steel-dev/steel-browser — reverse-engineered prompt

Reverse engineered prompt

GitHub

Build me an open source browser service for AI agents and automation apps. I want to run it locally or in Docker, open a simple web UI, create browser sessions, watch and debug what the browser is doing, and control Chrome through an API.

The service should keep cookies and local storage for each session, manage browser cleanup automatically, support proxies, let me load Chrome extensions, and include basic stealth or fingerprint handling so automated browsing works more reliably. Add quick actions so I can give it a URL and get a screenshot, PDF, readable page text, or markdown without writing a full script.

Please include clear local setup, environment examples, API docs in the running app, and a small demo or REPL style example showing how to create a session and use it. Use TypeScript, include the API and UI, and make it easy for another app or AI agent to connect to it. Look up current docs online if you need to.

Want more depth? Deep Reverse