AlibabaPage Agent is a JavaScript agent that runs inside a web page and operates its interface from natural-language instructions. Adding one script gives a page its own agent, with no browser extension, Python, or headless browser required. It reads and manipulates the page through a text representation of the DOM rather than screenshots, so it needs no multimodal model or special permissions.
It works with most mainstream models through an OpenAI-compatible endpoint, including locally deployed ones, and a free testing model is available for evaluation. The project targets client-side enhancement of a web app rather than server-side automation: an in-product AI copilot, filling multi-step forms in ERP, CRM, and admin systems from one sentence, and natural-language or voice access for accessibility. An optional Chrome extension extends tasks across multiple pages and tabs, and an MCP server in beta lets outside agent clients control the browser. Its DOM processing components and prompt are derived from Browser Use.
Features
- In-page agent: runs as page JavaScript with no extension, Python, or headless browser
- Text-based DOM manipulation: acts on the DOM as text, without screenshots or multimodal models
- Bring your own LLM: most mainstream models via an OpenAI-compatible endpoint, including local ones
- SaaS copilot: embed an assistant in an existing product without a backend rewrite
- Form filling: turn multi-click workflows into a single instruction
- Accessibility: natural-language and voice control of any web app
- Chrome extension: optional multi-page tasks across browser tabs
- MCP server (beta): control the browser from external agent clients
