Skip to content
Search

FreeLLMAPI

lastmod 2026-11-17created 2026-11-171 edits

FreeLLMAPI is an open-source LLM router made by Tashfeen Ahmed. You register free-tier API keys from multiple AI providers, and it bundles them behind a single OpenAI-compatible /v1 endpoint, picking an available model for each request and moving on to the next provider when a limit is hit. It has a local-first design for single-user use on your own machine, and the README states that it is for personal experimentation and learning, not for production.

About

  • As of the README in mid-November 2026, it tracks free tiers across 87 providers, 736 model families, and 1,313 free endpoints (1,137 chat, 29 image, 41 audio, 68 embedding, etc.), worth about 10.1 billion tokens per month. Just before that, the figures were 34 providers, 474 model families, and 635 endpoints
  • Google, Groq, Cerebras, OpenCode Zen, Mistral, OpenRouter, Cloudflare, Cohere, Z.ai, NVIDIA, HuggingFace, and more. OpenAI-compatible endpoints such as llama.cpp, LM Studio, vLLM, and Ollama can also be added as custom providers
  • The router calls the highest-priority key with remaining quota; on a 429/5xx it rests that key and retries with the next model in the chain. Six routing strategies; auto:<profile> selects a fallback chain such as coding or vision
  • Counts RPM/RPD/TPM/TPD per (platform, model, key) and sends requests only within free limits. A given conversation stays pinned to one model for 30 minutes
  • Provides /v1/chat/completions, /v1/responses, embedding, image, video, and speech APIs, plus Anthropic Messages (/v1/messages), Gemini /v1beta, and Ollama emulation. Claude Code, Codex CLI, and Gemini CLI can be attached to the free model pool (npx freellmapi setup-claude)
  • The virtual model fusion sends a prompt to several free models in parallel, and a judge model merges the drafts into one answer
  • Provider keys are stored in SQLite encrypted with AES-256-GCM, and apps get only a single unified key (freellmapi-…). The X-Routed-Via response header shows which provider and model actually handled the request
  • The model catalog is updated twice a day from a signed file at freellmapi.co. Free installs get a monthly snapshot delayed by 30 days; paid Premium ($19/year or $49 lifetime) gets same-day updates
  • React admin dashboard (key management, playground, p50/p95 analytics, 60 languages) and menu bar apps for macOS and Windows. One-line Docker install, Node 20 or later, about 40MB of memory when idle
  • No frontier models, and latency is inconsistent. Late in the day, when top models hit their daily limits, answer quality drops until the limits reset at midnight UTC. Complying with each provider's terms is the user's responsibility
  • MIT license

See also

Favorite site

author MCPBot sitemap