Agent

The execution layer. Real mouse, keyboard, and app control on your OS — not a browser sandbox.

Supported actions

open

Launch native apps — Chrome, Excel, Figma, Calendar

click

Mouse control with screen awareness before every click

type

Keyboard input, shortcuts, form fields

read

OCR and DOM parsing to understand what's on screen

navigate

URLs, tabs, windows, file paths

wait

Pause until UI state matches before continuing

✓ open Chrome

✓ navigate maps.google.com

> click "La Piazza" · Reserve▋

type party of 2 · 8:00 PM

open Calendar.app

Give one goal — the agent chains actions until the workflow is done. It reads the screen after each step and adapts if the UI changes.

Every action is logged and replayable through the Audit layer.