page-agent

To add natural language UI control to a web application, you embed a TypeScript-based GUI agent that reads the DOM as text and executes user instructions via LLM endpoints. You can integrate it via a single script tag or npm package.

Yes, you can use an AI agent to automate legacy web systems without frontend rewrites. The agent reads the existing DOM as text and translates natural language commands into direct DOM manipulation, modernizing accessibility without code changes.

Yes, running a browser-based GUI agent requires Node 22.13 or higher. You also need an OpenAI-compatible API provider configured for client-side execution to process the natural language instructions and drive the DOM interactions.

The best way to integrate an AI copilot into a SaaS admin panel is initializing a PageAgent class with an OpenAI-compatible LLM endpoint. This allows end-users to trigger actions like clicking or filling forms using simple text commands.

A web agent reads and interacts with the DOM by converting the page structure into text for the LLM to understand. The LLM then generates specific manipulation commands based on natural language user instructions, executing them directly on the web page.

Limitations when using natural language commands for DOM manipulation include relying entirely on the LLM's interpretation of the text-based DOM and requiring a stable OpenAI-compatible API endpoint for client-side execution to function correctly.