Steel provides an open-source browser API for AI agents and applications, simplifying web automation by managing browser instances and infrastructure. It targets developers building live web agents, offering a robust solution for programmatic web interaction without the overhead of managing browser processes.

How It Works

Steel leverages Puppeteer and the Chrome DevTools Protocol (CDP) for deep control over browser instances, enabling integration with Puppeteer, Playwright, or Selenium. It features session management for persistent browser state (cookies, local storage), proxy support for IP rotation, and custom extension loading. Anti-detection measures, debugging tools, and resource management are included to facilitate reliable and stealthy web automation.

Quick Start & Requirements

• Local Deployment: Clone the repository and run docker compose up. This starts the API on port 3000 and the UI on port 5173.
• Node.js: Requires Node.js and Chrome installed. Run npm install and npm run dev. The CHROME_EXECUTABLE_PATH environment variable can specify a custom Chrome location.
• SDKs: Python and Node.js SDKs are available for easier interaction.
• Documentation: API Reference, Cookbook, Node SDK Reference, Python SDK Reference.

Highlighted Details

• Full browser control via Puppeteer and CDP, compatible with Playwright and Selenium.
• Built-in proxy chain management and Chrome extension support.
• Anti-detection features and debugging tools, including request logging and a UI.
• Quick Actions API for direct scraping, screenshots, and PDF generation.

Maintenance & Community

• Active development with a public beta phase.
• Community engagement via Discord and GitHub issues.
• Twitter presence.

Licensing & Compatibility

• Licensed under the Apache 2.0 license.
• Permissive license suitable for commercial use and integration into closed-source applications.

Limitations & Caveats

The project is in public beta, indicating potential for evolving APIs and breaking changes. The Selenium integration notes that it does not support all features of the CDP-based sessions.