Skip to main content

​
Browser Viewer

The Browser Viewer is a browser-in-browser pane embedded in the Chat page’s session panel. It renders live screenshots from a Playwright Chromium instance running inside the agent container, giving real-time visibility into what an agent sees and does on the web.

​
Interaction Modes

​
Take Control

Direct interaction with the remote browser. All input is forwarded to the Playwright instance:
  • Click — click through to elements on the page.
  • Scroll — mouse wheel events forwarded as-is.
  • Keyboard — keystrokes sent directly to Playwright.

​
Describe

Guided interaction for instructing the agent about a specific element:
  1. Click an element to select it.
  2. A floating popover appears showing element details (tag, id, classes, CSS selector, raw HTML).
  3. Type a natural-language description of what the agent should do with that element.
  4. Send the description to the chat thread for the agent to act on.

​
Screenshot Refresh

Screenshots refresh automatically after every action (click, navigate, scroll, keypress, type). No manual refresh is required.

​
Proxy Chain

All interactions are proxied through the full service stack:
The UI never communicates directly with the agent container.

​
Agentbox Endpoints