browser-use/browser-use-pi

A tiny TypeScript web agent, hill climbed on real browser evals.

What it solves

It provides a way for AI agents to interact with the web by allowing them to write and execute JavaScript to control a Chrome browser. This eliminates the need for manual browser automation scripts by letting the agent build its own helpers and navigate the web autonomously.

How it works

The system uses Pi Mono as a foundation, combining a persistent V8 REPL and raw Chrome DevTools Protocol (CDP) to control Chrome. The agent processes the browser's accessibility (AX) tree and screenshots to understand the page, writes JavaScript code to perform actions, and iterates until the task is completed.

Who it’s for

Developers who want to integrate browser automation capabilities into their TypeScript applications using a single SDK.

Highlights

  • Persistent Sessions: Supports saved logins and persistent workspaces.
  • Cloud Integration: Works with Browser Use Cloud, removing the need for local Chrome installations.
  • Programmability: Offers streaming, hooks, and typed results for better control over agent actions.
  • Observability: Includes interaction highlights, recordings, and GIF exports to visualize the agent's work.

Related

  • Project
    krypticmouse/DSRs

    A high-performance Rust rewrite of the DSPy framework that enables programmatic, automated optimization of LM-powered application pipelines.

  • Project
    MobiusQuant/Gendangzou-skill

    A specialized AI Agent skill that provides traceable A-share market sector data, policy analysis, and company/ETF mapping to enable verifiable financial research.

  • Project
    microsoft/debug-gym

    A text-based interactive debugging framework for Python programs that allows LLM agents to use tools like pdb and bash to investigate and fix bugs in code repositories.

  • Project
    microsoft/project-oagents

    An opinionated .NET framework built on top of Semantic Kernel and Orleans for creating and hosting event-driven AI agents.