If you want to think about the implementation I guess you could bundle it into the kernel then expose some input/output?
https://lemmy.today/comment/21990242
Where do we install apt in the first place if no software is on it, I guess if it was just raw linux you would just implement wget or curl then download whatever you wanted
it would be unikernel so your essentially statically linking a binary to the the linux kernel image, so it all be running in kernelspace. You would have to statically link libc and other dependencies likely (i rem red hat patching glibc for this to work but not sure what they changed?).
My job involves doing performance optimizations for billion dollar tech companies so its kind of my job to know lol.
People already have vibe coded Operating Systems, are they well made? no, I don’t think TempleOS is some enterprise grade codebase ether but its cool that he got to make his own OS. Would be cool for anyone to be able to do that.
Well the meme says to add some new layer over syscalls so the plan is already to break userspace entirely to force you to make a new one. But yeah if you wanted to make it irl bundling another kernel would likely be easier then Linux unless you wanted the hw support.
Look, honestly this has merit. Not for Claude Code, that’s insane. But there are a lot of Risc V capable NPUs that could use a smaller model for internal locally sourced workflows. I’m thinking like a read only RAG paired with a tiny GLM 4.7 trimmed down to fit on a SpacemiT Key Stone K3.
how does apt work when glibc does not work?
How does Claude work with nothing?
If you want to think about the implementation I guess you could bundle it into the kernel then expose some input/output? https://lemmy.today/comment/21990242
Where do we install apt in the first place if no software is on it, I guess if it was just raw linux you would just implement wget or curl then download whatever you wanted
how do you think that works, if the kernel doesn’t even have ls ?
it would be unikernel so your essentially statically linking a binary to the the linux kernel image, so it all be running in kernelspace. You would have to statically link libc and other dependencies likely (i rem red hat patching glibc for this to work but not sure what they changed?).
how are you this savvy with the underpinnings of the OS, but this naive about this kind of scenario and use of AI in this context?
My job involves doing performance optimizations for billion dollar tech companies so its kind of my job to know lol.
People already have vibe coded Operating Systems, are they well made? no, I don’t think TempleOS is some enterprise grade codebase ether but its cool that he got to make his own OS. Would be cool for anyone to be able to do that.
You understand the irony of a claude code LLM as a unikernel if your background is perf op for an OS right?
well yeah but the sublemmy is Microblog Memes not StartUpIdeas
This breaks the prime directive of Linux. “Don’t break userspace”
If you’re working on the linux kernel everything you build drops once init/systemd launches.
https://unix.stackexchange.com/questions/235335/why-is-there-a-linux-kernel-policy-to-never-break-user-space#235532
Claude Code is an LLM that relies on multiple libraries not just glibc. What you’re proposing breaks the basic tenets of the Linux Kernel.
I think you’d probably be better served packaging this as a RISC V OS instead as binaries do run in the kernel there.
Well the meme says to add some new layer over syscalls so the plan is already to break userspace entirely to force you to make a new one. But yeah if you wanted to make it irl bundling another kernel would likely be easier then Linux unless you wanted the hw support.
Look, honestly this has merit. Not for Claude Code, that’s insane. But there are a lot of Risc V capable NPUs that could use a smaller model for internal locally sourced workflows. I’m thinking like a read only RAG paired with a tiny GLM 4.7 trimmed down to fit on a SpacemiT Key Stone K3.
I’ll reference you on Codeberg.
https://www.cnx-software.com/2026/01/23/spacemit-k3-16-core-risc-v-soc-system-information-and-early-benchmarks/ woah these do seem cool