What is ai-gateway?

ai-gateway is a self hosted proxy server for AI API requests, designed to sit in front of AI model providers like OpenAI and Claude and add management, routing, and billing features on top of them. The core idea is that instead of pointing your applications directly at an AI provider's API, you point them at this gateway instead. The gateway exposes OpenAI and Claude compatible endpoints and handles the rest: authenticating requests with your own tokens, routing them to the right upstream provider, tracking usage, and enforcing per user quotas. The architecture splits into two parts. The control plane, called the master, handles administration: managing users, tokens, channels, models, and routing rules through a web UI and admin API. The data plane, called an agent, handles the actual request traffic, relaying calls to upstream providers and reporting usage back. In the simplest setup, both run together in one process on a single machine. For larger deployments, you can run multiple agents across different locations or regions, kept in sync with the master over a WebSocket connection. Model routing lets you group several upstream providers or model variants under one name, with priority and weight settings so traffic spreads across them and failed requests retry automatically. A cross region routing feature can also send requests from one region to agents in another, which helps balance load or route around regional restrictions. A built in billing module tracks usage per request and rolls it up daily by user and channel. The project is written in Go, with the web frontend embedded directly into the compiled binary so no separate web server is needed. Deployment is via Docker Compose, or Kubernetes for larger setups. Building from source requires Go 1.25 and Node.js 20. It is licensed under MIT.