Skip to content
← Tools / Skills
ToolSteel/10 trustVerified

Steel

Open-source browser API for AI agents — self-hostable cloud Chrome sessions.

Visit official site →

About

Steel (steel-dev/steel-browser) is an Apache 2.0 open-source browser API built specifically for AI agents and web automation. Its core unit is a sandboxed Chrome session — isolated state, cookies, anti-detection, proxies, and a live session viewer — driven over Puppeteer, Playwright, Selenium, or the Chrome DevTools Protocol. The Sessions API launches and manages cloud browser sessions, runs stateless quick actions (scrape, screenshot, PDF, search), and records Agent Logs plus MP4 replays so failed runs are debuggable instead of opaque. The repo carries roughly 7,000 GitHub stars.

Deployment is a single docker run of the published image, with the API live on port 3000 and a session inspection UI included; the same runtime ships as managed Steel Cloud. Integrations cover Browser Use, LangGraph, CrewAI, Pydantic AI, Mastra, and Stagehand.

Limitation: it is an early-stage company with a smaller enterprise track record than incumbents like Browserless, and its Selenium integration does not support all features of the CDP-based sessions API — Selenium-first teams should test their exact flow before committing.

What it can do

  • Sandboxed Chrome sessions with isolated state, cookies, and anti-detection
  • Live session viewer, MP4 replays, and Agent Logs for debugging
  • Puppeteer, Playwright, Selenium, and raw CDP access
  • Self-hostable Apache 2.0 runtime or managed Steel Cloud

Setup

Self-host with docker run -p 3000:3000 -p 9223:9223 ghcr.io/steel-dev/steel-browser, or sign up for Steel Cloud.

Prerequisites

Docker for self-hosting; Puppeteer, Playwright, or Selenium client for driving sessions.

Use cases

browser-automationai-agentsweb-scrapingcomputer-use

Works with

Trust score — 8/10

/10 trustVerified

Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 61; Trending score 75/100 from researcher assessment.

★ 7,000 GitHub stars

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.