One API Key. 1 Billion Free AI Tokens. Every Month.
One API Key. 1 Billion Free AI Tokens. Every Month.14 free AI providers stacked behind one OpenAI-compatible endpoint. Self-hosted, MIT, no card.
One key, every free brain on the internet — Gemini, Groq, Cerebras, GitHub Models + 10 more. Auto-failover when one runs dry.
Repo: github.com/tashfeenahmed/freellmapi
🧠 How it works — 30 sec
Normal life: You want free AI → sign up at Google → sign up at Groq → sign up at OpenRouter → juggle 14 different keys → each one rate-limits you mid-task → rage quit.
FreeLLMAPI life: Sign up free at each → paste all keys into the FreeLLMAPI dashboard → it gives you ONE master key. Your app calls that one key. Behind the curtain, it picks whichever provider still has budget left and routes there. One runs dry? Auto-falls-back to the next. You never see the chaos.
Think of it like a credit-card juggling app for AI — except every “card” is a free tier and you never get a bill.
It speaks OpenAI’s language. Any app that works with ChatGPT’s API works with this — Cursor, LangChain, the OpenAI Python SDK, curl, whatever. Just swap one URL.
💰 The math — what 1 billion tokens actually buys
Per month, free, stacked together:
| Provider | Free quota/mo | Best for |
|---|---|---|
 Mistral La Plateforme |
~1 billion (the whale) | Mistral Large/Medium, Codestral |
 Cerebras |
~30M (1M tokens/day) | Fastest tokens you’ll ever see — 30 RPM |
 Groq |
~30M | Llama 3.3 70B, GPT-OSS 120B/20B |
 OpenRouter |
~6M (19 free models) | DeepSeek, Kimi, Qwen, Llama, Gemma |
 Google Gemini |
~3M (20 RPD per model) | Gemini 2.5 Flash, Pro previews |
 GitHub Models |
~18M est. (50 RPD) | GPT-4.1, GPT-4o (free Copilot tier) |
 SambaNova |
~3M | DeepSeek V3.1/V3.2, Llama 4 Maverick |
 Cloudflare Workers AI |
~20M | Kimi K2.5, Qwen3, GLM-4.7 |
 Z.ai (Zhipu) |
~30M | GLM-4.5 Flash, GLM-4.7 Flash |
 Cohere |
~1-2M (1000 calls/mo, 20 RPM) | Command R+ |
 NVIDIA NIM |
credits-based | Llama 3.1 70B |
That’s ~$50–150/month of API value, free, every month.
🍿 The smart-routing magic
| Feature | Plain English |
|---|---|
model="auto" |
The router picks the best available model for your request — you don’t even have to choose |
 Auto-failover |
One provider rate-limits? Skips it, retries the next. Up to 20 attempts. |
 Sticky sessions |
Multi-turn convos stay on the same model for 30 min → no hallucination spike from mid-chat switching |
 Per-key tracking |
Counts RPM/RPD/TPM/TPD per provider per model → never overshoots a free cap |
 Encrypted key storage |
API keys locked with AES-256-GCM, decrypted only when needed |
 Single bearer key for your apps |
Your apps see ONE key. Upstream chaos never leaks. |
 Health checks |
Marks dead keys as healthy / rate_limited / invalid / error. Auto-skips dead ones. |
 Admin dashboard |
Pretty React UI: add keys, reorder fallback order, see analytics, run prompts in a playground |
 Tool calling / streaming |
Full OpenAI-style tools and tool_choice — works across every provider that supports it |
 Runs on a Raspberry Pi |
~40 MB RAM at idle. Cheap home server = free AI for life. |
🎮 Setup — 3 minutes
Prereqs: Node.js 20+ and npm. That’s it.
Clone + install + run:
git clone https://github.com/tashfeenahmed/freellmapi.git
cd freellmapi
npm install
Generate an encryption key (locks your API keys at rest):
cp .env.example .env
echo "ENCRYPTION_KEY=$(node -e "console.log(require('crypto').randomBytes(32).toString('hex'))")" >> .env
Start the server + dashboard:
npm run dev
Open http://localhost:5173 → paste in your provider keys → reorder the fallback chain → grab your unified key from the header. That’s the one your apps point at.
For production (single server, no dev mode):
npm run build
node server/dist/index.js
🪝 Hooking it up to your apps
Any OpenAI-compatible app/SDK works. Just point base_url at your local proxy.
Python (the most common):
from openai import OpenAI
client = OpenAI(
base_url="http://localhost:3001/v1",
api_key="freellmapi-your-unified-key",
)
resp = client.chat.completions.create(
model="auto", # let the router pick, OR specify like "gemini-2.5-flash"
messages=[{"role": "user", "content": "Summarise the fall of Rome in one sentence."}],
)
print(resp.choices[0].message.content)
curl (raw test):
curl http://localhost:3001/v1/chat/completions \
-H "Authorization: Bearer freellmapi-your-unified-key" \
-H "Content-Type: application/json" \
-d '{"model": "auto", "messages": [{"role": "user", "content": "hi"}]}'
🛡️ Catch check
| Q | A |
|---|---|
| Actually free? | Yes. Every listed provider has a recurring free tier, no card. Sign up free, paste key, done. |
| Quality vs paid GPT/Claude? | Mistral Large + DeepSeek V3.1 + Llama 3.3 70B = roughly Sonnet-tier for most tasks |
| Safe? | MIT licensed, 3.5k  , open source, code auditable. Keys encrypted locally. |
| Will providers ban me? | You’re using their official free tiers within limits. Same as using each one directly. |
| Embeddings / image gen / audio? |  Not yet supported. Text + tool calls only. |
| Multi-user / billing? | Single-user by design. Personal proxy. |
| Logs my prompts? | Local SQLite only. Nothing phoned home. |
| Vision / multimodal? | Not yet — text content only. |
🎯 What to build with this
| Play | Free fuel makes it possible |
|---|---|
 Personal AI assistant |
Run 24/7 on a Pi, never pay a bill |
 Mass content gen |
Blog posts, SEO, social — burn 18M tokens/mo guilt-free |
 Side project with AI |
Stop ghosting your ideas because OpenAI charges money |
 RAG over your docs |
Free embeddings… wait no. Use a separate free embeddings provider. |
| Cursor / Continue / VS Code |
Point them at this, code with free AI all day |
 LLM testing / benchmarking |
Compare 60 free models on the same prompt, no card |
| Internal company tool |
Run on a $5 VPS, serve your whole team |
Simple-pimple: clone repo → npm install → paste 14 free API keys into the dashboard → get ONE master key back → use it in any OpenAI-style app. Auto-falls-back when one runs out. Free 1 billion tokens a month, forever.
One key, every free brain. The juggling act is now someone else’s problem.
!