What it is
A self-hosted router that turns dozens of scattered free AI tiers into a single working API. FreeLLMAPI sits on your own machine or server, takes keys for whichever free providers you already have, and exposes them all behind one OpenAI-compatible /v1 endpoint. It is open source (MIT) and built to be installed and forgotten about.
Why it matters
Each free tier on its own is a couple of million tokens a month with a handful of requests per day. Stacked together across all supported providers, that adds up to roughly 7 billion tokens of usable inference a month. FreeLLMAPI handles the mess of wiring that up.
How routing works
- Picks the best available model per request using live speed, capability and reliability scores.
- Fails over to the next provider on 429/5xx with cooldowns and key rotation.
- Tracks per-key usage so requests always stay under every provider cap.
- Updates its own model catalog from a signed feed, so new models and quota changes land without a manual update.
Supported surfaces
The router serves chat completions, responses, completions, embeddings, image generation, video generation, audio speech and transcription. It also exposes an Anthropic-compatible endpoint for Claude Code and a native Gemini route for the Gemini CLI. Custom OpenAI-compatible endpoints (Ollama, LM Studio, vLLM) can be added as a provider.
Setup
Docker one-liner that starts the container and generates an encryption key:
curl -fsSL https://freellmapi.co/install.sh | bash
Then add provider keys on the Keys page and point any OpenAI SDK at http://localhost:3001/v1. Desktop apps are available for macOS and Windows, and the dashboard ships in 60 languages.
Licensing and stats
MIT license, 26,000+ GitHub stars, 744 commits across a small active multi-author project. Provider keys are stored AES-256-GCM encrypted in a local SQLite database.
