FreeLLMAPI

Overview

Most AI providers give away a free tier, and on its own each one is small. FreeLLMAPI puts your accounts for all of them behind a single address that speaks the OpenAI API. Your editor, your coding agent, or your own code points at this app, and it picks a model, tracks how much of each free tier is left, and moves to the next provider when one runs out.

Features

• One OpenAI-compatible address for chat, embeddings, images, speech, and transcription, streaming or not
• Also speaks the Anthropic and Gemini formats, so Claude Code and Gemini CLI work without a shim
• Automatic failover: a rate-limited or failing provider is retried on the next one, with cooldowns and key rotation
• Six routing strategies, plus named fallback chains you can pick per request (a coding chain, a vision chain)
• Per-key counters for requests and tokens, so routing stays under every free-tier cap
• Your provider keys are encrypted at rest; your tools only ever see one key this app issues
• A dashboard for keys, routing, a playground, and speed and error charts
• Add any OpenAI-compatible endpoint of your own, including a model running on your own machine

Setting up

You need the one-time setup code to create the first account. The app writes it to its own log at every start until an account exists, so open this app's Logs from its page on your box, copy the code, then register. The account you create is the only one: this app has a single login, and there is no way to add a second.

After that, open Keys and paste in a key from each provider you have an account with. Nothing routes until at least one key is there. Then copy the key this app issues you and set it, plus this app's address, in whatever tool you want to use.

Good to know

• Your tools can reach it, and the rest stays private. moose keeps this app owner-only, so the dashboard is behind your box login. The API addresses your tools call are the exception: they answer without it, and the key this app issues is what guards them.
• The author says this is for personal experiments, not production. Free tiers have no service promise: models vary, the best ones run out late in the day, and none of it is guaranteed. Do not build something you depend on top of it.
• Using free tiers this way is between you and each provider. You sign up with them yourself, and the terms you accepted still apply when the traffic goes through this app. It is worth a read before you add a key.
• The model list updates itself over the internet. The app fetches a signed list of providers and models from the author's server twice a day. Your prompts, answers, and keys never go there.
• The free model list runs about a month behind. New models reach a free install 30 days after they go live. The paid upgrade from the author removes that delay. Nothing is switched off without it.
• The providers themselves may charge you. Their free tiers are free, but if you add a key to a paid plan, or pass a free limit into a paid one, that provider bills you directly.