Welcome
Unbrowse is the internet-use layer for AI agents.
It turns intent-level requests into reliable, auditable web actions.
It does this by learning browser-authenticated flows once, then reusing them as versioned skills.
What you’ll learn
This documentation covers:
What Unbrowse is and why it exists
How skills are captured from real browsing (HAR)
How discovery and routing pick the best skill for an intent
How execution preserves cookies, CSRF, and login state
How to contribute as an indexer
How to integrate Unbrowse into agents and apps
Quick navigation
New to Unbrowse
Start with What is Unbrowse?.
Indexers
Go to Getting Started.
Developers
Go to Quick Start.
Technology
Read How It Works (Simple).
The Problem We Solve
The internet was built for humans, not for AI. Today's AI agents are bottlenecked by:
Slow execution (5-60 seconds per action)
Poor reliability (70-85% success rates)
High operating costs
Limited coverage (APIs exist for only ~1% of websites)
The Unbrowse Solution
Unbrowse operates at the request level, not the pixel level.
It uses real browser sessions for capture and parity.
It can execute requests in a browser context when a site is strict about auth.
Up to 50x faster (<2 second per action)
90%+ reliability
Significantly cheaper than GUI automation
Universal coverage (works on most websites, even without a public API)
Key Features
✅ Traverse once, reuse everywhere - one capture becomes a reusable skill
✅ Browser-parity execution - replay inside a browser context when needed
✅ Crowdsourced ability index - Learn from billions of human browsing patterns
✅ Natural language interface - Agents use simple requests, we handle the complexity
✅ Live web data - Real-time information, not cached or stale data
✅ Versioned + verifiable skills - auditability, rollback, and reliability scoring
Get Started
Ready to dive in? Choose your path:
Questions? Check out Common Questions or reach out to us at lewis@getfoundry.app
Last updated