (Recommended)

   npm install -g gemini-cli-router

3. Alternative: Install from Source

   git clone https://github.com/Jasonzhangf/gemini-cli-router.git
   cd gemini-cli-router
   npm install -g .

4. Configure Your Provider

   # Edit configuration
   nano ~/.gemini-cli-router/.env
      
   # Add your settings
   GCR_PROVIDER=shuaihong
   GCR_TARGET_API_KEY=your_api_key_here

5. Start Using

   gcr chat "Hello, world!"

🗑️ Uninstallation

# Uninstall GCR
npm uninstall -g gemini-cli-router

# Optional: Uninstall Official Gemini CLI
npm uninstall -g @google/gemini-cli

🧪 Testing

Test your setup:

# Test proxy functionality
node test-proxy.js

# Test with real conversation
gcr chat "Hello, test message"

🐛 Troubleshooting

Common Issues

Port 3458 already in use:

# Kill existing processes
lsof -ti:3458 | xargs kill -9

Permission errors:

# Make scripts executable
chmod +x gcr-gemini install-gcr-simple.sh uninstall-gcr.sh

OAuth every time:

• Set GCR_API_KEY in ~/.gemini-cli-router/.env to avoid OAuth prompts

Official Gemini CLI not found:

• Make sure you installed it first: npm install -g @google/gemini-cli

Debug Mode

Enable debug logging:

export GCR_DEBUG=true
gcr chat "test"

🛠️ 技术原理：为什么需要修改Gemini CLI源码？

🎯 问题背景

虽然GCR项目最初设计为"零修改"方案，但在实际实现中发现，为了实现完整的第三方AI提供商路由转发，必须对Gemini CLI源码进行适度修改。这不是设计缺陷，而是Google Gemini CLI架构特性所决定的技术限制。

🔒 Google Gemini CLI的技术限制

1. API端点硬编码

// Google @google/genai SDK 内部硬编码
const API_ENDPOINT = "https://generativelanguage.googleapis.com";

• Google的SDK将API端点硬编码为Google服务器
• 无法通过配置文件或环境变量动态修改
• SDK内部包含端点验证逻辑，拒绝非Google域名

2. HTTPS证书和域名绑定

• Google SDK内置了证书验证机制
• 只信任*.googleapis.com域名的证书
• 即使DNS劫持也会被SSL/TLS层阻止

3. 认证机制深度绑定

// Google特定的认证流程
const auth = new GoogleAuth({
  scopes: ['https://www.googleapis.com/auth/generative-language']
});

• CLI使用Google特定的OAuth2认证流程
• Token验证与Google Identity平台绑定
• 第三方提供商无法验证Google issued tokens

4. 请求格式和响应处理

• Gemini API使用独特的请求格式（与OpenAI/Claude不同）
• CLI内置了Gemini特定的错误处理和重试逻辑
• 响应解析代码假设特定的JSON结构

❌ 为什么外部代理方案不可行？

1. 网络层拦截的局限性

# 这些方法都行不通
✗ DNS劫持: generativelanguage.googleapis.com -> 127.0.0.1
  - HTTPS证书验证失败
  - SSL/TLS握手被拒绝

✗ HTTP代理: --proxy http://127.0.0.1:3458
  - CLI的--proxy选项期待CONNECT隧道代理
  - 不支持HTTP应用层代理

✗ iptables重定向: 443端口转发
  - 证书不匹配导致连接失败
  - 影响系统其他HTTPS请求

2. 中间人攻击检测

// Google SDK内置安全检查
if (!cert.subject.includes('googleapis.com')) {
  throw new Error('Invalid certificate');
}

3. 加密通道的不可突破性

• HTTPS加密使内容检查和修改成为不可能
• 无法在传输层修改请求/响应格式
• 无法注入第三方认证Token

✅ 修改CLI源码的技术方案

1. API端点重定向

// packages/cli/src/config/config.ts
apiEndpoint: settings.apiEndpoint || 'http://127.0.0.1:3458'

// 效果：将所有API请求重定向到本地代理
// Google SDK -> http://127.0.0.1:3458/v1beta/models/*/generateContent

2. 代理服务器格式转换

// proxy-service/src/server.js
app.post('/v1beta/*', async (req, res) => {
  // 1. 接收Gemini格式请求
  const geminiRequest = req.body;
  
  // 2. 转换为OpenAI格式
  const openaiRequest = GeminiTranslator.translateRequest(geminiRequest);
  
  // 3. 转发到第三方提供商
  const response = await fetch(providerUrl, {
    method: 'POST',
    body: JSON.stringify(openaiRequest)
  });
  
  // 4. 转换响应回Gemini格式
  const geminiResponse = GeminiTranslator.translateResponse(await response.json());
  res.json(geminiResponse);
});

3. UI界面增强

// packages/cli/src/ui/components/Footer.tsx
// 实时显示代理状态
🔄 Proxy:3458 | SHUAIHONG | gpt-4o

// packages/cli/src/ui/App.tsx
// 每30秒检测代理服务器状态
useEffect(() => {
  const detectProxy = async () => {
    const response = await fetch('http://127.0.0.1:3458/health');
    setProxyInfo(await response.json());
  };
}, []);

4. 第三方模型支持

// -m参数增强支持
.option('model', {
  description: 'Model (supports third-party: gpt-4o, claude-3.5-sonnet, deepseek-chat)'
})

// URL路径传递模型信息
http://127.0.0.1:3458/v1beta/models/gpt-4o/generateContent

🎯 修改后的完整流程

用户输入: gemini-local -m gpt-4o -p "Hello"
    ↓
[修改后的CLI] 发送到: http://127.0.0.1:3458/v1beta/models/gpt-4o/generateContent
    ↓
[本地代理] 接收Gemini格式请求
    ↓
[格式转换] Gemini → OpenAI格式
    ↓
[转发请求] → https://ai.shuaihong.fun/v1/chat/completions
    ↓
[接收响应] ← OpenAI格式响应
    ↓
[格式转换] OpenAI → Gemini格式
    ↓
[返回CLI] ← Gemini格式响应
    ↓
[UI显示] Footer显示: 🔄 Proxy:3458 | SHUAIHONG | gpt-4o

💡 技术优势

1. 透明性: 用户体验与原版CLI完全一致
2. 完整性: 支持所有CLI功能和参数
3. 实时性: UI实时显示代理状态和模型信息
4. 扩展性: 轻松添加新的AI提供商
5. 安全性: 本地代理，API密钥不会泄露

🔧 本地构建和安装

# 一键构建修改版CLI
./build-and-install.sh

# 使用修改版CLI
gemini-local -m gpt-4o -p "Hello GPT-4o!"
gemini-proxy -m claude-3.5-sonnet -p "Hi Claude!"

总结：虽然修改CLI源码增加了复杂性，但这是实现完整第三方AI集成的唯一可行技术方案。我们的修改是最小化的、针对性的，保持了原有功能的同时增加了强大的路由转发能力。

📁 Project Structure

gemini-cli-router/
├── gcr-gemini                 # Main executable
├── proxy-service/            # Proxy server code
│   ├── src/
│   │   ├── server.js         # Express server
│   │   ├── config.js         # Configuration
│   │   └── gemini-translator.js
│   └── package.json
├── install-gcr-simple.sh     # Simple installer
├── install-gcr.sh            # Advanced installer  
├── uninstall-gcr.sh          # Uninstaller
├── setup-post-install.js     # Post-install setup
├── cleanup-pre-uninstall.js  # Cleanup script
├── test-proxy.js             # Test utility
└── package.json              # NPM package config

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

MIT License - see LICENSE file for details.

👨‍💻 Author

Jason Zhang - GitHub

🔗 Related Projects

• Official Gemini CLI - The official Google Gemini CLI
• Claude Code - Claude's official CLI

⭐ If you find GCR useful, please star this repository!