Steel Browser

Steel, developed by the steel-dev organization, provides browser infrastructure for AI agents and apps. Steel Browser is the open-source, self-hostable server released under Apache 2.0, while Steel Cloud is the managed version that adds scale, stealth, proxies and CAPTCHA solving. The README notes the project is in public beta, and the latest listed release is v0.5.4-beta.

What It Is

Steel Browser is a browser API that manages sessions, pages and browser processes so developers can build web agents and automation without building their own browser infrastructure. It uses Puppeteer and the Chrome DevTools Protocol, so Puppeteer, Playwright or Selenium scripts can connect to it. Steel describes itself as infrastructure rather than an agent: your own agent or framework connects to it and Steel handles the browser lifecycle.

How Sessions and Quick Actions Work

The Sessions API creates browser sessions with options such as proxies, ad blocking, custom dimensions and extensions, and keeps cookies, local storage and state across requests. A Selenium mode is available for existing WebDriver workflows, though it does not support every feature of the CDP-based sessions. For simple read-only jobs, the Quick Actions endpoints scrape pages, take screenshots and produce PDFs, and the browser tools can also convert pages to markdown or readability output. Node.js and Python SDKs and a REST API work with both Steel Cloud and self-hosted instances.

Self-Hosting and Cloud Options

The server can run from a pre-built Docker image that serves the API and a debugging UI, via docker compose, or directly with Node.js and Chrome. Railway and Render one-click deployments are listed. Steel's site says Steel Cloud adds session persistence, profiles, a live session viewer with human takeover, readable traces, residential proxies and CAPTCHA solving. It also lists Stealth Browser on its Enterprise plan and self-hosting options for enterprise customers.