Gemini Code: Anthropic-Compatible Gemini Proxy

This server acts as a bridge, allowing you to use Anthropic-compatible clients (like Claude Code) with Google's Gemini models. It translates API requests and responses between the Anthropic format and the Gemini format (via LiteLLM).

Features

• Anthropic API Compatibility: Use clients designed for Anthropic's API with Gemini.
• Gemini Model Mapping: Maps Anthropic model aliases (e.g., haiku, sonnet) to specific Gemini models.
• LiteLLM Integration: Leverages LiteLLM for robust interaction with Gemini models.
• Streaming Support: Handles streaming responses for interactive experiences.
• Tool Use Support: Translates tool/function calling between formats.
• Token Counting: Provides a /v1/messages/count_tokens endpoint.

Prerequisites

• A Google Gemini API key.
• Python 3.8+.

Setup

1. Clone the repository:

   git clone https://github.com/coffeegrind123/gemini-code.git # Or your fork
   cd gemini-code

2. Create and activate a virtual environment (recommended):

   python3 -m venv .venv
   source .venv/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

4. Configure Environment Variables: Copy the example environment file:

   cp .env.example .env

   Edit .env and add your Gemini API key:

   GEMINI_API_KEY="your-google-ai-studio-key" # Required
   
   # Optional: Customize which Gemini models are used for Anthropic aliases
   # BIG_MODEL="gemini-1.5-pro-latest"    # For 'sonnet' requests
   # SMALL_MODEL="gemini-1.5-flash-latest" # For 'haiku' requests

   The server defaults to gemini-1.5-pro-latest for BIG_MODEL (Claude's sonnet) and gemini-1.5-flash-latest for SMALL_MODEL (Claude's haiku).

5. Run the server:

   To run with auto-reload for development (restarts on code changes):

   uvicorn server:app --host 0.0.0.0 --port 8082 --reload

   To run normally:

   python server.py

   (The server.py file has a __main__ block that runs uvicorn for you. Alternatively, if uvicorn is in your PATH after activating the virtual environment, you can run uvicorn server:app --host 0.0.0.0 --port 8082 directly.)

Usage with Claude Code Client

1. Ensure Claude Code CLI is installed (npm install -g @anthropic-ai/claude-code).

2. Point the Claude Code client to your running proxy server:

   ANTHROPIC_BASE_URL=http://localhost:8082 claude

3. For optimal performance, disable conversation history compacting after starting Claude Code:

   /config
   

   Select the option to disable history compacting when prompted. This ensures proper handling of context between the client and proxy.

4. Utilizing the CLAUDE.md file:

   • The repository includes a CLAUDE.md file to improve Gemini's performance with Claude Code tooling.
   • Copy this file to any project directory where you'll be using Claude Code:

     cp /path/to/gemini-code/CLAUDE.md /your/project/directory/

   • When starting a new conversation with Claude Code in that directory, begin with:

     First read and process CLAUDE.md with intent. After understanding and agreeing to use the policies and practices outlined in the document, respond with YES
     

   • This ensures Gemini receives important context and instructions for better assistance.

How It Works

1. The proxy receives an API request formatted for Anthropic's Messages API.
2. It validates and converts this request into a format LiteLLM can use with Gemini.
3. The request is sent to the specified Gemini model via LiteLLM.
4. The Gemini response is received (either streaming or complete).
5. This response is converted back into the Anthropic Messages API format.
6. The Anthropic-formatted response is sent back to the client.

Model Mapping

• Requests for claude-3-haiku... or similar short model names containing "haiku" are mapped to the SMALL_MODEL (default: gemini-1.5-flash-latest).
• Requests for claude-3-sonnet... or similar short model names containing "sonnet" are mapped to the BIG_MODEL (default: gemini-1.5-pro-latest).
• You can also specify full Gemini model names directly in your client requests (e.g., gemini/gemini-1.5-pro-latest).

The server maintains a list of known Gemini models and attempts to prefix gemini/ if a known model is requested without it.

Endpoints

• POST /v1/messages: Main endpoint for sending messages to models (Anthropic compatible).
• POST /v1/messages/count_tokens: Estimates token count for a given set of messages (Anthropic compatible).
• GET /: Root endpoint, returns a welcome message.

Logging

The server provides detailed logging, including colorized output for model mappings and request processing, to help with debugging and monitoring.

Additional Files

CLAUDE.md

The repository includes a special CLAUDE.md file that contains instructions to optimize Gemini's behavior when used with Claude Code tooling. This file is designed to:

• Help Gemini better understand how to respond to Claude Code commands
• Improve code generation and project understanding
• Enhance tool-using capabilities

The file needs to be present in any directory where you run Claude Code, as Claude Code automatically reads from a CLAUDE.md file in the current working directory.

Contributing

Contributions are welcome! Please feel free to submit issues or pull requests.

Thanks

Special thanks to https://github.com/1rgs/claude-code-proxy for inspiration and original code.