page-agent

Why we included this project

Page Agent is a single script that gives any web page its own natural-language agent. You can tell it to click a button or fill in a form, and it carries out the action in the browser. It needs no browser extension or headless browser. It reads the page's DOM as text rather than taking screenshots, so you can drive it with your own LLM, including models you host locally. That makes it a good fit for product teams that want to add an AI copilot to a SaaS app without rewriting the backend, and for anyone tired of clicking through long forms in ERP, CRM, and other admin systems. Because it is plain in-page JavaScript, it also helps with accessibility, letting people operate web apps by voice or through a screen reader. An optional Chrome extension and an MCP server extend the same agent across tabs and let external agent clients control the browser.