alibaba/page-agent

JavaScript in-page GUI agent. Control web interfaces with natural language.

What it solves

Page Agent provides a way to embed an AI agent directly into a web page using a single script. It eliminates the need for complex server-side automation, headless browsers, or browser extensions for simple in-page interactions, allowing developers to turn web pages into interactive AI-driven interfaces.

How it works

Unlike many web agents that rely on screenshots and multimodal LLMs, Page Agent uses text-based DOM manipulation. It runs as in-page JavaScript, executing actions on the web page based on natural language commands. It is compatible with most mainstream LLMs, including locally deployed models, and can be extended via an optional Chrome extension for tasks that span multiple pages.

Who it’s for

  • SaaS developers wanting to add an AI copilot to their product without rewriting their backend.
  • Administrators of ERP, CRM, and admin systems looking to automate complex, multi-click workflows (e.g., smart form filling).
  • Accessibility experts aiming to make web apps accessible via natural language and voice commands.
  • Developers building browser-based agents that need to control the browser via an MCP server.

Highlights

  • Easy integration: Integrated via a single script tag or NPM package.
  • Text-based interaction: Operates on the DOM directly, removing the need for multimodal LLMs or special permissions.
  • LLM Agnostic: Supports various mainstream and local LLMs.
  • Extension support: Optional Chrome extension for multi-page tasks and an MCP server for external control.