I built a reverse-engineering agent for the web
Vision LLM annotates browsing sessions to generate requests-based scripts.
A tool that watches you browse, then writes HTTP-based automation scripts
Browse once, get requests-based scripts—no Playwright or Selenium dependencies needed.
Developers and data engineers building web scrapers and automation scripts
Browser-use · Playwright · Selenium
This makes it a lot easier to create/fix scripts, all we have to is browse the site. This is not a completely autonomous agent, you can pause, talk, guide the agent on its RE path, just like a normal agent.
I am aware of the limitations of this approach, but from my own research(please refer the blog for details) up to 60%-90% of websites can be automated/scraped with just requests.
I was able to run this against some popular websites like, bookmyshow, makemytrip. And to test it in the open, it one-shotted scraper solutions for 2 problems posted on r/webscraping.
It supports local models, cloud hosted models, and proprietary models.
Github: https://github.com/StoneSteel27/AutomatiQ
Blog for verbose explanation and techincal implementation: https://shiny-sushi-116.notion.site/automatiq
Demo video of using Automatiq on bookmyshow website to get current running movies, showtimes, seat layout of a particular screen to create an interactive script: https://www.youtube.com/watch?v=OfMvAQP5pkM
Thanks for your time, hope you have a great day/night ahead ;)
Vision LLM annotates browsing sessions to generate requests-based scripts.
Outputs requests-based scripts instead of browser automation—10-100x faster than Selenium.
Network-layer API discovery beats brittle DOM scraping that Firecrawl uses.
Requests-based automation beats browser overhead when the AI can reverse-engineer APIs.
Wire-protocol parsing blocks dangerous commands without needing agents on endpoints.
Tamper-evident event journals with cryptographic chains solve AI audit requirements.