CodeariaAcademy
GitHub repositorycommit this monthDevelopmentAutomation

FreeLLMAPI

39k

A local router that puts the free tiers of 34 providers (Gemini, Groq, Mistral, DeepSeek, Qwen and others) behind one address. Claude Code connects to it with one command and keeps working when your limit runs out. The answers come not from Claude but from free models of other labs. MIT.

We have not run it ourselves: described from the docs and the repository.

Open on GitHub

In numbers

GitHub stars as of 22.10.2026
free providers behind one address
tokens a month with keys for every provider (the authors' estimate)

What it is

Every AI provider has a free tier, but a small one: a hundred requests a day here, a per-minute cap there. One by one they are awkward to use; together they add up to a real amount. FreeLLMAPI installs on your computer, keeps the keys of every provider in an encrypted database and gives you one address and one key. An app or an agent sends its request there, and the router picks a free model itself and moves on to the next one when the current one runs out of quota.

For Claude Code this means the shell, the commands, your project files and your usual workflow stay, while models like Gemini, GLM, Qwen or DeepSeek do the answering. It does not match Claude in quality, but it is a good fallback when your subscription's daily limit runs out in the middle of a task. A conversation stays on one model for up to 30 minutes, and when a model switch is forced you can turn on a short note for the new model so it does not lose the thread.

How it works

Установка
On Windows the easiest route is the .exe installer from the repository's Releases, on macOS the .dmg. On a server or with Docker: curl -fsSL https://freellmapi.co/install.sh | bash. Node.js 20 or newer is required.
Ключи провайдеров
Open http://localhost:3001, add free keys on the Keys page (Google AI Studio, Groq, Mistral and others), set the order in the Fallback Chain and copy the unified key from the top of the Keys page. The more providers you add, the longer the limits last.
npx freellmapi setup-claude
Connects Claude Code to the router: npx freellmapi setup-claude --url http://localhost:3001 --api-key <your unified key>. If you would rather not write the key into a config file, start it with npx freellmapi launch.
FREELLMAPI_CONTEXT_HANDOFF
The note for the new model on a mid-conversation switch is off by default. Turn it on with the line FREELLMAPI_CONTEXT_HANDOFF=on_model_switch in the .env file.
X-Routed-Via
Every response carries a header naming the provider and model that served it, so you can see who is really answering instead of Claude.

Know before installing

This is not the Claude model. Claude Code stays as the shell, but free models of other labs do the answering, noticeably weaker on hard tasks.
The limits do not disappear: late in the UTC day the strong models hit their daily quotas and answers get weaker until UTC midnight. Response speed varies.
For personal experiments only: the authors say plainly it is not for production. You are responsible for each provider's terms: one account per provider, no reselling, no sharing your endpoint with other people.
The free version gets the model catalogue once a month, so new models show up about 30 days late. The paid subscription only speeds up the catalogue; the router itself is free.
Open on GitHub

Also read

Less code, more tests: Ponytail teaches Claude Code to stop writing what you don't needOctober 18, 2026Job Hunting With Claude Code: a Tailored Resume and Cover Letter for Every Posting in One Command (Setup From Scratch)October 17, 2026

Comments