I built a way for Grok Build to control a real Mac...open apps, click UI, type, and move the mouse just like you would
And I built the whole thing with Grok Build itself and optimised it to make the computer control as fast and efficient as possible
Most computer-use agents rely heavily on:
Screenshot → vision → figure out where to click → screenshot again → repeat
that works but in my experience they're not really practical for everyday tasks....as they are slow and can take forever to complete even basic steps
So I asked Grok Build to take a much faster approach with macOS Accessibility first to directly understand buttons, fields, menus and other UI elements....then only falls back to screenshots when they're actually needed
Previously Grok Build could execute shell commands on your Mac
Now, through this harness, it can actually interact with the Mac UI too
Shell → Accessibility → mouse + keyboard → screenshots when needed
Basically giving Grok Build hands for your Mac
Open source: