What is mini-router?

MiniRouter is a self-hosted tool that combines more than 16 free AI providers, such as Gemini, Groq, Mistral, Cerebras, and HuggingFace, into a single API you run on your own server. It is meant to work like OpenRouter, a paid hosted service that gives access to many AI models through one interface, except MiniRouter runs on hardware you control and does not charge a monthly fee, since it relies entirely on providers' free tiers. The API speaks the same format that OpenAI's own API uses, so existing code written for OpenAI's client libraries can point at MiniRouter with only the base address and key changed. You get your own free API keys from each provider you want to use, list them in a configuration file along with a priority order, and the router picks between them accordingly. Setup can be done by running the Python code directly, using Docker Compose, or, on Windows, a one-click start script. Beyond routing requests, MiniRouter includes an admin API for creating individual user accounts with their own daily request limits, listing existing users, and pulling analytics on usage. A web dashboard shows the health status of each connected provider, usage statistics, estimated cost savings compared to paying for a service like OpenRouter, and tools for managing users. Rate limits and cost tracking apply per user rather than being shared globally across everyone using the server. The README lists specific free tier limits and relative speed for each supported provider, ranging from very fast options like Groq and Cerebras to slower ones like HuggingFace, so someone setting this up can choose providers that fit their needs. The project is released under the MIT license, which explicitly allows commercial use, and the maintainers welcome outside contributions through issues or pull requests.</content>