FreeLLMAPI Setup Guide: Run Claude Code on 34 Free AI Providers

Almost every AI lab now gives away a free tier: a few million tokens a month, a few thousand requests a day. On its own each one is small. FreeLLMAPI stacks them. You add free keys from the providers you sign up for, and it gives you one OpenAI-compatible endpoint at http://localhost:3001 on your own computer. When one provider hits its limit, it automatically moves your request to the next one.

Per its README on October 2, 2026, the catalog tracks 34 providers, 474 model families and 635 free model endpoints, adding up to roughly 7.4 billion tokens per month of listed free-tier capacity. It also speaks Anthropic's message format at /v1/messages, which is what lets Claude Code run on it.

• Automatic failover: if a provider is rate-limited or errors, the next model in your chain picks up the request.
• Sticky sessions: a conversation stays on one model for 30 minutes, so answers stay consistent.
• Context handoff (optional): when a mid-chat switch does happen, it adds a short note telling the new model what it is picking up.
• Encrypted keys: your provider keys are stored encrypted on your machine. Your apps only ever see one FreeLLMAPI key.

• Claude Code already installed.
• Node.js 20.18 or newer (for the npx commands below).
• Either the FreeLLMAPI desktop app (macOS or Windows) or Docker for the one-line server install.
• Free accounts and API keys from the providers you want to use. You only need a few to start.

Easiest: download the desktop app from the repo's Releases page. The macOS .dmg and Windows .exe are attached to every release, and there is no account or password to set up. It runs the whole router and dashboard from your menu bar.

Or with Docker: this one line sets up a ~/freellmapi folder, generates an encryption key, pulls the image and starts the server. You can read the script first at https://freellmapi.co/install.sh.

Not sure which to pick? Paste this into Claude Code and let it walk you through it:

1. 1Open the dashboard at http://localhost:3001 (or from the tray icon if you used the desktop app).
2. 2Go to the Keys page and paste in the free API keys from the providers you signed up for.
3. 3Reorder the Fallback Chain so the models you like best are tried first.
4. 4Copy your unified API key from the Keys page header (or the tray popover). This is the one key Claude Code will use.

Our pick is the launcher. It starts a Claude Code session with the FreeLLMAPI credentials injected into that one process only, and never writes them to disk. Your normal Claude Code setup stays exactly as it was, so you can keep paid work on your paid plan and use the free pool in a separate window.

Want it permanent instead? The setup command writes FreeLLMAPI into your Claude Code config. Run it with --dry-run first to see the exact change. A real run merges into your existing config and takes a timestamped backup first.

• Learning Claude Code and practicing prompts without watching your usage meter.
• Personal side projects and throwaway prototypes with nothing private in them.
• Bulk, low-stakes jobs like renaming files, drafting test data or summarizing public docs.
• Trying lots of different models side by side from the built-in playground.

• Your prompts and code leave your machine. Every request goes to whichever third-party provider the router picks. Never use it on client code, secrets or private repos.
• Personal experimentation only. The README says plainly that it is for experimentation and learning, not production. If you build something real, swap in a paid API before you ship.
• Provider terms still apply. Your account with each provider is governed by the terms you accepted, even when traffic goes through FreeLLMAPI. You are responsible for following them.
• Free tiers change and rate-limit. Providers launch, retire and tighten free models without notice. The README's own limitations list says no frontier models, variable speed and no uptime guarantee, and quality can dip late in the day as top models hit their daily caps.
• It is not Claude. Claude Code runs on top of it, but the answers come from whatever free models are in your chain, so expect weaker results on hard coding tasks.

1. 1Install the desktop app or run the Docker one-liner.
2. 2Add a few free provider keys on the Keys page and copy your unified key.
3. 3Run npx freellmapi launch to start a separate Claude Code session on the free pool.
4. 4Keep client work, secrets and private repos on your normal paid Claude Code.