Steel

Introducing Steel CLI v0.2.0 - a super-powered browser automation CLI ✦ New Agent Skills ✦ agent-browser integration ✦ Stealth: captcha solving & proxies ✦ Run background browser sessions in parallel Demo: Claude uses 3 browser sandboxes to scrape @tryramp ChatGPT rankings

An interesting use of our new Steel CLI 👇 > let an agent fight the messy web UI once, > then export the exact steel browser commands as a deterministic bash runbook > you and your agent can rerun it anytime

Give your agent the browser it deserves by running: `npx skills add steel-dev/cli` Check out the docs here: docs.steel.dev/overview/steel… Looking forward to your feedback! It's still pretty experimental and has lots of room for contributions :)

> `steel login` Don't get slowed down managing api keys when working locally. Log in once and you're authenticated forever across all your sessions.

Introducing Steel CLI v0.2.0 - a super-powered browser automation CLI ✦ New Agent Skills ✦ agent-browser integration ✦ Stealth: captcha solving & proxies ✦ Run background browser sessions in parallel Demo: Claude uses 3 browser sandboxes to scrape @tryramp ChatGPT rankings

@leodev @convex makes sense 👀

@convex We need a convex component for @steeldotdev 👀

Time to hack

Steel works with Playwright, Puppeteer, and Selenium. Same APIs you already know. Just swap the connection URL.

If you save ~1s per loop and your agent runs 20 loops, you've just won ~20s on a single task. Multiply across hundreds of workers and you either delight users or grow a queue. steel.dev/blog/remote-br…

Mobile mode strips away nested navigation and hover menus. Linear flows mean clearer action spaces and fewer mistakes for vision agents.

Production agents get blocked even when intent is legit. Anti-bot systems score a stack, not a single event. If one layer drifts, the whole run fails.

Old replay: event-based recreation, diverges from reality. New replay: MP4 recording, 1:1 with what happened. No more "why did this fail?"

Your agent needs to complete auth across multiple pages? Human-in-the-loop setup: open the debug URL, handle 2FA yourself, then let the agent take over with valid credentials.

Profiles are like Chrome identities for your agents. Log in once, save the state, reuse it forever. Cookies, auth tokens, extensions—persisted across sessions.

Full article: steel.dev/blog/happy-pat…

You know Puppeteer. You already know Steel. The happy path is just: create → connect → run → release. Full walkthrough 👇