Agent Browser is an open-source automation framework for LLM-based agents that enables programmatic interaction with web interfaces through structured DOM navigation and action execution.

• Converts complex HTML structures into simplified markdown or JSON representations to optimize LLM context window utilization.
• Executes browser actions including clicks, text input, and scroll events via the Chrome DevTools Protocol (CDP) or Playwright.
• Implements visual grounding techniques to map coordinate-based clicks to specific DOM elements for high-precision navigation.
• Provides built-in session management to maintain cookies, local storage, and authentication states across multi-step agent workflows.
• Supports integration with OpenAI, Anthropic, and local models via LangChain or direct API calls.
• Runs in Dockerized environments with headless Chromium support for scalable background processing.

Ideal for building automated web research tools, competitive price monitoring systems, and cross-platform data extraction pipelines.