FreeLLMAPI: how to get free LLM API access

FreeLLMAPI is an MIT-licensed, self-hosted proxy that puts many free-tier LLM providers behind one OpenAI-compatible /v1 endpoint, with automatic failover when a provider hits its rate limit. The project lives at github.com/tashfeenahmed/freellmapi. You run it on your own machine, add your own free API keys from providers such as Groq, Cerebras, Google AI Studio and Mistral, and point your apps at a single local URL and a single unified key. This page was last updated October 22, 2026.

What the numbers say (and why they differ)

The project changes fast, and two sources give different counts:

Source	Providers	Model endpoints	Monthly free tokens claimed
BetterStack guide (community guide, earlier version)	28	339	about 4 billion, only if all 28 providers are set up
Repository README (checked October 22, 2026)	34	635	"7.4 billion tokens per month"

Both figures are theoretical sums of every provider's free allowance. You only get them if you create an account and key at every provider, and most people will use far less. The repository shows 38.9k GitHub stars and the MIT license.

How to set it up

Facts below come from the repository README and the BetterStack guide.

Item	Value
Fast install (needs Docker)	curl -fsSL https://freellmapi.co/install.sh | bash
Local development	npm install && npm run dev
Server and dashboard port	3001 (Vite dev dashboard on 5173)
Node.js	20 or newer
Manual .env	ENCRYPTION_KEY from openssl rand -hex 32, plus PORT=3001 (BetterStack guide)
Provider key storage	AES-256-GCM encrypted at rest (SQLite)
Unified key	created in the dashboard; apps see one freellmapi-… bearer token

Steps:

1. Clone the repository, run npm install, copy .env.example to .env and set ENCRYPTION_KEY.
2. Start it with npm run dev and open http://localhost:3001.
3. In the dashboard, add one free API key per provider you want to use.
4. Create the unified API key under Keys.
5. In your app, set the base URL to http://localhost:3001/v1 and use the unified key. With model="auto" the router picks the model for you.

For coding tools the README lists a helper: npx freellmapi setup-claude --url http://localhost:3001 --api-key <unified-key> (setup commands for Codex and others are also mentioned by BetterStack).

Which endpoints it exposes

According to the README it speaks several dialects, so one proxy can serve different clients:

Endpoint	Used by
/v1/chat/completions, /v1/responses, /v1/embeddings	OpenAI-style clients
/v1/messages	Anthropic-style clients
/v1beta	Gemini-style clients
/mcp	MCP clients
/v1/models	model list

Named providers include Google, Groq, Cerebras, OpenCode Zen, Mistral, OpenRouter, Cloudflare, Cohere, Z.ai, NVIDIA and HuggingFace; the README says 22 more exist, plus a custom provider slot for any OpenAI-compatible endpoint (llama.cpp, LM Studio, vLLM, Ollama).

Routing: how failover works

Per BetterStack, requests go to the best available provider under a chosen strategy: Manual, Balanced, Smartest, Fastest, Most Reliable or Custom. Balanced weights reliability 50%, speed 25% and intelligence 25%. When one provider is rate-limited, the next one answers. Failover can land on a less capable model and latency varies.

Limits and cautions

The README describes the project as "Personal experimentation only", with "no frontier models, variable latency, no SLA", and says answer quality "dips late in the day" as top models hit daily caps. BetterStack adds: use one account per provider, do not resell access or share the proxy, and do not run a production service on free tiers. The README also notes that live catalog updates are a paid Premium feature ($19 per year or $49 lifetime); the router itself is free.

Security: the proxy stores your provider keys, so keep port 3001 off the public internet and protect the dashboard.

FreeLLMAPI is an independent project and is not affiliated with this site or with the providers it routes to.

Related

If you want a different self-hosted gateway for Claude Code, see our OmniRoute setup guide.

FAQ

What is FreeLLMAPI?

FreeLLMAPI is a free, MIT-licensed proxy you host yourself that routes requests across many free-tier LLM providers through one OpenAI-compatible endpoint.

Is FreeLLMAPI really free?

The router is free and open source. You still need free accounts at each provider, each with its own limits, and the optional Premium catalog costs $19 per year or $49 lifetime.

What port does FreeLLMAPI use?

Port 3001 for the server and dashboard; the development dashboard runs on 5173.

Can I use it in production?

No. The README says personal experimentation only, with no SLA, and free tiers can throttle at any time. Use a paid API for anything critical.

Does it work with Claude Code?

It exposes an Anthropic-style /v1/messages endpoint and a setup-claude helper, so it can be pointed at from Claude Code; quality depends on which free models answer.