Features • Quick Start • Getting Started • API Reference • Architecture

---

Free Provider is a self-hosted, OpenAI-compatible API server and chat web UI that is powered by the ChatGPT web client — no API keys, no paid plan, zero-cost inference.

It exposes familiar endpoints like POST /v1/chat/completions and GET /v1/models, so any OpenAI SDK or tool can point at it locally and not pay a cent.

---

Features

• OpenAI-compatible — drop-in replacement endpoints (/v1/models, /v1/chat/completions)
• Auto model balancing — model: "auto" alternates each request between ChatGPT and Gemini, round-robin through paired per-provider windows (one window from each backend work together as a pair)
• Streaming (SSE) — real-time, token-by-token responses
• Stateless — every request is independent, so multi-turn context is preserved via the message array
• API-key auth — bearer-token validation out of the box
• Rate limiting + cooldown — exponential backoff, inter-request delay, and a global cooldown after repeated Cloudflare blocks (returns 429 + Retry-After)
• Proxy rotation — rotate through http/https/socks4/socks5 proxies on blocks, with per-proxy benching
• Streaming retry — SSE streams retry like ordinary requests instead of dying on first failure
• Polished web UI — dark, responsive chat interface with markdown + code highlighting
• Built-in logging — every user message and AI response is logged to the terminal and chat.log
• Anonymous or cookie mode — works with no account, or unlock more with your ChatGPT cookies
• Zero-cost — runs entirely against the public ChatGPT web backend

---

Quick Start

1. Install

git clone <your-repo-url>
cd <repo> && npm install     # client deps (undici for proxies)
cd <repo>/server && npm install    # server deps (express)

2. Run

npm start          # or: pnpm start

Open the UI at http://localhost:3000.

The terminal prints the server's API key and the exact export line to reuse it:

API Key:  sk-free-… (generated, saved in .api-key)
-> export OPENAI_API_KEY="sk-free-…"

3. Call it like OpenAI

curl http://localhost:3000/v1/chat/completions \
  -H "Authorization: Bearer sk-free-…" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4o","messages":[{"role":"user","content":"Hello"}],"stream":true}'

Or point any OpenAI SDK at http://localhost:3000/v1.

---

Why it exists

The bundled client library (in client/ & src/) reverse-engineers ChatGPT's web endpoints — the same requests the browser makes. The server wraps that client in a clean, OpenAI-compatible surface so you get a free, local AI provider without needing an OpenAI API key.

⚠️ This is an unofficial project. It relies on ChatGPT's web backend, which can change at any time. Use responsibly and at your own risk.

---

Docs

Doc	Description
Getting Started	Install, configure, and run everything
API Reference	Every HTTP endpoint with examples
Architecture	How the server + client fit together
Frontend	The web chat UI and its settings
Security	API keys, cookies, and logging
Client API	The underlying client library methods
Client Guide	Using the client directly
Testing	Running the 400+ test suite and 85% coverage gate

---

License

MIT © 2025 — see LICENSE.