FreeLLMAPI

FreeLLMAPI is a self-hosted router that aggregates the free tiers of 34 LLM providers, plus custom OpenAI-compatible chat, embedding, image, and audio endpoints, behind a single OpenAI-compatible API. Stacked together the free tiers add up to roughly 7.4 billion tokens a month across 474 model families and 635 provider endpoints. Any OpenAI client library points at the local server, which routes each request across whichever providers have keys configured, falls over to the next provider when one is rate-limited, and tracks per-key usage to stay under every free-tier cap.

The router picks the highest-priority model with a healthy key that is under all its rate limits, decrypts the key in memory, calls the provider, and on a 429 or 5xx cools that key down and retries the next model in the chain. Six routing strategies rank the chain on live per-model speed, capability, and reliability scores, the same model on several providers collapses into one entry with in-group failover, and named fallback-chain profiles (a coding chain, a vision chain) switch from the dashboard or per request. Provider keys are AES-256-GCM encrypted in SQLite and client apps see one unified bearer token. The model catalog syncs from a signed feed twice a day. Free installs receive a monthly snapshot (a model reaches them 30 days after it joins the live feed) and a paid tier receives it the same day.

Beyond the OpenAI-style surfaces (chat completions, responses, completions, images, videos, speech, transcriptions, embeddings, and models), the router speaks the Anthropic Messages API, a native Gemini surface, and opt-in Ollama emulation, so Claude Code, Codex, Gemini CLI, and local-model clients such as Zed and JetBrains run against the free pool. Documented clients include Claude Code, Codex, Gemini CLI, Cursor, Cline, Roo Code, OpenCode, Aider, Continue, Goose, Qwen Code, Kilo Code, Crush, OpenClaw, Hermes, and Pi, and most agents configure themselves with one command that reads the live catalog and merges into the existing config. A virtual fusion model fans a prompt out to a panel of free models in parallel and has a judge model synthesize one answer, tool calls round-trip across providers, and an opt-in compression pipeline deduplicates prompts and trims stale context before routing.

It runs as a Node server, a container, or a native menu-bar desktop app for macOS and Windows, with a React dashboard (60 languages) for keys, chain order, a playground, and p50, p95, and TTFT analytics, plus an MCP server that lets ChatGPT and other MCP clients run inference and inspect routing. It compares itself against OpenRouter, LiteLLM, and Portkey. The project is local-first and single-user by design, positioned for personal experimentation rather than production: the free pool has no frontier models, variable latency, no SLA, and its effective capability dips late in the day as top models hit daily caps.