LLM-powered scraper with a cyberpunk skin and Tor support

What it does CyberScraper 2077 is a Python Streamlit app that fetches web pages via Playwright, then passes the content to OpenAI, Gemini, or a local Ollama model for structured extraction. You type a URL and a natural-language request; it returns JSON, CSV, Excel, or SQL and can push results directly to Google Sheets. It also handles pagination ranges and can route requests through Tor for .onion sites.

The interesting bit The project treats the LLM as the parsing layer rather than writing brittle CSS selectors or XPath queries. A “current browser” mode reuses your local browser instance to slip past bot detection, and there’s a caching layer to avoid redundant API calls when you iterate on the same page.

Key highlights

• Supports OpenAI, Gemini, and Ollama backends; the author recommends the commercial APIs for reliability
• Exports to JSON, CSV, HTML, SQL, and Excel with one-click Google Sheets upload
• Tor network support with automatic .onion detection and circuit isolation
• Multi-page scraping in beta, accepting range syntax like https://example.com/page 1-5,7,9-12
• Stealth mode and local-browser reuse for anti-bot evasion
• LRU and custom dictionary caching to reduce LLM API costs

Caveats

• The author explicitly warns that open-source LLMs via Ollama need prompt tuning and extra filtering, and performance depends heavily on local hardware
• Captcha bypass only works natively, not under Docker
• Multi-page scraping is beta and may behave unexpectedly
• Windows users are told to use Docker because the author won’t maintain a native version

Verdict Worth a look if you want a low-code scraping frontend and are comfortable paying for OpenAI or Gemini tokens. Skip it if you need a headless, fully automated pipeline; the Streamlit GUI and LLM dependency make this a manual, interactive tool rather than a cron-friendly scraper.