steel-dev/steel-browser

Steel is an open-source tool that gives AI applications a way to control a real web browser. The idea is that AI agents often need to browse websites, fill out forms, take screenshots, or extract content, but wiring up a browser to an AI system from scratch involves a lot of low-level infrastructure work. Steel handles that infrastructure layer so the AI application can focus on what it actually needs to do. Under the hood, Steel runs a Chrome browser process and exposes an API your code can call to create browser sessions, navigate pages, capture screenshots, convert pages to readable text or PDF, and manage cookies and local storage. It uses an existing browser automation protocol called CDP (Chrome DevTools Protocol), which means it is compatible with popular automation libraries like Puppeteer, Playwright, and Selenium. It also includes some detection-avoidance features and support for routing traffic through proxies. You can run Steel yourself using a pre-built Docker image or by installing it locally with Node.js and a Chrome installation. One Docker command starts the server on port 3000 and a debugging UI at the same address. There is also a hosted cloud version at steel.dev if you do not want to manage the server yourself. The project provides two SDKs, one for Python and one for Node.js, as well as a REST API with full documentation. A separate Cookbook repository contains ready-to-use examples for common tasks like scraping, web research, and interacting with sites that require login. Steel is aimed at developers building AI agents and automation tools that need reliable, programmable browser access. It is in public beta, open source, and actively developed. The license is available in the repository.