Hi Enrico, Thanks for the questions – please see my answers inline:

“What exactly do you mean by that?” By “AI-native” I mean ASIOS is designed from the ground up for AI workloads, not retrofitted. It embeds ML-driven optimizations (e.g. reinforcement-learning schedulers, Bayesian latency predictors), deterministic scheduling, native multi-accelerator support and built-in alignment/safety probes directly into the OS stack.

“Why an old Ubuntu?” We chose Ubuntu 24.04 LTS because it gives us a rock-solid, long-term-support base with an upstream PREEMPT_RT HWE kernel (6.11) that we can rebase regularly. It’s reproducible, widely deployed in enterprise and cloud environments, and lets us focus on AI-centric enhancements rather than reinventing the distribution.

“If you’re talking AI, you mean datacenter, right? I’d trim things down…” Our primary target today is bare-metal and VM-based AI clusters, but ASIOS also supports very slim builds (we strip out unneeded services) and future multi-distro releases (we’re planning Devuan, Yocto/PTXdist, RHEL/SUSE variants).

“Which symbols? ELF symbols or kconfig defines?” The overlay logic deals with Kconfig symbols (CONFIG_*). asios-config-overlay.sh skips any missing CONFIG options instead of failing,

Re: extra-deb tooling Great pointer on deb-pkg – we’ll evaluate integrating it into our .deb build pipeline to streamline packaging. I did run into some issues packaging cross-architecture kernels with dpkg-buildpackage.

Please let me know if you have any other suggestions or questions! Cheers, — Vikram R KarLex AI, Inc.