Computer Use

The Desktop app’s Computer Use mode gives agents full control of your desktop — they can see your screen, move the mouse, type, click, and interact with any application.

How It Works

Agent takes a screenshot of your screen
Screenshot sent to vision LLM (Claude with computer-use)
LLM analyzes the screen and decides the next action
Agent executes the action (click, type, scroll, hotkey)
Repeat until task complete

Actions

Action	Description
`click`	Move mouse + click at coordinates
`double_click`	Double-click at coordinates
`type`	Type text (with natural rhythm)
`press`	Press a single key (Enter, Tab, etc.)
`hotkey`	Key combination (Ctrl+C, Cmd+V, etc.)
`scroll`	Smooth scroll up/down
`screenshot`	Capture current screen state
`wait`	Pause for a duration

Human-Like Behavior

All desktop actions use human-like patterns:

Mouse — Bezier curves with acceleration/deceleration
Typing — Variable speed, faster for common letter pairs
Clicks — Small position jitter (humans don’t click exact pixels)
Pauses — Random micro-delays between actions

Safety

Agent always asks before risky actions (installing software, deleting files)
You can pause/stop at any time
Take Over mode lets you grab control instantly
All actions logged in the timeline