Page Agent

Most browser agents drive a separate headless browser or an extension from outside the app. Page Agent runs in-page as JavaScript, reads the DOM as text, and acts without screenshots or a multimodal model. You bring your own LLM, including local ones, and can ship a SaaS copilot without rewriting the backend. Optional extension and MCP cover multi-page work. The useful shift is product-native agents inside the UI, not a robot clicking from another process.