For example, you could run VS Code on that machine as a window on your Mac. A more real world example is people accessing guis (e.g. matlab) on lab clusters.

The closest set up for x11 would be to use x11 forwarding with xpra.

1. I'd really like to run my development environment for things under Siri for its tiling window management but for better or worse I'm deeply embedded in the Apple ecosystem for everything else, this looks like it could be a really nice way of doing it (possibly once multi monitor support is in).

2. There are still a few applications which have supported Linux builds but no support for macOS (Iridium's Niagara Workbench application for configuration of building management systems springs to mind here). Since Apple ended support for Quartz this has been a bit of a pain to deal with.

But in the field of integrated circuit design there’s lots of apps that are Linux-only. I’ve tried to run some of them in containers on Mac. But XQuartz is awful.

If they ever transitioned to Wayland perhaps this would let us run these apps on Mac in a nice way.

On the other hand some of them have started getting ARM builds (for running simulations on certain cloud environments) so maybe native Mac GUI builds could happen someday soon.

For example, you may not want to run some graphical applications directly on your Mac for security, isolation or testing purposes.

If this software turns out to be lower latency than RDP and CRD, I could also see it being very useful for accessing a remote graphical workstation (e.g.: running heavy software on an beefy machine in a data center instead of taking up resources on my skinny laptop).

This is for the long tail. The compositor path dodges a pile of volunteer-port churn and runs the Linux build directly, which is a lot more appealing for niche GUI tools and dev apps that barely get maintained on Linux, never mind macOS.

Anything I want sandboxed or “grouped”.

Work on a project -> open the relevant container.

Similar to parallels window integration mode.

It’s all from shortcomings to have a hierarchical view on your data and applications.

Goal: isolation. Security-wise, and focus-wise

And also this chart is super weird:

    Solution  Latency  HiDPI  Native Integration  Setup Complexity
    Cocoa-Way   Low     Native windows   Easy
    XQuartz   High   Partial   X11 quirks   Medium
    VNC   High     Full screen   Medium
    VM GUI   High   Partial   Separate window   Complex

I also looked at some of the code and it's using OpenGL 3.3 Core which is... super old. For a project like this they should be targetting OpenGL 4.5 Core. But it makes sense in the context of this being LLM-generated since most of its training data is probably OpenGL 3.3 Core code....

Overall this project is very strange and it makes me think AI isn't all that great.

Here some history on how the command key came to be https://www.folklore.org/Swedish_Campground.html https://en.wikipedia.org/wiki/Command_key