Page-agent

Page-agent is an open-source JavaScript library that turns any website into an AI-native application. Developers add a single script tag — no Python server, no headless browser, no infrastructure — and suddenly users can control the page by typing commands like "Fill the expense report for Friday's trip" or "Click the submit button." The agent reads the DOM, understands the page layout, and performs the clicks, typing, and navigation automatically. It runs entirely in the user's browser, which means data never leaves the client, and the agent asks for confirmation before taking actions through a built-in collaborative panel.

What makes Page-agent stand out is its simplicity and flexibility. It supports a wide range of LLMs — OpenAI, Claude, DeepSeek, Qwen, Gemini, Grok, and even fully offline models via Ollama or LM Studio. The library is MIT-licensed, so there are no licensing fees or vendor lock-in. For developers who need multi-page workflows, an optional Chrome extension lets the agent control multiple tabs and popups, and it can be connected to external desktop agents like Claude Desktop through an MCP server. But the core library works without any extension at all.

Page-agent is ideal for SaaS teams that want to ship an AI copilot into their product without a months-long backend project. It is especially useful for complex enterprise interfaces — ERP systems, CRM modules, admin dashboards — where users regularly navigate multi-step forms and repetitive tasks. The accessibility angle is also strong: voice or screen-reader users can interact with any web app through natural language, removing barriers in standard point-and-click UIs.