December 2025 | Paul Hanchett
Not Another Browser Automation Tool
Most browser automation tools—Puppeteer, Selenium, Playwright—focus on one thing: controlling the browser. They’re built for testing and scripting, treating the browser as a machine to be programmed. The AI Browser Extension (AIBE) takes a fundamentally different approach: observing human behavior.
Instead of asking “How do I make the browser click this button?”, AIBE asks “What did the user just do, and what can an AI learn from watching them?”
This isn’t just a philosophical difference—it’s a practical one. By capturing user interactions from their perspective (what they see, what they click, what they type), AIBE makes available training data that teaches AI systems to navigate the web the way humans do: reactively, contextually, and adaptively.
Of course, AIBE can control the browser too. But that’s not the point. The point is teaching AI to understand web interaction, not just execute scripts.
What’s in Version 1.0.0?
Observer Channel: Watch and Learn
AIBE captures every meaningful user interaction:
-
Mouse events with target labels and final values
-
Keyboard input as consolidated final text (not raw keystrokes)
-
Screen updates showing visible elements and their states
-
Navigation events tracking page loads and URL changes
Events are labeled with their source—user for human actions, ai for actions triggered through the Actor channel—enabling clean separation of training data from AI-generated actions.
Story Assembly: A Linguistic Model for Web Behavior
Raw events are organized into a hierarchical structure that reveals natural patterns in browsing behavior:
-
Words: All interactions with a single element
-
Sentences: All Words within one URL/page
-
Paragraphs: All Sentences within one domain
-
Stories: The complete journey in one browser tab
This structure enables AI systems to understand context without semantic analysis—a login form looks like a login form regardless of the website.
MongoDB Integration: Persistent Memory
Stories are automatically stored in MongoDB, if it is available, creating a searchable archive of user behavior patterns. Query by domain, time range, interaction type, or any structural element of the Story hierarchy.
Actor Channel: Close the Loop
AI systems can send commands back to the browser— to navigate, click, type— and to observe the results. This enables:
-
Validation of learned behaviors
-
Interactive training with feedback
-
Autonomous web agents (with appropriate safeguards)
Simple Installation
For complete details, see the .
pip install aibe-server
aibe-server start
The Chrome extension downloads directly from the running server. MongoDB is optional but recommended for persistent storage.
The Generic User Framework
AIBE implements what I call the “Generic User Framework”—treating AI as a user of web interfaces rather than a programmer of them. This means:
-
AI sees only what a human would see (visible text and controls)
-
AI can only do what a human could do (click, type, navigate)
-
AI learns from watching humans, not from reading documentation
This constraint is liberating. By limiting the AI’s perception and action space to match human capabilities, we make web navigation a tractable learning problem.
Get Started
-
Documentation:
-
Installation:
pip install aibe-server -
Source: Contact paul@paulhanchett.com for contributor access
AIBE 1.0.0 represents years of research into human-AI collaboration on the web. I’m excited to share it and see what others build with it.