BrowserPilot is an open-source, vision-powered agentic browser focused on real-world scraping. You describe tasks in natural language (“visit site -> grab prices -> export CSV”), and it navigates like a human, handles anti-bot, and exports clean data.

What it does:

• Natural-language tasks -> actions (navigate, click, type, scroll)

• Anti-bot detection, proxy rotation, and CAPTCHA solving

• Exports in PDF/CSV/JSON (with timestamps + metadata)

• Works across arbitrary sites (Amazon, LinkedIn, blogs, etc.)

• Live session streaming; you can step in and control

How it works (high level)

• Playwright controls the browser.

• Gemini Vision analyzes page state and suggests interactions.

• FastAPI backend orchestrates tasks

• Proxy pool with health checks and rotation on blocks

• Universal extractor shapes data for PDF/CSV/JSON outputs

Why it’s different

• Open-source alternative to Perplexity Comet, director.ai, and Firecrawl

• Vision-first control reduces breakage from layout changes

• Built to handle protected sites with anti-bot + proxies

• Self-hostable, focused on low-maintenance scraping workflows

Would love your feedback!