# gpt-web-gateway

> Self-hosted gateway that drives a real ChatGPT session in Chrome: chat completions with thinking modes, image generation and editing. Docker, MIT.

Source code: https://github.com/stufently/gpt-web-gateway
License: MIT

- OpenAI-style endpoints: /v1/chat/completions, /v1/responses, /v1/images/generations and /v1/images/edits; works with the OpenAI SDK.
- Thinking modes instant, standard, extended, extra_high and pro, with an applied echo block that shows what the UI really activated.
- Image generation and editing: batches up to 10, aspect ratios, reference images and multipart uploads.
- Multi-turn conversations via conversation_id and OpenAI-style streaming with stream: true.
- Operational hardening: request queue with Retry-After, typed errors, Prometheus metrics and a liveness endpoint with stuck detection.
- One-time remote login at /login (SSO/2FA supported) or automatic login with e-mail, password and TOTP secret.

## Documentation

- [gpt-web-gateway](https://stufently.github.io/gpt-web-gateway/docs/): OpenAI-compatible API gateway backed by the ChatGPT.com web UI. A headless Real Google Chrome (patchright — a patched Playwright fork) drives a real…
- [Security Policy](https://stufently.github.io/gpt-web-gateway/docs/security/): Please report security issues privately via GitHub Security Advisories ("Report a vulnerability" on the repository's Security tab) rather than public…

### How do I run gpt-web-gateway?

Build the Docker image and run it with the auth/ and images volumes mounted and API_KEY set, or copy .env.example to .env and run docker compose up -d. Then open /login once to sign in to ChatGPT; the session is saved to the auth/ volume.

### Can I use it with the OpenAI SDK?

Yes. Point the client at the gateway, for example base_url http://localhost:3000/v1 with your API_KEY as api_key, and call chat completions or images.generate as usual. The model field is accepted and ignored.

### Is using it allowed by OpenAI?

No. Automating the ChatGPT web interface violates OpenAI's Terms of Use and the account can be suspended or banned without warning. Use a dedicated account with its own Plus or Pro subscription, never your main one.

### Does it bypass ChatGPT rate limits or the subscription?

No. You need your own Plus or Pro subscription, the account's limits still apply, and the gateway applies cooldowns when ChatGPT rate-limits the account.

### How do I secure the gateway?

Never expose it without authentication: set API_KEY, put auth on your reverse proxy or bind the port to localhost. The web UI at /, generated images under /images/, /health and /metrics stay outside API_KEY by design, and the auth/ volume holds session cookies, so keep it private.

### How many requests can it handle at once?

Up to MAX_CONCURRENT_JOBS jobs run in parallel (1 to 5, default 3), each on its own browser tab; further jobs wait in a queue capped by MAX_QUEUE_SIZE. It is a personal or team gateway, not a high-throughput proxy.

## Optional

- [Full text of all docs](https://stufently.github.io/gpt-web-gateway/llms-full.txt)