FreeLLMAPI

FreeLLMAPI is a self-hosted router that stacks the free tiers of many LLM providers behind one OpenAI-compatible /v1 endpoint. The stated pool is roughly 7.4 billion tokens a month across 34 providers and 635 free model endpoints, drawn from 474 model families. Provider keys are AES-256-GCM encrypted in a local SQLite database and decrypted in memory per request; your applications only ever see a single unified bearer token.

The router picks the highest-priority model with a healthy key that is under all of its rate limits, calls that provider, and on a 429 or 5xx cools the key down and retries the next model in your chain. Six ranking strategies use live speed, capability, and reliability scores. The same model offered by several providers collapses into one entry with in-group failover, and named fallback-chain profiles can be selected per request with auto: .

Beyond chat completions, it serves the OpenAI responses, completions, images, videos, speech, transcription, embeddings, and models endpoints, streaming and non-streaming; an Anthropic Messages surface at /v1/messages; a native Gemini surface at /v1beta; and opt-in Ollama emulation for local-model clients. A custom provider lets you point chat, embedding, image, or audio models at any OpenAI-compatible endpoint such as llama.cpp, LM Studio, vLLM, a local Ollama, or a remote gateway.

Setup is a Docker one-liner or a desktop installer for macOS and Windows, with a menu-bar app that runs the router and dashboard locally. More than fifteen generator commands (npx freellmapi setup-claude, setup-codex, setup-aider, and others) write client configuration from your live catalogue, support --dry-run, back up existing files, and merge rather than overwrite; freellmapi launch and launch-codex inject credentials into the child process instead of writing them to disk.

The catalogue updates itself from a signed feed twice a day; free installs track a monthly snapshot each model joins 30 days after it lands in the live feed, and a paid tier receives it the same day. The documented trade-offs of stacking free tiers are no frontier models, variable latency, no SLA, and effective capability dipping late in the day as top models hit daily caps before resetting at UTC midnight. It runs on Node 20 or newer at around 40 MB resident at idle and is designed for a single local user.

Features

• One endpoint, many free tiers: 34 providers and 635 free model endpoints behind a single OpenAI-compatible /v1 API
• Every OpenAI surface: chat completions, responses, images, videos, speech, transcriptions, embeddings, and models, streaming or not
• Anthropic, Gemini, and Ollama wires: /v1/messages, /v1beta, and opt-in Ollama emulation alongside the OpenAI surface
• Smart routing: six ranking strategies over live speed, capability, and reliability scores, with failover on 429 and 5xx plus key rotation and cooldowns
• Quota tracking: per-key RPM, RPD, TPM, and TPD counters that learn provider ceilings so requests stay under every free-tier cap
• Fusion: a virtual model that fans a prompt out to several models and synthesises the drafts with a judge model
• Tool calling and structured outputs: OpenAI-style tools round-trip across providers, plus response_format, seed, and logprobs
• Sticky sessions: conversations stay on one model for 30 minutes, with an optional handoff note on a mid-chat switch
• Opt-in prompt compression: a fail-open pipeline that deduplicates prompts, filters tool output, and trims stale context before routing
• Encrypted keys: provider keys AES-256-GCM encrypted in SQLite, exposed to clients only as one unified token
• Client generators: setup commands for more than fifteen agents and editors, with --dry-run and backups
• Dashboard and MCP: a React admin UI with latency analytics in 60 languages, and an /mcp server agents can query for usable models and provider health