睦MU-AI Book a demo

Product

One toolset for AI
to read and operate a real computer.

Computer use lets AI work through a user interface. Mu-AI can use the screen, keyboard, mouse or interface controls to operate existing websites, desktop apps and phones. Available actions depend on the environment.

47 tools in 8 families

Content reviewed on

7ObserveTake screenshots, read the screen, find targets, wait until the screen is readycapture_screen · read_screen · locate · crop · wait_for
10Point and clickClick, drag and scroll by description or by coordinatesclick · drag · scroll · click_at
3KeyboardType text, press keys and shortcutstype_text · key · hotkey
9UI structureRead desktop app controls and operate them directlyui_tree · ui_find · ui_press · ui_set_value
6WebExact reads and writes in Chrome and Firefox tabsweb_eval · web_list_tabs · web_crop
8SystemWindows, clipboard, reading tables, display informationfocus_window · read_grid · clipboard_get
2MobileList and operate Android and iOS devicesmobile_list_devices · mobile_shell
2Diagnostics and element inventoryCheck the environment; list the clickable elements on screendiagnose · enumerate_elements

Machine vision

Mu looks at the screen the way a person does, so it can operate software that has no interface to connect to. With a vision model connected, AI can find "the Save button at the top right" from one description. Without a vision model, the UI structure, web and coordinate tools still work.

  1. On-premises
    Connect a compatible vision model running in your own data center.
  2. Mix and match
    Click targeting, screen description and region recognition can each use the model that suits it best.
  3. Managed
    If you would rather not host a model, use a managed vision model service.

Know if each action worked

An optional change feed reports what happened on screen after each action: the page changed, focus moved, a field value changed, a dialog opened. The AI can confirm an action took effect without taking another screenshot.

Connect through MCP

Mu uses a standard protocol. 15 AI clients can use it directly, and you are not tied to one model vendor.

Common questions

Do we need a vision model?
Not always. The UI structure and web tools work without one. You need a vision model only to find targets by description or to read the screen.
Does data leave our organization?
Mu runs on the local machine. Whether screens leave your network depends on the model you connect: a cloud model receives screen content, while a vision model in your own data center keeps it inside.
Common questions

Want to see these tools
operate your systems?

翔耀實業股份有限公司 × 賀舟有限公司