Page Agent
In-page JavaScript agent that operates web interfaces from text commands
What it is
A TypeScript library you drop into a web page so an LLM can read the DOM as text and click, type, and fill forms in response to plain-language instructions. It needs no screenshots, headless browser, or Python, and works with most OpenAI-compatible models, including local ones. An optional Chrome extension handles multi-page tasks and a beta MCP server lets outside agents drive the browser. Web developers use it to add an AI copilot or natural-language form filling to SaaS, admin, and CRM-style apps without changing their backend.