7.4 billion free AI tokens a month

This is the setup for freellmapi, the thing from the reel. Every number below is from the project's own README, checked on 30 August 2026.

What it actually does: it stacks the free tiers of 34 different AI providers behind one OpenAI compatible /v1 endpoint. You point your existing OpenAI SDK at it, and a router picks whichever free model is available and under its rate limit.

The numbers

Receipt	Value
Free tokens a month	7.4 billion
Providers	34
Free model endpoints	635
Memory at idle	~40 MB RSS
Runs on	anything with Node 20+, Raspberry Pi included
License	MIT
Stars	22,315

Setup, the fast path

Docker required. This creates ~/freellmapi, generates an encryption key, pulls the image and starts the container.

curl -fsSL https://freellmapi.co/install.sh | bash

Then open localhost:3001, add your provider keys on the Keys page, drag the fallback chain into the order you want, and copy the unified key from the top of that page. That single key is what your OpenAI SDK points at from then on.

Watch the port. It serves on 3001, which a lot of dev tools also use. Change it before you install rather than after.

There is a native menu bar app in the repo if you would rather skip containers, a Windows .exe in Releases, and an experimental Termux guide for Android.

How the router behaves

One request comes in. It picks the highest priority model that has a healthy key and is under all its rate limits, decrypts that key in memory, and calls the provider. On a 429 or a 5xx it puts that key on cooldown and moves to the next model in your chain.

So you are describing a preference order, and letting it degrade down that order as caps get hit.

The catches, and there are three real ones

The author says do not use it in production. Straight from the README: "This project is for personal experimentation and learning, not production. Free tiers exist so developers can prototype against them; they aren't a stable, supported inference substrate." If you build something real on it, swap in a paid API before you ship.

It gets worse in the evening. This is the one nobody mentions. The effective intelligence of the endpoint dips late in the day as the top models hit their daily caps, then resets at UTC midnight. Not slower. Worse, because you have slid further down your own fallback chain.

Your provider terms still apply. Routing a free tier through a proxy does not change what you agreed to when you signed up. The author reviewed this provider by provider in May 2026 and published what he found, which is more care than this category usually gets.

No frontier models, variable latency, no SLA.

Sources

• Repo: github.com/tashfeenahmed/freellmapi

Every number here is from that README on 30 August 2026. It moves fast: two weeks earlier it was 29 providers and 358 endpoints. If it has changed, trust the repo over this page.

Which providers are you stacking?

Ask me in The Rabbit Hole and I will tell you what order to put your fallback chain in.

It is the room where I keep the parts that don't fit on a page like this: the full run of numbers, what broke the first time, and the ones that looked good and then didn't work at all.

Two catches, up front, because that's the rule here. It's private, so you have to be inside to read anything. And it's new, so you'd be early rather than joining something busy.

Free: skool.com/the-rabbit-hole-3206/about