Gemgate

OpenAI-compatible API server powered by Gemini via Google AI Studio.

gemgate/
├── server.js         # Relay server
├── models.js         # Model catalog
├── package.json
├── processor/        # AI Studio app source
│   └── index.tsx
├── README.md
└── LICENSE

How it works

Client (OpenAI SDK) → Gemgate → WebSocket → AI Studio Processor → Gemini API

The server speaks the OpenAI API. An ngrok tunnel makes it publicly reachable. The processor runs as a hosted app in AI Studio and handles the Gemini API calls.

Quick start

1. Start the server

npx gemgate

2. Open the processor

https://ai.studio/apps/4bf06673-9b53-4f03-9002-3822741dcd88?fullscreenApplet=true

Enter the WebSocket URL and Processor Token from the server console.

3. Use it

Point any OpenAI-compatible client at the API Endpoint with the API Key from the console.

curl https://your-tunnel.ngrok-free.dev/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gemini-flash-latest","messages":[{"role":"user","content":"Hello"}]}'

CLI

gemgate [options]

Options:
  --port <n>      Port (default: 3777)
  --no-tunnel     Skip ngrok, localhost only
  --help          Show help

Environment

Create a .env file or set env vars:

| Variable | Description | |----------|-------------| | GEMGATE_KEY | Fixed API key (auto-generated if unset) | | GEMGATE_PROC | Fixed processor token (auto-generated if unset) | | NGROK_AUTHTOKEN | ngrok auth token | | NGROK_DOMAIN | specificed domain | | PORT | Port number | | REQUEST_TIMEOUT | Timeout in ms (default: 120000) |

API

| Method | Path | Auth | Description | |--------|------|------|-------------| | POST | /v1/chat/completions | Yes | Chat completions (streaming + sync) | | POST | /v1/embeddings | Yes | Text embeddings | | GET | /v1/models | Yes | List models | | GET | /health | No | Health check |

Supported

• Chat completions (streaming + sync)
• Function/tool calling
• System instructions
• Temperature, top_p, stop sequences
• JSON mode and structured output
• Reasoning effort control
• Embeddings
• Multi-modal input (images, audio)

Requirements

• Node.js 18+
• ngrok free account (or --no-tunnel for localhost)

License

MIT