AI Agent Scraper

Report Abuse

Basic Information

BrowserPilot is an AI-powered, agentic browser designed to automate web navigation and robustly extract data from arbitrary websites using vision-based reasoning. It is intended to replace fragile, selector-based scrapers by letting you describe tasks in plain English and letting the system figure out navigation, interaction and extraction. The project combines a FastAPI backend, Playwright browser automation and Google Gemini vision to analyze pages, detect anti-bot measures, solve CAPTCHAs, rotate proxies and output results in common formats. The README documents installation steps, required environment variables including a Gemini API key and optional proxy configuration, and shows example prompts like exporting Amazon search results to CSV or saving Hacker News top stories as JSON. The app can be run with uvicorn and includes a live browser stream and a dashboard for monitoring sessions and proxy health.

Links

Categorization

App Details

Features
Vision-driven page analysis powered by Google Gemini to identify page types and relevant elements. Smart browser controller that detects anti-bot measures, rotates proxies, restarts browsers when blocked and handles logins and infinite scroll automatically. CAPTCHA recognition and solving integrated with the vision model. Universal extractor that produces outputs in PDF, CSV or JSON and preserves timestamps and metadata. Real-time session streaming with remote click and type control and multi-user viewing for debugging. Proxy management with health tracking, performance-based selection, site-specific blocking lists and automatic failure recovery. Built on FastAPI with Playwright for browser automation and configurable via environment variables and a .env file.
Use Cases
BrowserPilot reduces the effort and brittleness of traditional scrapers by using AI vision and natural language instructions so non-experts can request data extraction without writing brittle selectors. It handles common scraping roadblocks such as Cloudflare challenges, rate limits, CAPTCHAs and layout changes by switching proxies, solving challenges and adapting interactions. Outputs can be requested in familiar formats (JSON, CSV, PDF) and include structured metadata for downstream use. Live streaming and remote control make it easier to debug sessions or demonstrate workflows. The proxy dashboard and statistics help maintain reliable scraping at scale and track AI token usage and session performance.

Please fill the required fields*