Agent Browser Automation

•
9
•

Automates web browser interactions using accessibility tree snapshots and deterministic element selection optimized for AI agents.

Agent Browser is a specialized headless browser CLI designed to bridge the gap between AI models and web interfaces. By utilizing accessibility trees to generate stable 'refs' for elements, it allows AI agents to perform deterministic clicks, form fills, and navigation with higher reliability than traditional visual tools. It is particularly effective for complex single-page applications (SPAs), multi-step workflows requiring session isolation, and scenarios where state persistence like cookies and local storage must be managed across browser instances.

Key Features

01Deterministic element selection via accessibility tree snapshots and refs
02Comprehensive interaction suite including hover, drag-and-drop, and scroll
039 GitHub stars
04State persistence to save and load authentication cookies and storage
05Isolated session management for simultaneous multi-context browsing
06Granular network control for blocking ads or mocking API responses

Use Cases

01Automating multi-step web workflows and complex form submissions
02High-reliability web scraping and data extraction from dynamic SPAs
03Simultaneous multi-user testing and session-isolated browser automation