Skip to content
/ parser Public

Monte Carlo Tree Search (MCTS) adjacent DOM for computer-use agents

Notifications You must be signed in to change notification settings

LuxenAI/parser

Repository files navigation

Parser

State-of-the-art Monte Carlo Tree Search (MCTS) for computer-use agents, built around the browser DOM as the primary planning interface.

Why this project exists

Most computer-use agents still plan from pixels first and structure second. That makes long-horizon reasoning brittle, expensive, and hard to debug.

This project takes a DOM-first approach:

  • Use structured page state (DOM tree + metadata + interaction history) as the search state.
  • Use MCTS to plan multi-step action sequences before executing risky clicks or form edits.
  • Keep the agent grounded in real browser affordances (click, type, select, scroll, navigate), not just token-level guesses.

About

Monte Carlo Tree Search (MCTS) adjacent DOM for computer-use agents

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •