Gemini CLI OpenAI Proxy

A lightweight Node.js server that acts as a bridge between OpenAI-compatible clients and the Google Gemini CLI.

This project allows you to use the Gemini CLI programmatically with tools, libraries, and UIs designed for the OpenAI API (like generic chat interfaces, VS Code extensions, or AutoGPT-style scripts) by mimicking the v1/chat/completions endpoint.

Features

• OpenAI Compatibility: Implements /v1/chat/completions and /v1/models.
• Streaming Support: Converts Gemini's stream-json output into OpenAI Server-Sent Events (SSE) for real-time token streaming.
• Context Management: Automatically converts OpenAI conversation history (Messages API) into a structured prompt format suitable for the stateless Gemini CLI.
• Tool Grounding: Includes system prompt injection to prevent tool hallucinations (e.g., stopping the model from trying to run shell commands when only file search is available).
• Authentication: Simple Bearer token authentication.

Prerequisites

1. Node.js: v14+ installed.
2. Gemini CLI: You must have the Gemini CLI installed and authenticated on the host machine.
   • Ensure running gemini --prompt "Hello" works in your terminal.

Installation

1. Clone the repository:

   git clone https://github.com/apuryear/gemini-cli-openai-proxy.git
   cd gemini-cli-openai-proxy

2. Install dependencies:

   npm install express body-parser cors

3. Start the server:

   node server.js

   By default, the server runs on port 3000.

Configuration

You can configure the server using environment variables:

Variable	Description	Default
PORT	The port the server listens on.	3000
OPENAI_API_KEY	The Bearer token required by clients.	sk-dummy-key

Example:

export OPENAI_API_KEY="my-secret-key"
export PORT=8080
node server.js