38K ·GitHub
What is this?
FreeLLMAPI aggregates the free tiers of 34 LLM providers (635 free model endpoints, 474 model families) behind a single OpenAI-compatible /v1 endpoint, with smart routing, automatic failover, and encrypted key storage.
FreeLLMAPI aggregates the free tiers of 34 LLM providers (635 free model endpoints, 474 model families) behind a single OpenAI-compatible /v1 endpoint, with smart routing, automatic failover, and encrypted key storage.Is this a library I import, or a standalone server I run?It's a self-hosted proxy server you run (via Docker or the desktop app); your apps simply point an OpenAI SDK at its unified endpoint instead of importing any code.
It's a self-hosted proxy server you run (via Docker or the desktop app); your apps simply point an OpenAI SDK at its unified endpoint instead of importing any code.How do I get it running?Run the one-line Docker install script, which sets up ~/freellmapi, generates an encryption key, pulls the image and starts the container; then add your provider keys on the Keys page and grab your unified API key.
Run the one-line Docker install script, which sets up ~/freellmapi, generates an encryption key, pulls the image and starts the container; then add your provider keys on the Keys page and grab your unified API key.Do I need Docker to use it?Docker is the recommended quick-start path; Windows users can skip it and use the desktop .exe installer instead, and building from source requires Node 20+.
Docker is the recommended quick-start path; Windows users can skip it and use the desktop .exe installer instead, and building from source requires Node 20+.What's the catch with the free version?It describes itself as deliberately narrow in scope, and free (non-premium) installs only get a monthly snapshot of the model catalog — a new model takes 30 days to reach you versus same-day for the $19/yr premium tier.
It describes itself as deliberately narrow in scope, and free (non-premium) installs only get a monthly snapshot of the model catalog — a new model takes 30 days to reach you versus same-day for the $19/yr premium tier.Can this run on something as small as a Raspberry Pi?Yes — it runs on Windows, macOS, Linux servers, or a small ARM board like a Raspberry Pi, using roughly 40MB RSS at idle.
Yes — it runs on Windows, macOS, Linux servers, or a small ARM board like a Raspberry Pi, using roughly 40MB RSS at idle.What is Fusion?Fusion is a virtual model that fans a prompt out to a panel of diverse free models in parallel, then uses a judge model to synthesize one combined answer from the drafts.
Fusion is a virtual model that fans a prompt out to a panel of diverse free models in parallel, then uses a judge model to synthesize one combined answer from the drafts.Is there a desktop app instead of the Docker container?Yes, a native menu-bar app for macOS (.dmg) and Windows (.exe) runs the full router and dashboard locally from your tray, with a popover showing live request stats and no account setup needed.
Yes, a native menu-bar app for macOS (.dmg) and Windows (.exe) runs the full router and dashboard locally from your tray, with a popover showing live request stats and no account setup needed.Can it generate images, video, or speech too, not just text?Yes, it exposes /v1/images/generations, /v1/videos/generations, and /v1/audio/speech endpoints that route across providers serving those media models, normalizing both synchronous and queued video jobs into a finished MP4.
Yes, it exposes /v1/images/generations, /v1/videos/generations, and /v1/audio/speech endpoints that route across providers serving those media models, normalizing both synchronous and queued video jobs into a finished MP4.Which coding agents can I hook this up to out of the box?Claude Code, Codex CLI, Gemini CLI, Aider, Cline, Roo Code, Continue, OpenCode, Goose, Qwen Code, Kilo Code, Crush, Cursor, Zed, JetBrains AI, DeepSeek Harness, and AtomCode each get a one-command setup generator like 'npx freellmapi setup-claude'.
Claude Code, Codex CLI, Gemini CLI, Aider, Cline, Roo Code, Continue, OpenCode, Goose, Qwen Code, Kilo Code, Crush, Cursor, Zed, JetBrains AI, DeepSeek Harness, and AtomCode each get a one-command setup generator like 'npx freellmapi setup-claude'.What are similar projects?casbin-gateway, gpt-load, osw, ThinkWatch-Lite, OmniRoute, opencodex, litellm, magpie
casbin-gateway, gpt-load, osw, ThinkWatch-Lite, OmniRoute, opencodex, litellm, magpieWhich topics is it in?llm-gateway, load-balancer, menu-bar, raspberry-pi, openai-compatible, encryption, gemini, sqlite, proxy, local-first, mcp-server, codex
llm-gateway, load-balancer, menu-bar, raspberry-pi, openai-compatible, encryption, gemini, sqlite, proxy, local-first, mcp-server, codex