Most browser agents sit outside the page: a headless browser, a Python process, or a screenshot-driven multimodal loop that a site owner can't realistically ship to end users. Page Agent flips this — it runs entirely as in-page JavaScript that reads the live DOM as text, so the agent acting on your UI is the same code you already ship to the browser. No vision model, no special permissions, no separate runtime.
What Sets It Apart
- Text-based DOM, not screenshots — works with any text LLM, needs no multimodal model or elevated permissions, and stays cheap enough to run on the client.
- Zero-install integration — a single script tag or
npm install page-agent; no browser extension and no backend rewrite, so the copilot ships with your frontend. - Bring your own LLM — point it at any OpenAI-compatible endpoint (e.g. Qwen via DashScope), keeping cost, latency, and data flow under your control.
- Reuses browser-use internals — its DOM processing and prompts derive from browser-use, but retargeted for client-side enhancement rather than server-side scraping.
Great Fit If
Great fit if you want to add a natural-language copilot to a SaaS product in a few lines, collapse multi-click ERP/CRM/admin forms into one sentence, or make an existing web app reachable by voice and plain language. Look elsewhere if you need server-side automation, cross-site scraping, or a headless pipeline — Page Agent is explicitly for client-side enhancement of pages you control, not background automation. An optional Chrome extension and a beta MCP server extend it to multi-page and multi-tab tasks.