A free, open-source, OpenAI-compatible gateway that pools the free tiers of 16 LLM providers.
freellmpool is a free, open-source Python tool that pools the free tiers of 16 LLM
providers — Groq, Cerebras, NVIDIA, Google Gemini, OpenRouter, Cloudflare, Mistral, Cohere and more —
behind one OpenAI-compatible endpoint. It runs as a command-line tool, a Python library, a
local proxy, and an MCP server, with automatic failover
when a provider is rate-limited and per-day quota tracking across tiers. Two providers (Pollinations
and OVHcloud) need no API key, so pip install freellmpool answers immediately — no signup.
The proxy speaks both the OpenAI and the Anthropic API, so existing apps and coding agents such as
Codex, Claude Code, and aider run on pooled free models with no code changes.
pip install freellmpool
freellmpool ask "Explain the CAP theorem in one sentence." # no API key neededGitHub · PyPI · Changelog · MIT license
| Tool | What it is | Install | Keyless start | CLI / library / proxy / MCP |
|---|---|---|---|---|
| freellmpool | Pools many providers' free tiers | pip install | Yes (2 providers) | All four |
| OpenRouter | Hosted paid aggregator (some free models) | API key | No (account + key) | API only |
| LiteLLM | Multi-provider SDK/proxy (bring your own keys) | pip install | No | Library + proxy |
| Self-hosted free-API servers | A server you deploy | Docker + config | No (deploy first) | Server only |
freellmpool's niche: the keyless, pip-installable client for squeezing the hosted free tiers — not a server you deploy and not a paid aggregator.
freellmpool ask "...", pipe stdin, pin a provider/model.OPENAI_BASE_URL at freellmpool proxy; existing code is unchanged.freellmpool code claude (also codex, aider, cline, continue, cursor) routes them to pooled free models — see the step-by-step guide.from freellmpool import Pool, AsyncPool.freellmpool mcp lets Claude Desktop/Code/Cursor offload subtasks to free models.freellmpool benchmark + latency-aware routing.Per-provider free tiers: Groq · Gemini · Cerebras · NVIDIA · Mistral
Yes — freellmpool is a free, open-source, MIT-licensed gateway that exposes one OpenAI-compatible
endpoint backed by the free tiers of 16 providers. Install it with pip install freellmpool
and point any OpenAI client at the local proxy.
freellmpool pools them: each request goes to a provider you have access to, and if that one is
rate-limited or down it automatically fails over to the next. It tracks per-day usage so load spreads
evenly across the free tiers, and an optional fast routing mode prefers the
lowest-latency provider.
Yes. The proxy implements both the OpenAI and the Anthropic APIs, so you set
OPENAI_BASE_URL or ANTHROPIC_BASE_URL to the local proxy and run Codex,
Claude Code, aider, Cline, Continue, or Cursor on pooled free models without changing any code.
Run freellmpool code claude for the exact setup. (The Claude Code path is experimental:
text and tool use, no vision yet.)
No key is required to start: Pollinations and OVHcloud work anonymously, so a fresh install answers immediately. Add free keys for the other 14 providers to unlock more models and higher limits.
16 providers: Pollinations, OVHcloud, LLM7, Groq, Cerebras, NVIDIA NIM, OpenRouter, Google Gemini, GitHub Models, Cloudflare Workers AI, Mistral, Cohere, SambaNova, Z.ai/Zhipu, Ollama Cloud, and LongCat — 235 live-validated chat models, plus free embeddings.
OpenRouter is a hosted aggregator you pay per token (with some rate-limited free models) and access
through its API. freellmpool is a local, open-source tool you pip install that pools the
providers' own free tiers directly, works with no key to start, and ships a CLI, a Python library, a
proxy, and an MCP server.
Yes — it is MIT-licensed and free to use. Source is on GitHub; the package is on PyPI.
Free-tier models are smaller than frontier models — good for drafting, classification, triage, and everyday coding, not GPT-class reasoning on the hardest problems. Limits reset at UTC midnight and capacity varies through the day. The proxy is meant for local/single-user use.