I've been working on browser automation infrastructure for a while and kept hitting the same ceiling: Playwright and Puppeteer are great for scripting but fall apart when you need stealth at scale. Cloud-based solutions like Browserbase work but you're sending all your data to a third party and can't control the environment.

So we built Owl Browser, a self-hosted Chromium-based engine that runs in a single Docker container. The core technical problem we tried to solve was fingerprint consistency. Most browser automation tools either ignore fingerprinting entirely or apply surface-level patches (spoofing navigator.userAgent, etc.) that fail because the underlying hardware signals remain inconsistent. We went deeper: GPU renderer strings, canvas noise, AudioContext fingerprints, WebGL parameters, font metrics, and timing side-channels all need to be internally consistent for a profile to pass detection.

Current numbers: 256 parallel sessions per instance, sub-12ms cold start, 157 MCP tools, RFC 9421 HTTP Message Signatures with Ed25519 for Cloudflare Verified Bot enrollment.

For March we're running free stress tests: 100 concurrent real browser sessions against any web app, you get a breakdown of where things fail. Different from k6 or Locust because those send raw HTTP requests and skip everything that runs client-side. Signup at owlbrowser.net.

Happy to go deep on the fingerprinting approach or the MCP integration if anyone's curious.