Stop using chromium to build scrapers.

Found this yesterday and have been digging through it since. Obscura. Open-source headless browser written from scratch in Rust, single static binary, drop-in for Puppeteer and Playwright through CDP. Apache 2.0.

The README numbers (author-supplied, so I'm quoting them, not citing as fact). 30 MB memory versus 200+ MB for headless Chrome. 70 MB binary versus 300+ MB Chromium install. 85 ms page loads versus around 500 ms. Instant startup versus roughly 2 seconds. One curl-and-tar deploy with no Node, no Chromium download, no system dependencies.

The actual interesting bit isn't the speed though. It's that stealth ships in the binary, not as a plugin you maintain forever. Per-session randomization of GPU, canvas, audio, and battery fingerprints. event.isTrusted = true on dispatched events, which is the specific thing most stealth-Chromium forks fake poorly and which detection scripts actually look at. navigator.webdriver returning undefined. A 3,520-domain tracker blocklist baked in.

It's shown as "the headless browser for AI agents," which honestly is doing a lot of work the repo hasn't earned yet (no MCP server, no Claude binding, no browser-use bindings anywhere in there). But every property it ships is exactly what an agent driving 50 parallel sessions inside a container would actually want, so the marketing skipped the implementation step and the substrate happens to fit anyway.

Worth a look:

https://github.com/h4ckf0r0day/obscura

13 comments