BrowserPilot: Tell your browser what to do. It actually does it.
An AI-powered, vision-first browser that navigates any website, handles anti-bot walls, solves CAPTCHAs, and delivers data in your format — PDF, CSV, or JSON.
See it in action
A quick demo showing navigation, extraction, and export flow.
Powerful by design, resilient in practice
From vision-first interaction to anti-bot evasion and universal export formats, every piece is battle-tested for the real web.
It Actually Sees Websites
Vision model understands layouts like a human — resilient to redesigns and anti-bot tricks.
Anti‑Bot Ninja
Detects Cloudflare and rate limits, rotates proxies, solves CAPTCHAs, and adapts on the fly.
Agentic Control
Clicks, types, scrolls, navigates. Like a tireless intern who never makes mistakes.
Recovery & Resilience
Restarts browsers, swaps identities, and continues tasks even when sites fight back.
Any Website
Amazon, LinkedIn, random blogs — even weird edge cases. If humans can do it, so can it.
Live Streaming
Watch the browser session in real time. Step in to click or type if needed.
JSON
Clean, structured output with timestamps and metadata.
CSV
Perfectly shaped tables for analysis and reporting.
Beautifully rendered documents for sharing or archiving.
Real examples you can run today
Speak in plain English. The agent figures out the rest — reliably.
“Go to Hacker News and save the top stories as JSON”
“Search Amazon for wireless headphones under $100 and export the results to CSV”
“Go to LinkedIn, find AI engineers in San Francisco, and save their profiles”
“Visit this random e‑commerce site and grab all the product prices”
The cool technical stuff
Built for resilience and clarity — from dynamic vision to robust proxy rotation.
Smart Format Detection
Say “save as PDF / CSV / JSON” — output is shaped perfectly with metadata.
Anti‑Bot Mode
Spots challenges early, switches proxies, solves CAPTCHAs, and backs off gracefully.
Proxy Management
Health tracking, performance‑based selection, site‑specific block lists.
Universal Extractor
AI‑powered extraction that organizes content with timestamps and structure.
Vision Integration
Understands dynamic layouts, anti‑bot patterns, and interaction targets.
Meaningful Dashboard
Live sessions, proxy performance, and token spend that actually helps.
Built with Open Source + Gemini
Designed for low-cost maintenance: open-source building blocks on the backend with a vision model for robust, layout‑agnostic interaction. Minimal glue, maximum leverage.
Browser automation, backend framework, and UI all rely on permissive OSS libraries.
Gemini looks at pages like a human, so layout changes don't break your flows.
Fewer brittle selectors, less maintenance. Proxies and anti‑bot handled centrally.
How BrowserPilot compares
A practical look at scraping and automation tradeoffs. The goal: resilient automation with minimal upkeep.
| Feature | Traditional scraping | director.ai | Perplexity Comet | BrowserPilot |
|---|---|---|---|---|
| Open Source | ||||
| Natural-language tasks | ||||
| Vision-based interaction | ||||
| Anti-bot detection | ||||
| Proxy rotation / identity | manual | |||
| Scrape protected sites | difficult | limited | ||
| Export formats (PDF/CSV/JSON) | manual | |||
| Live session streaming | ||||
| Self-hostable |
Notes: Feature mapping is based on our understanding of public information and typical usage. Availability and capabilities may change.
Vision: an open‑source Comet‑style browser focused on scraping
We're building BrowserPilot as the open‑source alternative to Comet‑like browsers — with a special emphasis on reliable scraping. Instead of brittle selectors and manual parsing, describe the task in natural language and let the agent handle the rest.
Say what you want — “scrape prices from this site and export to CSV” — and the agent executes.
Anti‑bot detection, CAPTCHAs, and proxy rotation keep tasks running with anonymity.
Export as PDF, CSV, or JSON. We add timestamps and metadata automatically.
Gemini sees the page like a human, so redesigns break less often than CSS/XPath scrapers.
Stream the session live, step in when needed, and compose repeatable workflows.
Built on OSS libraries and a pay‑as‑you‑go model for low maintenance burden.
Getting started — it's actually easy
From cloning to running your first task in minutes. Configure proxies if you plan to scrape heavily.
python --version # 3.8 or newergit clone https://github.com/ai-naymul/AI-Agent-Scraper.git
cd ai-agentic-browsercurl -LsSf https://astral.sh/uv/install.sh | sh
uv pip install -r requirements.txt# .env (gitignored)
GOOGLE_API_KEY=your_actual_api_key_here
SCRAPER_PROXIES=[{"server":"http://proxy1:port","username":"user","password":"pass"}]python -m uvicorn backend.main:app --reload
# Open http://localhost:8000{
"SCRAPER_PROXIES": [
{ "server": "http://proxy1.example.com:8080", "username": "user1", "password": "pass1", "location": "US" },
{ "server": "http://proxy2.example.com:8080", "username": "user2", "password": "pass2", "location": "EU" }
]
}