Gemini API Proxy (Compatible with Gemini-Cli)

A lightweight Google Gemini API-compatible proxy server that allows you to call OpenAI-compatible LLM services using the Gemini API format.

🔥 Quick Setup for Gemini CLI

Currently, Gemini CLI cannot easily use models other than Gemini. This Python tool was developed to meet this need.

Usage Steps:

1. Modify config.json to add your API provider configurations.
2. Install dependencies and run python gemini_proxy_for_kimi.py.
3. When prompted, select the API provider you want to use.
4. Set environment variables:

   export GOOGLE_GEMINI_BASE_URL=http://localhost:8000/
   export GEMINI_API_KEY=sk-1234

5. In Gemini CLI, use /auth and select "Use Gemini API Key".

⚠️ Important Note: The current version has only been fully tested and optimized on Moonshot Kimi. Other OpenAI-compatible services require your own testing and adjustments.

中文文档 | Chinese Documentation

✨ Features

• 🔄 Complete API Compatibility - Supports all Gemini API endpoints
• 🔀 Intelligent Format Conversion - Seamless Gemini ↔ OpenAI format conversion
• 🌊 Streaming Response Support - Complete Server-Sent Events streaming processing
• 🛠️ Function Calling Support - Bidirectional conversion of tool calls
• 🗣️ Multi-turn Conversations - Complete conversation history handling
• 📊 Model Mapping - Flexible model name mapping configuration
• 📝 Detailed Logging - Configurable access logs and detailed request logs
• ⚙️ Configuration Files - Unified management through JSON configuration files
• 🔁 Automatic Retries - Automatically retries requests on failure

🚀 Quick Start

1. Requirements

• Python 3.8+
• Dependencies: fastapi, uvicorn, openai

2. Install Dependencies

pip install -r requirements.txt

3. Configuration File

Create a config.json file, which now supports multiple providers:

{
    "providers": [
        {
            "name": "Moonshot",
            "openai_api_key": "sk-1234",
            "openai_base_url": "https://api.moonshot.cn/v1",
            "daily_limit": -1,
            "model_mapping": {
                "gemini-2.5-pro": "kimi-k2-0711-preview",
                "gemini-2.5-flash": "moonshot-v1-auto"
            },
            "default_openai_model": "kimi-k2-0711-preview"
        },
        {
            "name": "OpenAI",
            "openai_api_key": "sk-5678",
            "openai_base_url": "https://api.openai.com/v1",
            "daily_limit": -1,
            "model_mapping": {
                "gemini-2.5-pro": "gpt-4o",
                "gemini-2.5-flash": "gpt-4o-mini"
            },
            "default_openai_model": "gpt-4o-mini"
        }
    ],
    "server": {
        "host": "0.0.0.0",
        "port": 8000,
        "log_level": "info"
    },
    "logging": {
        "enable_detailed_logs": false,
        "enable_access_logs": true,
        "log_directory": "logs"
    },
    "retry": {
        "max_retries": 999,
        "wait_fixed": 5
    }
}

4. Start the Service

When you start the service, you will be prompted to select a provider:

python gemini_proxy_for_kimi.py

The service will start at http://0.0.0.0:8000.

📖 API Usage

Basic Request

curl -X POST http://localhost:8000/v1beta/models/gemini-2.5-pro:generateContent \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "parts": [{"text": "Hello, how are you?"}]
    }],
    "generationConfig": {
      "temperature": 0.7,
      "maxOutputTokens": 200
    }
  }'

Streaming Request

curl -X POST http://localhost:8000/v1beta/models/gemini-2.5-pro:streamGenerateContent \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "parts": [{"text": "Write a short story"}]
    }]
  }'

Token Counting

curl -X POST http://localhost:8000/v1beta/models/gemini-2.5-pro:countTokens \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "parts": [{"text": "Count tokens for this text"}]
    }]
  }'

Function Calling

curl -X POST http://localhost:8000/v1beta/models/gemini-2.5-pro:generateContent \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "parts": [{"text": "What is the weather like in Beijing?"}]
    }],
    "tools": [{
      "functionDeclarations": [{
        "name": "get_weather",
        "description": "Get weather information for a city",
        "parameters": {
          "type": "object",
          "properties": {
            "city": {"type": "string", "description": "City name"},
            "date": {"type": "string", "description": "Date in YYYY-MM-DD format"}
          },
          "required": ["city"]
        }
      }]
    }]
  }'

🔧 Configuration

Complete Configuration File (config.json)

{
    "providers": [
        {
            "name": "Moonshot",
            "openai_api_key": "sk-1234",
            "openai_base_url": "https://api.moonshot.cn/v1",
            "daily_limit": -1,
            "model_mapping": {
                "gemini-2.5-pro": "kimi-k2-0711-preview",
                "gemini-2.5-flash": "moonshot-v1-auto"
            },
            "default_openai_model": "kimi-k2-0711-preview"
        }
    ],
    "server": {
        "host": "0.0.0.0",
        "port": 8000,
        "log_level": "info"
    },
    "logging": {
        "enable_detailed_logs": false,
        "enable_access_logs": true,
        "log_directory": "logs"
    },
    "retry": {
        "max_retries": 999,
        "wait_fixed": 5
    }
}

Configuration Options

Option	Description	Default
providers	A list of API provider configurations.	[]
providers[].name	The name of the provider, for selection at startup.	Unnamed Provider
providers[].openai_api_key	OpenAI API key for the provider.	Required
providers[].openai_base_url	OpenAI API base URL for the provider.	https://api.openai.com/v1
providers[].daily_limit	Daily request limit for the provider. -1 means unlimited.	-1
providers[].model_mapping	Gemini to OpenAI model mapping for the provider.	{}
providers[].default_openai_model	Default OpenAI model for the provider.	gpt-3.5-turbo
server.host	Listen address	0.0.0.0
server.port	Listen port	8000
server.log_level	Log level	info
logging.enable_detailed_logs	Enable detailed request logs	false
logging.enable_access_logs	Enable access logs	true
logging.log_directory	Log directory	logs
retry.max_retries	Maximum number of retries on failure	3
retry.wait_fixed	Fixed wait time between retries in seconds	2

Provider Failover Logic

The service includes an automatic failover mechanism based on the daily_limit of each provider to ensure high availability:

1. Primary Provider: The service will always try to use the provider you selected at startup first.
2. Unlimited Failover: If the primary provider reaches its daily limit, the service will automatically search for and switch to the first available provider with an unlimited quota ("daily_limit": -1).
3. Limited Failover: If no unlimited providers are available, the service will then search for and switch to the first available provider that still has a remaining request quota.
4. Service Unavailable: If all configured providers have reached their daily limits, the API will return a 503 Service Unavailable error.

📊 Supported LLM Services

✅ Tested Services

Kimi (Moonshot) - Recommended

{
  "openai_api_key": "sk-xxx",
  "openai_base_url": "https://api.moonshot.cn/v1",
  "model_mapping": {
    "gemini-2.5-pro": "kimi-k2-0711-preview",
    "gemini-2.5-flash": "kimi-k2-0711-preview"
  }
}

⚠️ Services Requiring Your Own Testing

Note: The following services are theoretically compatible but require your own testing and adjustments. You can enable enable_detailed_logs in the config file to output detailed request and response information for targeted adaptation.

OpenAI

{
  "openai_api_key": "sk-xxx",
  "openai_base_url": "https://api.openai.com/v1",
  "model_mapping": {
    "gemini-2.5-pro": "gpt-4o",
    "gemini-2.5-flash": "gpt-4o-mini"
  }
}

Azure OpenAI

{
  "openai_api_key": "your-azure-key",
  "openai_base_url": "https://your-resource.openai.azure.com/openai/deployments/your-deployment",
  "model_mapping": {
    "gemini-2.5-pro": "gpt-4",
    "gemini-2.5-flash": "gpt-35-turbo"
  }
}

DeepSeek

{
  "openai_api_key": "sk-xxx", 
  "openai_base_url": "https://api.deepseek.com/v1",
  "model_mapping": {
    "gemini-2.5-pro": "deepseek-chat",
    "gemini-2.5-flash": "deepseek-chat"
  }
}

Zhipu AI

{
  "openai_api_key": "your-zhipu-key",
  "openai_base_url": "https://open.bigmodel.cn/api/paas/v4",
  "model_mapping": {
    "gemini-2.5-pro": "glm-4",
    "gemini-2.5-flash": "glm-4-flash"
  }
}

Ollama (Local Deployment)

{
  "openai_api_key": "ollama",
  "openai_base_url": "http://localhost:11434/v1",
  "model_mapping": {
    "gemini-2.5-pro": "llama3:8b",
    "gemini-2.5-flash": "llama3:8b"
  }
}

🛠️ Customizing Other Services

If you need to adapt other OpenAI-compatible services:

1. Clone the project: git clone
2. Modify configuration: Update API endpoints and model mappings in config.json
3. Test functionality: Focus on testing streaming responses, function calls, multi-turn conversations. You can enable enable_detailed_logs in the config file to output detailed request and response information for targeted adaptation.
4. Optimize code: Adjust conversion logic based on target service characteristics

📝 Logging System

Access Logs

Concise access logs showing basic information for each request:

🚀 POST /v1beta/models/gemini-2.5-pro:generateContent - 200 - Model: gemini-2.5-pro - ID: abc12345 - 2.341s
🚀 POST /v1beta/models/gemini-2.5-pro:streamGenerateContent - 200 - Model: gemini-2.5-pro(stream) - ID: def67890 - 5.123s

Detailed Logs

Complete request/response conversion process (optional):

• 1_GEMINI_REQUEST - Original Gemini request
• 2_OPENAI_REQUEST - Converted OpenAI request
• 3_OPENAI_RESPONSE - Raw OpenAI response
• 4_GEMINI_RESPONSE - Final Gemini response

🚀 Production Deployment

Using Gunicorn (Recommended)

1. Install Gunicorn:

pip install gunicorn

2. Create startup script start.sh:

#!/bin/bash
gunicorn gemini_proxy_for_kimi:app \
  --worker-class uvicorn.workers.UvicornWorker \
  --workers 4 \
  --bind 0.0.0.0:8000 \
  --access-logfile - \
  --error-logfile - \
  --log-level info

3. Run:

chmod +x start.sh
./start.sh

Using systemd Service

Create /etc/systemd/system/gemini-proxy.service:

[Unit]
Description=Gemini API Proxy
After=network.target

[Service]
Type=exec
User=your-user
Group=your-group
WorkingDirectory=/path/to/your/app
ExecStart=/path/to/your/venv/bin/gunicorn gemini_proxy_for_kimi:app \
  --worker-class uvicorn.workers.UvicornWorker \
  --workers 4 \
  --bind 0.0.0.0:8000
Restart=always
RestartSec=3

[Install]
WantedBy=multi-user.target

Start the service:

sudo systemctl daemon-reload
sudo systemctl enable gemini-proxy
sudo systemctl start gemini-proxy

Nginx Reverse Proxy

Create Nginx configuration:

server {
    listen 80;
    server_name your-domain.com;

    location / {
        proxy_pass http://127.0.0.1:8000;
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto $scheme;
        
        # Support streaming responses
        proxy_buffering off;
        proxy_cache off;
        
        # Increase timeout
        proxy_connect_timeout 60s;
        proxy_send_timeout 60s;
        proxy_read_timeout 60s;
    }
}

Performance Optimization

1. Worker Count: Set based on CPU cores, recommended 2 * CPU_CORES + 1

2. Memory Optimization:

# Limit memory usage
gunicorn --max-requests 1000 --max-requests-jitter 100 ...

3. Connection Pool: Add connection pool settings in configuration

4. Caching: Consider adding Redis caching layer for responses

🔍 Monitoring and Maintenance

Health Check

curl http://localhost:8000/health

Returns:

{"status": "healthy", "service": "gemini-proxy"}

Log Rotation

Use logrotate to manage log files:

# /etc/logrotate.d/gemini-proxy
/path/to/your/app/logs/*.log {
    daily
    rotate 30
    compress
    delaycompress
    missingok
    create 644 your-user your-group
    postrotate
        systemctl reload gemini-proxy
    endscript
}

Monitoring Metrics

Access logs include:

• Request method and path
• Response status code
• Model used
• Request ID
• Response time

🛠️ Development and Debugging

Development Mode

# Enable detailed logs
# Set in config.json
{
  "logging": {
    "enable_detailed_logs": true,
    "enable_access_logs": true
  }
}

# Start development server
python gemini_proxy_for_kimi.py

Debugging Tips

1. View detailed logs: Enable enable_detailed_logs to see complete conversion process
2. Model mapping testing: Test mapping with different Gemini model names
3. Streaming response debugging: Observe SSE data streams
4. Function call debugging: Check format conversion of tool calls

🤝 Contributing

Issues and Pull Requests are welcome!

Development Environment Setup

git clone 
cd gemini-proxy
pip install -r requirements.txt

Code Standards

• Follow PEP 8 code style
• Add appropriate type annotations
• Write clear docstrings
• Ensure backward compatibility

📄 License

This project is licensed under the MIT License. See LICENSE file for details.

🆘 FAQ

Q: Which Gemini API endpoints are supported?

A: Supports generateContent, streamGenerateContent, countTokens, and health endpoints.

Q: Which LLM services are currently supported?

A: Currently only fully tested on Moonshot Kimi. Other OpenAI-compatible services are theoretically usable but require your own testing and adjustments.

Q: How to adapt other LLM services?

A: 1) Clone the project code 2) Modify config.json configuration 3) Focus on testing streaming responses, function calls, etc. 4) Adjust code logic as needed

Q: Why only support Kimi?

A: Different LLM services have variations in API details, response formats, error handling, etc., requiring targeted testing and optimization. Currently, efforts are focused on complete Kimi adaptation.

Q: What if streaming responses don't work?

A: Check if the client correctly handles text/event-stream format and ensure the network environment supports SSE.

Q: How to handle high concurrent requests?

A: Increase Gunicorn worker count, use load balancers, consider adding caching layers.

🔗 Related Links

• Google Gemini API Documentation
• Moonshot Kimi API Documentation

---

⭐ If this project helps you, please give us a Star!