Steel is an open-source browser API for AI agents and apps. It manages browser sessions, pages and browser processes behind a REST API with Puppeteer, Playwright, Selenium and CDP access, and it can run as the hosted Steel Cloud or as a self-hosted Docker image under Apache-2.0.[1][2]
Quick verdict
Steel is an open-source browser API for AI agents that manages sessions, proxies and stealth behind a REST, CDP, Puppeteer, Playwright and Selenium interface. It suits teams that want to self-host browser infrastructure under Apache-2.0 or use Steel Cloud, with the caveat that it is in public beta and enterprise terms are undocumented.[1][2][3]
What Steel does
Sessions keep browser state, cookies and local storage across requests, so multi-step authenticated flows survive between API calls.[1][2]
Full browser control runs over Puppeteer and CDP, and clients can connect with Puppeteer, Playwright or Selenium, including a Selenium-compatible session mode.[1]
Quick action endpoints convert pages to scraped content, screenshots or PDFs for simple stateless read-only jobs.[1]
Built-in proxy chain management, stealth plugins with fingerprint handling, custom Chrome extension loading, request logging and a UI for viewing and debugging sessions ship with the platform.[1][2]
Who it fits
Teams building AI web agents that need persistent authenticated sessions with proxy rotation and stealth rather than one-off page fetches.[1][2]
Developers who want to own their browser infrastructure and run it as a container in their own environment under an Apache-2.0 core.[1][2]
You want a self-hostable Apache-2.0 core with the option of Steel Cloud, rather than a hosted-only proprietary dependency.[1][2]
Your stack already speaks Puppeteer, Playwright or Selenium and you want session persistence, proxies and stealth without building that infrastructure yourself.[1]
Limits to know
Steel is in public beta and describes itself as evolving every day, so APIs and behavior may change under you.[1]
No SSO, data-residency, enterprise support or offline-use documentation was observed, so regulated or enterprise buyers should verify those points directly before committing.[1][2]
You need only simple stateless page extraction, where the quick scrape, screenshot and PDF actions already cover the job without sessions.[1]
You prefer a fully managed service with documented enterprise terms over operating a public-beta container yourself.[1]
Compatibility and deployment
Questions worth asking
Can I self-host Steel instead of using Steel Cloud?
Does Steel work with my existing Playwright or Selenium code?
Yes. The README shows Puppeteer, Playwright, Selenium and raw CDP connections to Steel sessions, so existing automation clients can point at Steel with small configuration changes rather than a rewrite.[1]
What are Steel quick actions for scraping and screenshots?
Sources and editorial review
Source research. Reviewed on October 13, 2026. Written by Sanjib Sah. AI assisted; reviewed by a human editor.
We have not run or tested this product.
- Steel - Open-source browser API
official · Observed October 13, 2026
Official homepage positions Steel as the open-source browser API for AI agents and apps, linking docs and GitHub with live sessions, proxies and integrations.
- steel-dev/steel-browser on GitHub
official · Observed October 13, 2026
Repo page shows Apache-2.0 license, Puppeteer and CDP browser control, session management, debug UI, and Docker plus compose deploy files.
- Steel browser README
official · Observed October 13, 2026
README details session, scrape, screenshot and PDF APIs, Playwright and Selenium compatibility, proxies and stealth, Steel Cloud accounts, Docker deploys and Apache 2.0 licensing in public beta.
- Steel documentation introduction
official · Observed October 13, 2026
Docs introduction describes open-source self-hosting or Steel Cloud, session and concurrency model, user-provided API keys, and undetectable proxy-backed browsing.