← Back to projects

OpenBrowser AI

AI browser agent that learns from exploration — turns one-time actions into reusable, self-healing automation assets.

PythonLangGraphCDPAI AgentGitHub ↗

Overview

OpenBrowser AI is an AI-powered browser automation framework. Give it a natural language task, and it perceives, plans, and executes actions in a real browser — clicking, typing, navigating, and extracting data autonomously.

Workflow Assets

The system transforms ephemeral agent explorations into permanent, reusable assets. First run records the full workflow; future runs replay it 10-100x faster with zero LLM calls. When websites change, it self-heals.

Key Features

  • Perceive → Plan → Execute loop with vision (screenshots + bounding boxes)
  • Dual asset generation — UI workflows + direct HTTP crawlers
  • Self-healing — AI-powered local repair when steps break
  • 12+ LLM providers — OpenAI, Anthropic, Google, Groq, Ollama, and more
  • MCP integration — works inside Claude Desktop

Tech Stack

  • Python 3.12+ with LangGraph orchestration
  • Chrome DevTools Protocol for browser control
  • FastAPI + Next.js web UI with real-time streaming