Building Another Browser AI
I didn't set out to build a browser agent. But when Scrapezy users kept asking for more sophisticated features—complex navigation, anti-bot detection—it became clear that existing solutions weren't cutting it.
The main issue? I wanted to use Camoufox (a Firefox-based anti-detection browser), but most tools like Stagehand, Playwright, and Browser-Use rely on CDP (Chromium DevTools Protocol). Incompatible. So I built from scratch.
The result: a browser agent that works with any LLM model and any Playwright-compatible browser. The key was nailing the agent loop (observe → decide → act → repeat) and implementing LLM result caching to cut costs and boost speed.
The breakthrough moment? When it successfully scraped three of Product Hunt's top five launches.
Now I'm deciding: open-source it, build a standalone product with hosted capabilities, or integrate it into Scrapezy.
Just build things. You'll learn something every time.