That sounds great. How well does the base OS + "pet" containers work with all the crazy dependencies you need to do modern ML work, e.g. some exact combination of nvidia drivers + CUDA + torch + other random stuff? That's the pain point I'd be motivated enough to solve that I'd switch distros.
Your "pet" containers basically become your traditional OS, in a way. They use filesystem overlays, so your container can see all of the files on your system, plus it's own layered files, ie. each container has it's own "view" of the filesystem.
You can install anything inside your "pet" containers that you would normally install on your actual system. The container keeps everything tidy and self-contained. I have a container for development, another for music/DAW, another for certain games that needed weird deps.
Fedora Kinoite/Silverblue come with `Toolbx`[1] preinstalled, but I found `distrobox`[2] to be more flexible for my needs. I layered distrobox onto my base image, and it just works.
Many GUI apps are available via Flatpaks, and can be installed directly or via the Software Center. You can enable Flathub[3] as a source, so there's a ton of available software, including Steam, Chrome, Firefox, Discord, Spotify and more. Flatpaks are also sandboxed and self-contained, so they can't pollute/break your system either.
