freellmpool

# freellmpool

> Free, open-source, OpenAI-compatible gateway that catalogs free-tier routes across 22 LLM
> providers (Groq, Cerebras, NVIDIA, Gemini, OpenRouter, Cloudflare, Mistral, Cohere
> and more) behind one endpoint, with automatic failover and per-day quota tracking.
> Works as a CLI, a Python library, a local proxy, and an MCP server. Keyless to
> start (`pip install freellmpool`); the proxy speaks the OpenAI API and has an experimental Anthropic-compatible path,
> so Codex, Claude Code, and aider run on pooled free models unchanged. MIT licensed.
> Capacity tools (`freellmpool capacity status`, `providers health`, `keys add`) show which
> free tiers are usable right now and help configure more. Capacity status is offline/cache-first;
> use `--refresh` only when you want to update the advisory external catalog.
> Text-only Responses and Anthropic requests stream incrementally. Tool calls and richer content remain
> buffered, and stream failover is allowed only before the downstream response is committed.

## Release status
- Latest release: 0.13.0. GitHub and PyPI include the audited provider catalog,
  bounded streaming and sentinel hardening, the Hermes profile, `spread` routing,
  `/livez`, `/readyz`, `/v1/providers`, `/v1/models?ready=true`, and OpenCode
  registry-readiness hardening.
- OpenCode npm registry publication is pending; `opencode-freellmpool` and
  `opencode-freellmpool-tui` were not published as of 2026-08-29.

## Docs
- [README](https://github.com/0xzr/freellmpool/blob/main/README.md): install, CLI, library, proxy, MCP, provider keys, routing.
- [Landing page / FAQ](https://0xzr.github.io/freellmpool/): what it is, how it compares, frequently asked questions.
- [Capacity & provider health](https://github.com/0xzr/freellmpool/blob/main/docs/CAPACITY.md): capacity status, health checks, key inventory, external catalog. Guide: https://0xzr.github.io/freellmpool/capacity-management.html
- [Local runtimes](https://github.com/0xzr/freellmpool/blob/main/README.md#local-runtimes-explicit-pin-only): explicitly preview fixed literal-loopback runtimes, affirmatively import pin-only routes, and remove managed imports without credentials or network scanning.
- [CHANGELOG](https://github.com/0xzr/freellmpool/blob/main/CHANGELOG.md): release history.
- [Integrations](https://github.com/0xzr/freellmpool/blob/main/docs/INTEGRATIONS.md): OpenCode, Hermes, metaswarm review adapter, aider, Continue, Cline, Cursor, Open WebUI, LangChain, LlamaIndex, and more.
- [Run OpenCode on free models](https://0xzr.github.io/freellmpool/run-opencode-on-free-models.html): OpenCode setup with an embedded TUI dashboard, long-run agent routing (`freellmpool/agent`), raised stream timeouts, and usage tools.
- [MCP server](https://github.com/0xzr/freellmpool/blob/main/docs/MCP.md): use free models from Claude Desktop/Code/Cursor.
- [Frontier model source audit (2026-07-28)](https://github.com/0xzr/freellmpool/blob/main/docs/FRONTIER_MODEL_SOURCE_AUDIT_2026-07-28.md): 207 public immutable Git blob objects plus current first-party model/runtime sources checked before catalogue and agent-routing changes.

## Install
- PyPI: `pip install freellmpool` (https://pypi.org/project/freellmpool/)