Docker on Mac runs Linux containers inside a Linux VM - you can't run macOS in Docker. So if you need Claude / Codex / OpenCode to interact with:

- macOS GUI apps (Xcode, Numbers, Safari, etc.) - macOS desktop automation (screenshots, mouse/keyboard input, accessibility APIs) - macOS CI/CD (building iOS/macOS apps, running XCTest)

...you need an actual macOS VM, which is what Lume provides.

I wonder what the additional layer of virtualization changes with respect to this in a project like this one: https://github.com/dockur/macos

The unattended setup is a large improvement, which also begs the question: Mac OS doesn't have an unattended.xml alternative for its installer?

re: https://github.com/dockur/macos

A closer comparison here is Lumier, which provides a "Docker-like" interface to spin up VMs with a noVNC server: https://cua.ai/docs/lume/guide/advanced/lumier/docker

The key difference: dockur/macos uses QEMU+KVM, which only works on Linux hosts. It can't run on macOS hardware since Apple doesn't expose KVM. See: https://github.com/dockur/macos/issues/256

macOS has unattended setup options via MDM or Apple Configurator…

Can you do zero touch without having an Apple Business account (so, a DUNS number) and a MDM?

I thought this was a silly way to do it too, but upon reflection I don’t know if you can zero touch setup a Mac without registering a device in DEP.

re: unattended setup.

You're both right - Apple's official zero-touch setup requires MDM + DEP, which needs Apple Business Manager (and yes, a DUNS number).

But for VMs specifically, DEP doesn't work anyway - VMs don't have real serial numbers that can be enrolled in Device Enrollment Program.

VNC-based setup automation is the only practical option - it's what the ecosystem has converged on for macOS VMs. Lume connects to the VM's VNC server and programmatically tabs, clicks, types through Setup Assistant.

I wish the virtualization framework would allow you to simulate your own MDM stuff. Would be very useful for integration testing MDM implementations themselves...

If you run say ollama in docker vs native you lose MPS (metal core like the way CUDA works but Mac) which is really a deal breaker to do any AI work via docker for doing things with local models, which is my use case to have a full fast VM inside my mac.