The LiberNT project is working on an alternative implementation of the Windows NT architecture, using Windows 11 version 24H2 as its compatibility reference. Unlike Linux distributions, LiberNT does not use the Linux kernel. It builds on the ReactOS codebase while replacing and extending several core components.
One of the project’s main priorities is ARM64 support, particularly for single-board computers and embedded devices. LiberNT also supports x86-64 and 32-bit x86, while RISC-V 64-bit and PowerPC ports are under development.
According to the developers, the implementation follows a clean-room approach. Windows behavior is studied using Microsoft’s public documentation, SDK and WDK headers, public symbols, and black-box testing. The team states no proprietary Windows code is copied or transcribed.
On the technical side, one of the biggest changes is NVS, a new memory manager and kernel pool implementation replacing the corresponding ReactOS components. LiberNT also introduces a new cache manager, with developers claiming better results than the original ReactOS memory manager in their memory and cache benchmarks.
Other changes include a new NTFS driver, an updated StorPort implementation with NVMe support, USB xHCI and SD bus drivers, and work on the Windows Display Driver Model and Desktop Window Manager.
On the application compatibility side, LiberNT implements WoW64 on its 64-bit targets, while ARM64 systems use the FEX emulator to run x86 applications. ARM64EC support is also part of the project.
Early compatibility results are already available on the project’s compatibility page. According to tests published by the developers, Google Chrome, the open-source strategy game 0 A.D., and Command & Conquer: Yuri’s Revenge are reported to work. Steam is listed as usable, while the RPCS3 PlayStation 3 emulator currently starts but has not been classified as usable.
However, the compatibility database contains only five entries, all tested against a specific development build.
LiberNT also targets Raspberry Pi hardware, with the same ARM64 disk image designed to boot on Raspberry Pi 3, Raspberry Pi 5, and generic ARM64 virtual machines running under QEMU. The project includes drivers for Raspberry Pi display, networking, and audio hardware.
For those interested in trying it, the developers provide downloadable builds in three formats: a preinstalled raw disk image, a VHD image for virtualization platforms, and a bootable installation ISO. The preinstalled image is recommended, with support for both UEFI and BIOS booting.
Still, LiberNT remains a work in progress. The developers caution that the available builds are experimental and should be used in virtual machines or on spare hardware rather than production systems. The source code is available on GitHub.
For anyone wondering, yes, it is vibecoded. I ain’t surprised.
Actually reading the outputs and understanding them is not vibecoding.
This worked better on my old computer than ReactOS ever did and I have been testing out ReactOS since 2007. If AI is what it takes to deliver a working FOSS Windows-like OS then I am not complaining.
If I understand correctly, this new fork LiberNT is Windows 11 version 24H2 only compatibility, without support for any prior generation. ReactOS is about the look and feel and compatibility since Windows Server 2003 and later (which initially started as Windows 95 clone), without a focus on a specific Windows release. Is this correct?
https://libernt.com/ from their website it looks like typical Ai generated website. Opening the page spins up my PCs fans like a videgame would. https://github.com/LiberNT/LiberNT their source code doesn’t list any contributors, so I’m not sure what’s going on. Ah, there we go, they use Ai off course: https://github.com/LiberNT/LiberNT#use-of-ai
We want to make it clear: AI is part of how LiberNT is built. It helps us review and write code, and everything it writes is then reviewed by a human. Before anything reaches the main branch, a person runs and validates the end-to-end tests. We care about this project too much to let a regression slip through.
It also helps us read documentation, extract definitions and fix bugs. For any given bug, it compares what the official Microsoft Learn documentation says with what our function actually does, so we can see exactly where our behavior differs.
Like any other tool, AI is there to help us achieve our goals. It follows the same rules we hold ourselves to, set out in AGENTS.md:
- We build from the Windows SDK and WDK headers, Windows 11 public symbols, Microsoft documentation and black-box runs on Windows 11.
- We stay clean-room: we never transcribe disassembled or decompiled Windows code, and we never copy GPL code from Linux.
- A fix is only a fix once we have seen it work in a real boot or test run. Once tests pass, a future CI will keep them passing and preserve our overall progress.
- We compare our test results check by check with the same tests run on Windows 11.
Looks like they are having Claude strip out the original licenses in the source files and claim copyright for themselves. Shameful.
Theyre replacing the original licenses for ReactOS?
Objectively speaking this contains internal leaked Microsoft code this is simply speaking unavoidable due to their heavy use of vibe coding.
Yes, but Microsoft can’t prove that in court without also admitting that they also owe every artist copilot (and all other ais) were trained on. Actually no wait. Microsoft please sue these guys, it will better the world.
No. WTF, we shouldn’t want AI training to be considered copyright infringement (not when Microsoft does it, not when someone else does it from Microsoft’s code). We should want copyright to be weakened, not strengthened; I’ve believed that for 20 years and have no intention to stop believing it now just because AI has now become a thing.
We should want copyright to be weakened, not strengthened
I’m not sure if this is a good idea.
They would probably just set a precedent that you can only sue for AI copyright infringement if you have tons of money
Objectively speaking this contains internal leaked Microsoft code
Objectively speaking, usage of Ai does not mean that the resulting code does contain internal leaked Microsoft code. It just mean there is potential and without comparing to the original source, we are not able to confirm. Overall the LberNT might or might not contain leaked code, but we are just unable to make definitive statements yet. Unless I am missing something obvious.
Strictly speaking it doesn’t mean that but they’re replicating windows 11 and doing vibe coded reverse engineering, we know these AI models are trained on leaked code. The chance that it contains stolen code is therefore extremely high and increases as time goes on.
I’m not saying its impossible, and there is some likelihood of that happening. If this comes true and Microsoft finds about it, then expect them to close the doors soon. In their README they say its clean room implementation, and my reply with clean room + ai is always “if you cannot proof the tools aren’t trained on code you try to implement, then you cannot proof it is clean room”. So I’m right there with your concerns. The only objection I had was that using Ai isn’t proof of having stolen code in the produces output. Otherwise I’m 100% with you.
Considering how the tech giants collected the material they used to train their AIs, they don’t really have a leg to stand on when it comes to copywritten material.
ReactOS is a great idea in concept, but development has been too slow to make an impact. Maybe this will help? Not sure why they’re making a fork instead of contributing to the main project.
Some make a fork to make their own name, since they become the fork maintainers. And I believe that the sweet opportunity to make a name raised in the history dimming the original authors is one the main reasons of such forks, unfortunately… in addition to the fear of dependence from the original authors’ decisions.
For example, Gitea, the one I appreciate the authors for, did contribute, and self-host. Forgejo got popularized by Codeberg, but too few people know Forgejo is a fork of Gitea, and some question (What is Gitea ?…) what the latter/origin is even knowing the former/fork!
In the result, this happens:
- https://news.ycombinator.com/item?id=49407020 (Is this how they dim so recent history, but why?…)Gitea deserves to be forgotten.
Nothing deserves to be forgotten. Every details in the infinitely marvelous history is critically crucial and invaluably important… it must be respected, praised, preserved safely, and learned from…
Sorry, but your statement is of unfair, destructive, sorrowful, or even fascists’ attitude…
People invest their human precious finite life time, hope, love, ideas in the code…
or even fascists’ attitude
???
