LiteBox is a sandboxing library OS that drastically cuts down the interface to the host, thereby reducing attack surface. It focuses on easy interop of various "North" shims and "South" platforms. LiteBox is designed for usage in both kernel and non-kernel scenarios.
LiteBox exposes a Rust-y nix/rustix-inspired "North" interface when it is provided a Platform interface at its "South". These interfaces allow for a wide variety of use-cases, easily allowing for connection between any of the North--South pairs.
Example use cases include:
- Running unmodified Linux programs on Windows
- Sandboxing Linux applications on Linux
- Run programs on top of SEV SNP
- Running OP-TEE programs on Linux
- Running on LVBS
With how buggy their flagship OS has become, why would I trust anything else they release to be better? Or even if it does work well now, why should I expect it to stay that way? Microsoft has burned through all possible goodwill at this point, at least for me.
This isn't supposed to replace Windows, and it isn't a GUI desktop operating system at all. I doubt anyone working on this has anything to do with the modern Windows desktop UX.
OP wasn't suggesting it was, just that the lack of quality in one significant area of the company's output leads to a lack of confidence in other products that they release.
>Kernel and low level stuff are actually very stable and good.
This. A while ago a build of Win 11 was shared/leaked that was tailored for the Chinese government called "Windows G" and it had all the ads, games, telemetry, anti-malware and other bullshit removed and it flew on 4GB RAM. So Microsoft CAN DO IT, if they actually want to, they just don't want to for users.
You can get something similar yourself at home running all the debloat tools out there but since they're not officially supported, either you'll break future windows updates, or the future windows updates will break your setup, so it's not worth it.
Talked about back in the Vista days publicly (I cannot find the articles now) - Microsoft has commitments to their hardware partners to help keep the hardware market from collapsing.
So they are not incentivized to keep Win32_Lean_N_Mean, but instead to put up artificial limits on how old of hardware can run W11.
I have no insider knowledge here, just this is a thing which get talked about around major Windows releases historically.
> Kernel and low level stuff are actually very stable and good.
In their intended applications, which might or might not be the ones you need.
The slowness of the filesystem that necessitated a whole custom caching layer in Git for Windows, or the slowness of process creation that necessitated adding “picoprocesses” to the kernel so that WSL1 would perform acceptably and still wasn’t enough for it to survive, those are entirely due to the kernel’s archtecture.
It’s not necessarily a huge deal that NT makes a bad substrate for Unix, even if POSIX support has been in the product requirements since before Win32 was conceived. I agree with the MSR paper[1] on fork(), for instance. But for a Unix-head, the “good” in your statement comes with important caveats. The filesystem is in particular so slow that Windows users will unironically claim that Ripgrep is slow and build their own NTFS parsers to sell as the fix[2].
But there's another issue which is what cripples windows for dev! NTFS has a terrible design flaw which is the fact that small files, under 640 bytes, are stored in the MFT. The MFT ends up having serious lock contention so lots of small file changes are slow. This screws up anything Unixy and git horribly.
WSL1 was built on top of that problem which was one of the many reasons it was slow as molasses.
This is not due to slowness of the file system. Native ntfs tools are much faster than Unix ones in some situations. The issue is that running Unix software on windows will naturally have a performance impact. You see the same thing in reverse using Wine on Linux. Windows uses a different design for IO so requires software to be written with that design in mind.
Microsoft doesn't have a very good track record with security or privacy. Maybe it works, but yeah you'll probably get screwed over at some point.
Still, the fact that it's open source is a good thing. People can now take that code and make something better (ripping out the AI for example) or just use bits and pieces for their own totally unrelated projects. I can't see that as anything but a win. I have no problem giving shitty companies credit where its due and they've done a good thing here.
It's a library that is linked to in place of an operating system - so whatever interface the OS provided (syscalls+ioctls, SMC methods, etc.) ends up linked / compiled into the application directly, and the "external interface" of the application becomes something different.
This is how most unikernels work; the "OS" is linked directly into the application's address space and the "external interface" becomes either hardware access or hypercalls.
Wine is also arguably a form of "library OS," for example (although it goes deeper than the most strict definition by also re-implementing a lot of the userland libraries).
So for example with this project, you could take a Linux application's codebase, recompile it linked to LiteBox, and run it on SEV-SNP. Or take an OP-TEE TA, link it to LiteBox, and run it on Linux.
The notable thing here is that it tries to cut the interface in the middle down to an intermediate representation that's supposed to be sandbox-able - ie, instead of auditing and limiting hundreds of POSIX syscalls like you might with a traditional kernel capabilities system, you're supposed to be able to control access to just a few primitives that they're condensed down to in the middle.
> So for example with this project, you could take a Linux application's codebase, recompile it linked to LiteBox
If you have to recompile, you might as well choose to recompile to WASM+WASI. The sandboxing story here is excellent due to its web origins. I thought the point of LiteBox is that recompilation isn’t needed.
Looking more closely, it looks like there are some "North" sides (platforms) with ABI shims (currently Linux and OP-TEE), but others (Windows, for example), would still require recompilation.
> If you have to recompile, you might as well choose to recompile to WASM+WASI.
I disagree here; this ignores the entire swath of functionality that an OS or runtime provides? Like, as just as an example, I can't "just recompile" my OP-TEE TA into WASM when it uses the KDF function from the OP-TEE runtime?
I think that's an OS in the form of a library, like Wine for example. From what I get from the description it allows you to run programs on your real OS and make it see a cut down API to your actual system to reduce the attack surface.
yeah, same here, I was like "wow what an interesting side to their business, a whole operating system intended to serve public and academic libraries!"
My understanding of this is that it is a sandbox. Providing a common interface like if it was an OS for the program to run inside, but avoiding the program to use the OS directly.
What is unclear is if it uses its own common ABI or if you use the one of the host os.
I don't know why but from the project description I have a little bit of feeling that this is another vibe coded project.
A library OS is an OS that is linked directly to your program instead of being a separate program accessed through a syscall to kernel mode. About the same as a “unikernel”, but a more recent term.
Basically it lets your program run directly on a hypervisor VM, though this one will also run as a Linux/Windows/BSD process.
No mention of starting with a design specification & then tied to formal verification the whole way?
It sounds interesting and a step forward (never heard of library Os itll now), but why won't this run into hundreds of the same security bugs that plague Windows if it's not spec'd and verified?
The lack of integrated sandboxing in windows compared to android/iphone is still frankly unacceptable. I've become increasingly paranoid about running any application on Windows (not that your average linux distro is even remotely better) and yet Apple and Google seem to be far, far ahead in user permissions (especially with GrapheneOS, god bless that team) and isolation of processes.
Consumers and businesses deserve better. It's crazy to me that in 2026 Notepad++ being compromised means as much potential damage as it does, still.
The sandboxing on mobile platforms puts the OS vendor in a special position to enforce a monopoly on apps and features. Apple enforces it aggressively, while Google only reluctantly so far. It also prevents the user from exerting full control of the system. Apple does it by locking things down directly, while Google punishes you for owning your devices with attestation.
There has to be a better way. I think Linux's flatpak is a reasonable approach here, although the execution might be rather poor. I want a basic set of trusted tool that I can do anything with, and run less trusted tools like GUI programs in sandboxes with limited filesystem access.
Those are policy decisions not really connected to the sandboxing technology. They control what sort of signing the system will accept and make it so that it only runs things they approve, and they only approve things that are sandboxed a certain way. The exact same sandboxing could be used with a system where an admin user can decide what gets to run and what kind of sandboxing is required for each thing.
> I've become increasingly paranoid about running any application on Windows (not that your average linux distro is even remotely better)
Linux excels over Windows in the area of security by a wide margin, I have no qualms about running an app on Linux versus Windows, any day of the week.
IIUC, if you have the source you can recompile said Windows app with LiteBox to statically link in the Windows OS kernel dependencies, so it'll run on any compatible processor regardless of OS (since it won't be making syscalls anymore). It's a unikernel basically.
That's the theory, but I don't know how far LiteBox is along to supporting that workflow.
> It focuses on easy interop of various "North" shims and "South" platforms.
For replacing wine on Linux the "North" would be kernel32 API or similar, the "South" would be Linux sys all API.
However this is meant as a library, thus require linking the Windows program to it and eine is more than the system interface, it has all the GUI parts etc of win32 API
A library os to me would typically mean it's aimed at hosting a single user program on bare hardware. I don't see that here, but maybe I'm just confused
It's both; it's aimed at hosting a single user program on another userspace, but also seems to have its own kernel as well?
The "North" part seems to be what I think you'd traditionally think of as a library OS, and then the "South" part seems to be shims to use various userlands and TEEs as the host (rather than the bare hardware in your example).
I'm really confused by the complete lack of documentation and examples, though. I think the "runners" are the closest thing there is.
I read this type of (sour) comment more and more on this forum. To me it reads very cynical and I wonder what the author is trying to say with this. Are you perhaps negatively impacted by automatic coding?
I read your comment as ignorant to AI's capabilities and their negative outcomes with relying on vibe coding.
The implication is that MS is forcing AI adoption on users at a point of absurd recklessness, and that they should not be trusted - especially not blindly trusted.
Perhaps the reason you're seeing comments similar to my original comment more frequently is because actual software engineers whom know the capabilities of AI and how much of a bad decision it is to assume it's as good as a competent engineer. Many engineers have had years of experience working with management, whom while have legit concerns about the capabilities of software as they are ultimately responsible for it and the financials, see them turning to vibe coding and relying on it. Non technical folks think software is kinda easy to do, and because LLMs can generate code that it just proves their assumptions.
I'm not sure whether Microsoft, the makers of Windows 95 (after which I stopped taking them seriously), are the sharpest tool in the box when it comes to security.
From the GitHub page:
LiteBox is a sandboxing library OS that drastically cuts down the interface to the host, thereby reducing attack surface. It focuses on easy interop of various "North" shims and "South" platforms. LiteBox is designed for usage in both kernel and non-kernel scenarios.
LiteBox exposes a Rust-y nix/rustix-inspired "North" interface when it is provided a Platform interface at its "South". These interfaces allow for a wide variety of use-cases, easily allowing for connection between any of the North--South pairs.
Example use cases include:
More links with discussion:
Reddit discussion: https://www.reddit.com/r/linux/comments/1qw4r71/microsofts_n...
Project lead James Morris announcing it on social.kernel.org: https://social.kernel.org/notice/B2xBkzWsBX0NerohSC
FYI, I am not the project lead for Litebox. It is led by Microsoft Research.
Copilot
https://github.com/microsoft/litebox/blob/main/.github/copil...
To be expected, given how many organisations now require employees to use AI if they want to meet their OKRs, especially all that sell AI tools.
https://files.catbox.moe/cq6xf4.png
With how buggy their flagship OS has become, why would I trust anything else they release to be better? Or even if it does work well now, why should I expect it to stay that way? Microsoft has burned through all possible goodwill at this point, at least for me.
This isn't supposed to replace Windows, and it isn't a GUI desktop operating system at all. I doubt anyone working on this has anything to do with the modern Windows desktop UX.
> This isn't supposed to replace Windows,
OP wasn't suggesting it was, just that the lack of quality in one significant area of the company's output leads to a lack of confidence in other products that they release.
UI of Windows is buggy and inconsistent. Kernel and low level stuff are actually very stable and good.
>Kernel and low level stuff are actually very stable and good.
This. A while ago a build of Win 11 was shared/leaked that was tailored for the Chinese government called "Windows G" and it had all the ads, games, telemetry, anti-malware and other bullshit removed and it flew on 4GB RAM. So Microsoft CAN DO IT, if they actually want to, they just don't want to for users.
You can get something similar yourself at home running all the debloat tools out there but since they're not officially supported, either you'll break future windows updates, or the future windows updates will break your setup, so it's not worth it.
Talked about back in the Vista days publicly (I cannot find the articles now) - Microsoft has commitments to their hardware partners to help keep the hardware market from collapsing.
So they are not incentivized to keep Win32_Lean_N_Mean, but instead to put up artificial limits on how old of hardware can run W11.
I have no insider knowledge here, just this is a thing which get talked about around major Windows releases historically.
Never heard of Windows G .. that sounds exactly what I want for my older Thinkpads :-)
Is this not just Windows LTSB/LTSC? Which has been a thing forever.
Maybe, could also be that for a 9 figure government contract they'll provide a custom LTSC branch just for you with only the features you want.
> Kernel and low level stuff are actually very stable and good.
In their intended applications, which might or might not be the ones you need.
The slowness of the filesystem that necessitated a whole custom caching layer in Git for Windows, or the slowness of process creation that necessitated adding “picoprocesses” to the kernel so that WSL1 would perform acceptably and still wasn’t enough for it to survive, those are entirely due to the kernel’s archtecture.
It’s not necessarily a huge deal that NT makes a bad substrate for Unix, even if POSIX support has been in the product requirements since before Win32 was conceived. I agree with the MSR paper[1] on fork(), for instance. But for a Unix-head, the “good” in your statement comes with important caveats. The filesystem is in particular so slow that Windows users will unironically claim that Ripgrep is slow and build their own NTFS parsers to sell as the fix[2].
[1] https://lwn.net/Articles/785430/
[2] https://nitter.net/CharlieMQV/status/1972647630653227054
This is on the mark.
But there's another issue which is what cripples windows for dev! NTFS has a terrible design flaw which is the fact that small files, under 640 bytes, are stored in the MFT. The MFT ends up having serious lock contention so lots of small file changes are slow. This screws up anything Unixy and git horribly.
WSL1 was built on top of that problem which was one of the many reasons it was slow as molasses.
Also why ReFS and "dev drive" exist...
The file system isn't slow. The slowness will be present in any file system due to the file system filters that all file system calls pass though.
This is not due to slowness of the file system. Native ntfs tools are much faster than Unix ones in some situations. The issue is that running Unix software on windows will naturally have a performance impact. You see the same thing in reverse using Wine on Linux. Windows uses a different design for IO so requires software to be written with that design in mind.
Microsoft fanboy spotted...
Why aren't you on the hate wagon with the rest of us, boy? You think yer BETTER? HUH?
NTFS, not so great.
NTFS is just fine. Stable, reliable, fast, plenty of features for a general purpose file system.
...But no way can you wrap it into something that looks posix-y from the inside
Why would you want to?
Microsoft doesn't have a very good track record with security or privacy. Maybe it works, but yeah you'll probably get screwed over at some point.
Still, the fact that it's open source is a good thing. People can now take that code and make something better (ripping out the AI for example) or just use bits and pieces for their own totally unrelated projects. I can't see that as anything but a win. I have no problem giving shitty companies credit where its due and they've done a good thing here.
What is a 'library OS'?
It's a library that is linked to in place of an operating system - so whatever interface the OS provided (syscalls+ioctls, SMC methods, etc.) ends up linked / compiled into the application directly, and the "external interface" of the application becomes something different.
This is how most unikernels work; the "OS" is linked directly into the application's address space and the "external interface" becomes either hardware access or hypercalls.
Wine is also arguably a form of "library OS," for example (although it goes deeper than the most strict definition by also re-implementing a lot of the userland libraries).
So for example with this project, you could take a Linux application's codebase, recompile it linked to LiteBox, and run it on SEV-SNP. Or take an OP-TEE TA, link it to LiteBox, and run it on Linux.
The notable thing here is that it tries to cut the interface in the middle down to an intermediate representation that's supposed to be sandbox-able - ie, instead of auditing and limiting hundreds of POSIX syscalls like you might with a traditional kernel capabilities system, you're supposed to be able to control access to just a few primitives that they're condensed down to in the middle.
> So for example with this project, you could take a Linux application's codebase, recompile it linked to LiteBox
If you have to recompile, you might as well choose to recompile to WASM+WASI. The sandboxing story here is excellent due to its web origins. I thought the point of LiteBox is that recompilation isn’t needed.
Looking more closely, it looks like there are some "North" sides (platforms) with ABI shims (currently Linux and OP-TEE), but others (Windows, for example), would still require recompilation.
> If you have to recompile, you might as well choose to recompile to WASM+WASI.
I disagree here; this ignores the entire swath of functionality that an OS or runtime provides? Like, as just as an example, I can't "just recompile" my OP-TEE TA into WASM when it uses the KDF function from the OP-TEE runtime?
I think that's an OS in the form of a library, like Wine for example. From what I get from the description it allows you to run programs on your real OS and make it see a cut down API to your actual system to reduce the attack surface.
https://en.wikipedia.org/wiki/Operating_system#Library
at first I thought library OS might have meant an OS meant for use at a library.
Honestly far less interesting to know I was wrong.
Is it not? You link the "library os" and you no longer need an os (when running in a supervisor) IIUC.
yeah, same here, I was like "wow what an interesting side to their business, a whole operating system intended to serve public and academic libraries!"
I’m not sure I understand what a library OS is; can someone here elaborate?
My understanding of this is that it is a sandbox. Providing a common interface like if it was an OS for the program to run inside, but avoiding the program to use the OS directly.
What is unclear is if it uses its own common ABI or if you use the one of the host os. I don't know why but from the project description I have a little bit of feeling that this is another vibe coded project.
A library OS is an OS that is linked directly to your program instead of being a separate program accessed through a syscall to kernel mode. About the same as a “unikernel”, but a more recent term.
Basically it lets your program run directly on a hypervisor VM, though this one will also run as a Linux/Windows/BSD process.
No mention of starting with a design specification & then tied to formal verification the whole way?
It sounds interesting and a step forward (never heard of library Os itll now), but why won't this run into hundreds of the same security bugs that plague Windows if it's not spec'd and verified?
People seem to believe writing things in rust means it's correct.
The lack of integrated sandboxing in windows compared to android/iphone is still frankly unacceptable. I've become increasingly paranoid about running any application on Windows (not that your average linux distro is even remotely better) and yet Apple and Google seem to be far, far ahead in user permissions (especially with GrapheneOS, god bless that team) and isolation of processes.
Consumers and businesses deserve better. It's crazy to me that in 2026 Notepad++ being compromised means as much potential damage as it does, still.
The sandboxing on mobile platforms puts the OS vendor in a special position to enforce a monopoly on apps and features. Apple enforces it aggressively, while Google only reluctantly so far. It also prevents the user from exerting full control of the system. Apple does it by locking things down directly, while Google punishes you for owning your devices with attestation.
There has to be a better way. I think Linux's flatpak is a reasonable approach here, although the execution might be rather poor. I want a basic set of trusted tool that I can do anything with, and run less trusted tools like GUI programs in sandboxes with limited filesystem access.
Those are policy decisions not really connected to the sandboxing technology. They control what sort of signing the system will accept and make it so that it only runs things they approve, and they only approve things that are sandboxed a certain way. The exact same sandboxing could be used with a system where an admin user can decide what gets to run and what kind of sandboxing is required for each thing.
There are containers, and one of their users is the Windows Sandbox - https://learn.microsoft.com/en-us/windows/security/applicati...
> I've become increasingly paranoid about running any application on Windows (not that your average linux distro is even remotely better)
Linux excels over Windows in the area of security by a wide margin, I have no qualms about running an app on Linux versus Windows, any day of the week.
UWP, and MSIX on Win32 via Appstore.
There is also sandboxing configuration via Intune for enterprises.
The cargo.lock file is 2200+ lines long. Did they spend a reasonable amount of time auditing these dependencies?
That's 238 dependencies (counting multiple versions of the same crate).
* Many of them are part of families of crates maintained by the same people (e.g. rust-crypto, windows, rand or regex).
* Most of them are popular crates I'm familiar with.
* Several are only needed to support old compiler versions and can be removed once the MSRV is raised
So it's not as bad as it looks at first glance.
What would be a reasonable amount of time to audit the dependencies?
I would let them decide based on their security policy.
If Microsoft states that they don't have any for a project like this, I would be wary of taking it too seriously.
They ran it through Copilot which gave it the all-clear.
I've always done 'sort | uniq'. Never bothered to check for the the unique flag to sort. Although 'uniq -c' is quite nice to have.
Yeah, to see the packages with multiple versions:
Package windows-sys has the highest number of versions included, 3: 0.59.0, 0.60.2, and 0.61.2.Edit: Also, beware of the unsorted uniq count:
grep -v '1 name' excludes 11, 21, etc., but I take your point.
Given, you know, Microsoft, I'd demand proof even if they said they did.
Hmmm. Another, admittedly interesting, step towards the complete digital lockdown. Isolate and virtualize everything, now also governed by AI!
I wonder if they, the industry as a whole, eventually will make being able to freely use a PC a subscription, bastardizing "freedom" completely.
Another layer (ouch) to abstract away Windows (ouch * ouch).
Use Linux or BSD and ignore that approach for Vendor Lock-in* into their “library OS”.
No deployment instructions?
Baaah! Microsoft, security-focused in a single sentence!
Can it replace Wine to run Windows apps on Linux?
IIUC, if you have the source you can recompile said Windows app with LiteBox to statically link in the Windows OS kernel dependencies, so it'll run on any compatible processor regardless of OS (since it won't be making syscalls anymore). It's a unikernel basically.
That's the theory, but I don't know how far LiteBox is along to supporting that workflow.
They say
> It focuses on easy interop of various "North" shims and "South" platforms.
For replacing wine on Linux the "North" would be kernel32 API or similar, the "South" would be Linux sys all API.
However this is meant as a library, thus require linking the Windows program to it and eine is more than the system interface, it has all the GUI parts etc of win32 API
A library os to me would typically mean it's aimed at hosting a single user program on bare hardware. I don't see that here, but maybe I'm just confused
It's both; it's aimed at hosting a single user program on another userspace, but also seems to have its own kernel as well?
The "North" part seems to be what I think you'd traditionally think of as a library OS, and then the "South" part seems to be shims to use various userlands and TEEs as the host (rather than the bare hardware in your example).
I'm really confused by the complete lack of documentation and examples, though. I think the "runners" are the closest thing there is.
The reddit conversation seems to allude to you being correct.
"We did not find any viable commercial use for it, but maybe you will."
Cool
What % of it is vibe-coded in copilot ?
Just assume the only thing a human did was name write the initial prompt.
I read this type of (sour) comment more and more on this forum. To me it reads very cynical and I wonder what the author is trying to say with this. Are you perhaps negatively impacted by automatic coding?
Do you want to enable Copilot ?
we are ALL negatively impacted by generative excrement
I have to use Windows at my day job
and my god, I'd prefer Windows 3.1
Nope, not at all.
I read your comment as ignorant to AI's capabilities and their negative outcomes with relying on vibe coding.
The implication is that MS is forcing AI adoption on users at a point of absurd recklessness, and that they should not be trusted - especially not blindly trusted.
Perhaps the reason you're seeing comments similar to my original comment more frequently is because actual software engineers whom know the capabilities of AI and how much of a bad decision it is to assume it's as good as a competent engineer. Many engineers have had years of experience working with management, whom while have legit concerns about the capabilities of software as they are ultimately responsible for it and the financials, see them turning to vibe coding and relying on it. Non technical folks think software is kinda easy to do, and because LLMs can generate code that it just proves their assumptions.
I'm not sure whether Microsoft, the makers of Windows 95 (after which I stopped taking them seriously), are the sharpest tool in the box when it comes to security.