The questions people ask first

No. Rabbit is an x86_64 kernel written from scratch that is compatible with the Linux userspace ABI: the same system calls, the same struct layouts, the same ELF loading rules. It is not a distribution and it does not replace Linux.

It runs in QEMU on x86_64, with -M q35. It is not bare metal yet, and without KVM or HVF the timings describe an emulator. Hardware acceleration is the open issue.

QEMU 8.2 or newer, 512 MiB of memory for the operations or 1024 MiB for the model and the agent, and a virtio network device for the network. The full comparison against mainline Linux needs a Linux host.

SmolLM2-135M-Instruct, in bfloat16, with the revision pinned and the SHA-256 verified. A quantized checkpoint is produced by the same repository and measured too: Q8 reproduces the reference tokens, Q4 changes the answer, and that change is published as a loss.

The model is local, inside the initramfs. The agent can also reach the network: a tool speaks HTTP to an API and another sends a datagram. It is an agent with a local brain and hands that reach the world.

It is not production ready, and it is not hardened on purpose: no KPTI, no retpoline, no SMAP enforcement. The threat model is a private single tenant VM, and the limitation is documented rather than quietly dropped.

GPL-2.0-only. Kernel crates may only use MIT, BSD or ISC licensed code, because GPL-2.0-only is incompatible with Apache-2.0-only.

Same host, same QEMU version, same machine model, same rootfs, same binary and same scripts. The raw CSV is committed, the losses are included, and the kernel with mitigations off is measured too, so the comparison cannot be cherry-picked.

All of it, and that is stated on purpose. The design decisions, the milestone criteria and the verification belong to a human; every milestone has an objective exit criterion that a machine checks in CI. The repository is the evidence.