Skip to main content

​
Overview

Magnitude can control a browser visually using screenshots to observe the page and coordinate-based actions to interact with it. Common uses include:
  • Testing web UI behavior
  • Verifying interface changes
  • Web scraping and scripted browsing tasks

​
Setup

The browser agent requires Chromium.
  • Run /browser-setup in Magnitude, or
  • Install manually with:
npx patchright install chromium

​
How it works

On each turn, the browser agent receives a fresh screenshot of the current page. It can then perform actions such as clicking, typing, scrolling, dragging, navigation, tab switching, or JavaScript evaluation. After actions, it waits for page stability before continuing so results are based on settled page state.

​
Supported models

The browser agent requires a visually grounded model. See Providers & Models for compatibility details.

​
Capabilities

  • Click
  • Double-click
  • Right-click
  • Type (including special keys)
  • Scroll
  • Drag
  • Navigate
  • Go back
  • Tab management
  • Screenshots
  • JavaScript evaluation