Vibe Hacking + Open-Source Rover: reverse-engineer site APIs in your browser. Every website becomes a callable agent surface.
|
|
|
New release
Vibe Hacking is here.
Reverse-engineer site APIs from inside your browser.
Turn any web interface agentic with open-source Rover.
Every website becomes a callable agent surface.
|
Websites are just API wrappers
The agent acts on the page, uncovers network calls, and generates reusable extraction scripts at scale.
The hard part of raw HTTP scraping has always been reverse-engineering the endpoints and recreating auth — this release does both directly in your browser.
|
|
|
Demo results
We used Vibe Hacking to scrape the X profiles that OpenClaw founder Peter Steinberger follows — despite X's UI limits to 50 profiles.
Check out the task results →
|
|
|
Why this matters
|
•
|
HTTP scraping is the fastest & cheapest method. Instead of a thousand page actions, the agent generates code for 1,000 API calls and runs them in seconds for pennies.
|
|
•
|
Raw HTTP scraping just got democratized. No more manually reverse-engineering endpoints and auth flows. The browser session gives the agent the right context to do it from within the page.
|
|
•
|
Browser context = strongest anti-bot footing. Headers and authorization propagate automatically — no rebuilding the site's auth stack by hand.
|
|
|
|
Open source
Rover harness is now Open Sourced
Rover turns any web interface agentic. Drop it into your own stack or pair it with our cloud browser for cross-domain workflows.
|
•
|
Browser automation — inject from Playwright/Puppeteer to handle messy agentic actions
|
|
•
|
Extensions & Electron — embed agentic capabilities in any Chromium-based app
|
|
•
|
Cloud browser — use our infrastructure when workflows need cross-domain interactions
|
In the demo — Rover looked up a menu URL in a cloud browser and then updates menu items in Grab. That's the kind of cross-surface workflow we want Rover to make easy.
|
|
|
Where Rover is going
Accessibility for Humans & Agents
|
doordash.com/rover?q=checkout-with-burrito
|
|
•
|
Agent accessibility — agent just opens URL with Rover params to transact/execute/apply
|
|
•
|
WebMCP accessibility — auto expose sites for WebMCP with controls for abuse and preventing disintermediation
|
|
•
|
Voice accessibility — every Rover-enabled site is automatically accessibility compliant
|
|
|
|
Free daily-driving
Upgraded to Gemini 3.1 Flash Lite
|
•
|
Confirmed as the default model. We've validated that 3.1 Flash Lite is good enough for daily-driving agentic tasks — performance holds across the workflows that matter.
|
|
•
|
We always believed smaller models would get here. 500 requests/day free on the AI Studio tier — enough to daily-drive a web agent without spending a penny.
|
|
|
What's next
|
Reusable scripts + cloud scale
Save network scripts as reusable tools; just call network tool to post to Twitter with zero LLM inference.
|
|
|
Code-first vibe scraping
Hybrid network + DOM scripts for 1,000× extraction scale and efficiency.
|
|
|
Rover partnerships
Enterprise & website-builder integrations in progress, so you will see more of Rover across the web.
|
|
|
BrightData enrichment
Post-scrape, the agent recommends the best datasets to join onto your data.
|
|
|
Try the release
Read the launch post, star Rover on GitHub, and start using the cloud product or extension with Flash Lite by default.
Happy shipping,
Arjun & Bhavani — rtrvr.ai
|
|
