I have to say this is nice packaging. It is weirdly not obvious how nice it is to cleanly use a different “machine” inside your desktop.
I never thought I’d prefer WSL to even my MacBook for working with remote servers and dev but somehow I do.
I of course know the many ways to roll something like this for myself, yes dev containers are better for many things etc. but it’s wierd how good the ergonomics of a WSL like container are.
Thanks for saying so! That's exactly why I wrote this. The UX of wsl2 is just right. There are a lot of other ways to accomplish this, but none touch that same experience.
I'm very curious about this - as a mise user this seems like a redundant extra level of abstraction that I can't figure out the benefits of (& WSL definitely comes with at least some downsides).
From the OP I see:
> my long journey to keep my host installation free from all the changing and breaking dev dependencies that force a reinstall every few months
mise does this 100% for me (assuming we don't count mise itself as one of those dev dependencies). I'm struggling to think of what falls outside of it?
i agree. you can quickly import distro images into named WSL VMs (distros) to keep things separated. and you get close-to-native performance (especially by disabling kernel mitigations)
Yeah, I can see why you'd want this. Being able to cobble it together with other tools is one thing. Having it ready to go without a bunch of setup is still a win.
Got a quick demo? Just opening an editor, running a dev server, and showing how you access the files would help me get a feel for it.
How does it differ from existing tools kinda do exactly same, like e.g. toolbx. (And to a slightly lesser degree flatpack, snap, etc.)?
And what prevented adopting/supporting existing projects, instead of further tool ecosystem fragmentation? (kinda the same question, just differently phrased)
And then I read your comment and my first reaction was, oh, they typoed toolbox. But, nope, Red Hat created toolbx which confusingly has the CLI binary of `toolbox`: https://containertoolbx.org
... it's not. Just another way to keep my host clean and do the dirty work somewhere else. Where did reproducible builds come through? Another comment?
Most of these tools you mention are designed to work best as an overlay to your $HOME. toolbx, distrobox, and to some extent flatpak and snap. NSL is built to give you the "pet" virtual machines with limited integration to the host - you can get to the host filesystem if you want, and you can use the host's Wayland session. I think a better comparison would be using NSL instead of using a remote computer or VM.
that repo is retired, now all the Frostyard atomic images build out of a single mkosi repo at https://github.com/frostyard/snosi. the website is https://frostyard.org - I'll take down the older snowlinux.org or redirect it. Thanks!
the README is probably 50/50 hand written and LLM written. We haven't really done much to publicize the org yet, we were waiting for bootc to stabilize on Debian. Tomorrow is actually retirement day for the non-bootc installation paths so we may do some clean up in the future. Don't really have a burning need to become the next hot distro though. If atomic debian desktops and servers sound good we're here.
I'm confused why VM + systemd-nspawn? From my understaing WSL 2 runs a single VM + something like systemd-nspawn per "linux installation", but it runs a VM because it needs linux kernel. Why not just do systemd-nspawn if you alread on linux?
Way better isolation, is my guess. Plus, you can use a different kernel this way.
I used to poo-poo when people said that containers aren't a _real_ security boundary, at least for personal stuff, and not a multi-tenant server. But I bet even mid-tier LLMs can break out of LXC/Docker/nspawn at this point.
That has to be the host kernel not the guest kernel. Old Nvidia systems without supported kernel drivers often lack IOMMU to forward PCIe memory, too. I wouldn't recommend using an outdated host kernel with 2026 AI-powered vulnerability scanners.
Windows (10 LTSC or 11 with dTPM) actually works better for old Nvidia systems with WSL. You even get CUDA libraries within WSL and security updates for the next 5 years.
CVEs for runc are much more frequent than CVEs for KVM. The attack surface area is bigger, and containers were never intended as a security boundary, but rather as a resource management tool.
it is quite likely that you can break out even with no linux containers related CVEs. --isolate does seem to fix that somehow but is explicit opt. in and "more painful to use" ... (which creates a UX challenge unlikely to end well from a security POV).
I was just testing this and it's not clear that it works out of the box. `krun` shows a different kernel than with `crun` but it doesn't reflect the dropped capabilities in the same way. I'm probably holding it wrong but I'm not sure what to look for at the moment.
"During a test conducted by Trail of Bits researcher Artem Dinaburg, a preview version of GPT 5.6-Cyber was tasked with breaking out of a Debian 12 virtual machine. Initially, the agent exploited a known Linux kernel vulnerability, CVE-2026-53359, by developing its own exploit. After the host was updated, the agent found another pathway through libslirp, chaining a known vulnerability (CVE-2026-9539) with a previously unassigned bug to gain arbitrary host memory access. Even after QEMU and libslirp were updated, the agent analyzed system components and constructed a new escape chain using three zero-day vulnerabilities and one KVM flaw that had not yet reached the distribution kernel.
These findings suggest that general-purpose VMs may not be adequate security boundaries for highly capable AI agents, especially in older systems with delayed security updates. Trail of Bits recommends using specialized isolation systems like Firecracker, restricting VM access, and implementing rapid patching to mitigate these risks."
Can confirm. My abliterated models kept breaking out of qemu VMs due to BIOS implementation quirks in qemu.
I'm using firecracker now with a very defensive systemd-as-separate-non-admin-user seccomp sandbox on top, which seems to hold them off long enough for me to see an agent going rogue and intervening.
Currently I still have hopes that eBPF sandboxing will help, but just a couple days ago my agent discovered a use after free bug in the ebpf kernel-side verifier... so there's that.
I was surprised I hadn't heard of this as a separate tool from systemd-nspawn. Then I realized why... the GitHub repo shows version 0.6 released in 2022, then version 1.0.0 last week, followed by a flurry of releases up to 1.8.0 yesterday.
So basically, it's been all Claude'd up extremely recently.
That said, the landing page, docs, and git README are much higher quality than I normally see out of LLM-generated projects, so at least the author knows how to reign in the needless verbosity and write for a technical audience. So I will give him credit for that at least.
I'd trust OP to make good architectural decisions/provide guidance even if AI mostly wrote the docs and UI. Looking into their GitHub, seems they are an engineering manager at Microsoft and contribute semi regularly to uBlue and Project Bluefin. Should be better quality than some random vibeslopper.
Distrobox mounts $HOME in the container at $HOME by default. WSL and NSL do not, and that's my preference. This means you can install tools that change your default path, change your dotfiles, etc, without affecting the host. The other obvious difference is that distrobox is powered by podman or docker, while NSL uses a VM that runs systemd-nspawn containers.
I was wondering the same thing. Distrobox mounts $HOME by default, but it’s trivial to give a distrobox its own $HOME.
Can you explain the advantages of systemd-nspawn containers versus podman/docker? I’m not familiar with systemd-nspawn, but I’m a regular user of distrobox and podman.
I'm a bit of an outsider here (I don't understand computer) but a few weeks ago I was wrestling with the question of deploying an application on many boxen. After looking into Docker (and unikernels, lol) I arrived at the conclusion that "the operating system is part of my application."
A very bloated "part", which I did not write, don't understand very well, and which spontaneously self-modifies. An OS is a big bag of global mutable state! Ew!
This made me very sad. (At which point I was informed that I had basically reinvented NixOS from first principles?)
Also something about "we don't break userspace -- that's libc's job!"
This is most useful to people who run Linux as their daily driver so rather than saying its like WSL can you explain what it offers that existing Linux containers do not?
Its also sounds different from WSL which I thought is a VM rather than a container.
> It's yet another step in my long journey to keep my host installation free from all the changing and breaking dev dependencies that force a reinstall every few months.
Something that also require a bit of explanation. What do you do that makes this such a common problem.
WSL2 uses a shared VM that does host integration (networking, shared files, permanent storage for each instance). NSL uses the same model but with Linux native technical implementations.
As for keeping my host clean - it's the developer's curse that always gets me. Install libWhatever3.2-dev because you need it to compile something, then don't realize until next time you open Chrome that it broke your system in some subtle way. There are dozens of ways to solve this like devcontainers, docker, incus, fully separate or remote vms. I like the WSL2 model so I wanted that same UX.
Hah! One would wish so but they do. `/usr/local` usually has preference over `/usr` and libraries that use a plugin architecture often break when you update but forget to recompile plugins.
And once you go into FFI for interpreted or JIT compiled languages it turns into hell. Both Python and Java creates huge headaches when you have multiple versions of the same libraries in various /usr subdirs.
I often helped lab students who read a simple `make; make install` tutorial for OpenCV and permanently messed up their Python installs.
So a single VM with multiple containers within it? Interesting, but I think you should make it immediately clear on your site. Persistence is not the USP because there are lots of ways that you can get persistent VMs or sandboxes.
It makes an image and runs it in separate namespace. It can start at bash or init. It's very fast but there seem to be a problem creating an Ubuntu image on Debian and vice versa.
I've wanted a little more isolation than what distrobox can give me for a while now, and I did enjoy WSL when I used Windows. I've also never heard of Frostyard before, and the other repos look pretty interesting as well. Thanks for sharing!
Nope. This is a VM, with containers inside. And the containers weren't "re-implemented", it uses existing systemd containers. That's what I was able to learn by skimming the site for 30 seconds. Did you not even do that? And why even be so condescending to someone else's project?
Because it does seem like we already have tools that skin this particular cat, it's valid to wonder what value this project provides over existing solutions.. even if they chose the wrong example to compare it to. Why do you think that's condescending?
LXC is a jail and qemu is an extra layer of emulation. Why pay that extra cost? Just because it serves a wsl image isn't really an answer. How is that worth the extra layer?
That was my go-to for the past 8+ years as well. In fact, a different angle on this same problem I wrote is Blincus: https://blincus.dev I think I lost the plot on it though and tried to do too much, automate too much.
The name WSL has always struck me as a bit back area (LSW would seem to make more sense, since it is a Linux subsystem for Windows). I guess they mean it as a Windows Subsystem for Linuxing.
Anyway, should this be called LSL or WSLL? Or maybe LSWSL.
The naming is a bit awkward now, but an original design goal of the Windows NT kernel was the ability to host different types of OS "personalities" (userspace, essentially) on top of the Windows kernel, and these are called "subsystems." Win32 is one such subsystem, although there was no need to formally call it "Windows Subsystem for Win32" because everyone understood that it was just Win32 ported to NT.
The NT kernel designers came from VMS, and during development MS made various half-hearted promises that NT would be able to run VMS software as well but never actually followed through. If they had, there would likely have been a Windows Subsystem for VMS.
It's an issue because Linux is a trademark. Microsoft cannot call something a "Linux something" without infringing a trademark, but they can say something is "for Linux" the same way third-party companies could market e.g., WordPerfect for Windows. Hence Windows Subsystem for Linux, mirroring the old Windows Services for Unix:
It's a subsystem instead of "services" because WSL1 really was an NT subsystem: a "kernel personality" that knows how to interpret Linux system calls just like the DOS and OS/2 subsystems of early NT.
Is the VM layer mostly for host hygiene, or are you also treating the instances as a real security boundary when something untrusted (deps or agent-written code) runs inside?
It really depends on how much of that other distro you really want/need. NSL's benefit for me is that it's headless unless I need a graphical app, then it shares the host's Wayland. My most common use case is an IDE inside a completely different distro from the host. If you want the whole desktop environment from your VM then NSL isn't your best route.
none of the above. a single vm started by systemd-vmspawn which then starts one or more containers for your instances using systemd-nspawn. It ends up being fast and lightweight.
Exactly opposite, gonna test it out today for a pet project since I can let Claude go haywire inside if it needs to install any deps from systemside. Similar stuff exists, but this seems lightweight enough for experimentation.
I never thought I’d prefer WSL to even my MacBook for working with remote servers and dev but somehow I do.
I of course know the many ways to roll something like this for myself, yes dev containers are better for many things etc. but it’s wierd how good the ergonomics of a WSL like container are.
From the OP I see:
> my long journey to keep my host installation free from all the changing and breaking dev dependencies that force a reinstall every few months
mise does this 100% for me (assuming we don't count mise itself as one of those dev dependencies). I'm struggling to think of what falls outside of it?
Got a quick demo? Just opening an editor, running a dev server, and showing how you access the files would help me get a feel for it.
And what prevented adopting/supporting existing projects, instead of further tool ecosystem fragmentation? (kinda the same question, just differently phrased)
And then I read your comment and my first reaction was, oh, they typoed toolbox. But, nope, Red Hat created toolbx which confusingly has the CLI binary of `toolbox`: https://containertoolbx.org
https://snowlinux.org/
I submitted the Github page as a new story:
https://github.com/frostyard/snowdesktop
https://news.ycombinator.com/item?id=49907568
The company behind rebranded to SUE.
Is that massive amount of bumf machine-generated?
Gosh. OK then.
I used to poo-poo when people said that containers aren't a _real_ security boundary, at least for personal stuff, and not a multi-tenant server. But I bet even mid-tier LLMs can break out of LXC/Docker/nspawn at this point.
Any kernel could be the last to support old nVidia drivers for your $10k card.
Windows (10 LTSC or 11 with dTPM) actually works better for old Nvidia systems with WSL. You even get CUDA libraries within WSL and security updates for the next 5 years.
> Your files and your account [..]
> Ports and windows on the host
it is quite likely that you can break out even with no linux containers related CVEs. --isolate does seem to fix that somehow but is explicit opt. in and "more painful to use" ... (which creates a UX challenge unlikely to end well from a security POV).
---
"During a test conducted by Trail of Bits researcher Artem Dinaburg, a preview version of GPT 5.6-Cyber was tasked with breaking out of a Debian 12 virtual machine. Initially, the agent exploited a known Linux kernel vulnerability, CVE-2026-53359, by developing its own exploit. After the host was updated, the agent found another pathway through libslirp, chaining a known vulnerability (CVE-2026-9539) with a previously unassigned bug to gain arbitrary host memory access. Even after QEMU and libslirp were updated, the agent analyzed system components and constructed a new escape chain using three zero-day vulnerabilities and one KVM flaw that had not yet reached the distribution kernel.
These findings suggest that general-purpose VMs may not be adequate security boundaries for highly capable AI agents, especially in older systems with delayed security updates. Trail of Bits recommends using specialized isolation systems like Firecracker, restricting VM access, and implementing rapid patching to mitigate these risks."
---
https://www.scworld.com/brief/ai-agent-repeatedly-escapes-vi...
I'm using firecracker now with a very defensive systemd-as-separate-non-admin-user seccomp sandbox on top, which seems to hold them off long enough for me to see an agent going rogue and intervening.
Currently I still have hopes that eBPF sandboxing will help, but just a couple days ago my agent discovered a use after free bug in the ebpf kernel-side verifier... so there's that.
So basically, it's been all Claude'd up extremely recently.
That said, the landing page, docs, and git README are much higher quality than I normally see out of LLM-generated projects, so at least the author knows how to reign in the needless verbosity and write for a technical audience. So I will give him credit for that at least.
To be fair, WSL mounts all your drives under /mnt/ by default, which I would argue isn't any better, arguably even worse.
Can you explain the advantages of systemd-nspawn containers versus podman/docker? I’m not familiar with systemd-nspawn, but I’m a regular user of distrobox and podman.
- https://sdme.io
- https://www.youtube.com/watch?v=1eVfgzd_xyQ
I'm a bit of an outsider here (I don't understand computer) but a few weeks ago I was wrestling with the question of deploying an application on many boxen. After looking into Docker (and unikernels, lol) I arrived at the conclusion that "the operating system is part of my application."
A very bloated "part", which I did not write, don't understand very well, and which spontaneously self-modifies. An OS is a big bag of global mutable state! Ew!
This made me very sad. (At which point I was informed that I had basically reinvented NixOS from first principles?)
Also something about "we don't break userspace -- that's libc's job!"
https://blog.hiler.eu/win32-the-only-stable-abi/
Its also sounds different from WSL which I thought is a VM rather than a container.
> It's yet another step in my long journey to keep my host installation free from all the changing and breaking dev dependencies that force a reinstall every few months.
Something that also require a bit of explanation. What do you do that makes this such a common problem.
As for keeping my host clean - it's the developer's curse that always gets me. Install libWhatever3.2-dev because you need it to compile something, then don't realize until next time you open Chrome that it broke your system in some subtle way. There are dozens of ways to solve this like devcontainers, docker, incus, fully separate or remote vms. I like the WSL2 model so I wanted that same UX.
And once you go into FFI for interpreted or JIT compiled languages it turns into hell. Both Python and Java creates huge headaches when you have multiple versions of the same libraries in various /usr subdirs.
I often helped lab students who read a simple `make; make install` tutorial for OpenCV and permanently messed up their Python installs.
FWIW, I'm currently using systemd-nspawn via mkosi: https://github.com/systemd/mkosi
It makes an image and runs it in separate namespace. It can start at bash or init. It's very fast but there seem to be a problem creating an Ubuntu image on Debian and vice versa.
Now all you have to do is run NSL under WSL.
The benefit is that its super low overhead (no VM).
Anyway, should this be called LSL or WSLL? Or maybe LSWSL.
The NT kernel designers came from VMS, and during development MS made various half-hearted promises that NT would be able to run VMS software as well but never actually followed through. If they had, there would likely have been a Windows Subsystem for VMS.
https://en.wikipedia.org/wiki/Windows_Services_for_UNIX
It's a subsystem instead of "services" because WSL1 really was an NT subsystem: a "kernel personality" that knows how to interpret Linux system calls just like the DOS and OS/2 subsystems of early NT.
I think I'll stick to Nix & Podman.
It's neat seeing this experience come to linux.
Sure you "could" do this with a bunch of tools, but it's the UX that matters.
Creative idea to fix something that nobody thought was broken, but actually was.
Which?
https://news.ycombinator.com/item?id=49896422
The website's Claudisms are unbearable.
It's an interesting project.
Are these systemd containers comparable to lxc containers or better maybe?
This could be a really interesting alternative to fully featured LXC containers.
Ubuntu has a similar cli tool you probably know about called multi-pass but it just spins up Ubuntu distros.
The issue I had with multipass was how difficult it was to access and backup files, I guess it was meant for more of a disposable container.
Will try to give it a spin later.