# Page Agent

A JavaScript library that adds an in-page AI agent to any website, so a user can control web interfaces such as forms and multi-step workflows with natural language.

Page Agent turns a web page into an agent-controllable interface by adding a single script. It manipulates the page through text-based DOM access rather than screenshots, so it needs no browser extension, headless browser or multimodal model, and it works with mainstream LLMs as well as locally deployed ones.

Described use cases include adding a SaaS copilot without a backend rewrite, turning multi-click workflows such as ERP or CRM forms into a single sentence, making a web app accessible through natural language or voice commands, and, with an optional Chrome extension, controlling tasks across multiple browser tabs. An MCP server also lets external agent clients control the browser through Page Agent.

It installs from npm and is licensed under the MIT license.

Website: https://alibaba.github.io/page-agent/