steel-browser

Why we included this project

Agents that have to actually browse the web usually run into a wall of infrastructure before they reach the interesting part. Steel runs real Chrome instances behind a simple HTTP API, so your code can open pages, keep sessions and cookies alive across requests, and save the results as markdown, screenshots, or PDFs. It manages browser processes, local storage, proxy chains for IP rotation, and stealth and fingerprint handling, yet you can still drive the browser directly through Puppeteer, Playwright, or Selenium when you want finer control. The project ships as a Docker image, so the same setup works for a few local containers during experiments and for a cloud deployment under heavier load. A built-in session viewer and request logging let you see what the browser actually did, which is often the hardest part of agent development.