Skip to main content
Project Intelligence

Collaborative browser

The collaborative browser opens a real browser (headless Chromium on your host) in the working area: the page is shown exactly as rendered — with images, tables, and interactive elements. The agent and you work in one live session: your clicks and input are visible to the agent, and its actions are visible to you.

What the agent can do​

  • Open and navigate. Opens a URL, navigates back/forward, and reloads the page.
  • Read. Gets the page content in readable form (Markdown) and the interface structure — the agent reasons and answers based on them.
  • Act. Clicks, types, selects values, hovers — only on elements it sees in the current page structure.

Human control: the "AI edit" toggle​

The changes the agent makes to the page (clicks, typing) are allowed only when two conditions hold: the page is open for you in the working area (you are watching) and the "AI edit" toggle is on.

SituationAgent actions on the page
You are not watching the pageRead only: open, navigate, analyze content
You are watching, "AI edit" offSame as the row above
You are watching, "AI edit" onFull set: clicks, typing, selects

Your own actions are never restricted: you can take over control at any moment, fill a form, or close the site.

Profiles and security​

  • A profile per user: cookies, logins, and site state are stored in each user's browser profile on your host — one user cannot see another's sessions.
  • One session per user: only one active browser session at a time; idle sessions are closed automatically.

When to use it​

  • Working with external portals: a client, a vendor, government services — the agent fills forms and collects data while you watch and confirm.
  • Checking and monitoring: the agent opens a page and explains what is on it; you see exactly what it sees — no "trust me."
  • Research: together with the external research tool (see Navigator), the agent can open found sources and break down their content with you.