Limitations¶
Honest scope notes for the current tree. These are intentional cuts and sequencing choices, not a backlog of unknown bugs.
Backends are asymmetric¶
| Track | Maturity |
|---|---|
| HVF / aarch64 | Platform layer live: Linux boot, virtio (blk/net/vsock/gpu), snapshot-fork, control plane, govern, observe, meter, SMP |
| KVM / x86-64 | PVH Linux boot, virtio-blk, IOAPIC; platform layer wired and run-verified on metal for control plane, vsock, metering, watchdogs, and slirp (compile path). Remaining gaps: virtio-net guest interface, SMP AP boot, snapshot/restore, GPU |
KVM parity is tracked in the roadmap.
Snapshot / restore¶
Snapshot save, rewind demo, and COW restore (nether-restore / restore_from=) are HVF-only today. KVM snapshot work is planned (KVM_GET/SET_* vCPU state + dirty pages) but not implemented.
Not a general-purpose VMM (yet)¶
- OVMF / UEFI is deferred. The edge path is PVH direct boot, not full firmware emulation.
- Windows guests are future scope (Phase 4+).
- Live migration and VFIO passthrough are roadmap items, not shipping.
- 3D virtio-gpu / virgl is explicitly out of core. 2D framebuffer only; 3D would be out-of-process.
Embedding¶
The shipping artifact is one swerver binary with embedded nether. The integration
contract (vsock spine, per-VM-per-worker, eventfd registration into IoRuntime) is
designed but not fully wired yet. The standalone nether executable remains for
dev and bringup only.
API stability¶
The library root (src/root.zig) and control protocol may change before 1.0. Downstream embedders should pin commits.
Platform¶
nether is the isolate + govern layer inside the swerver binary. It does not own routing, TLS, or billing — those stay in swerver and x402 above the embed boundary.
Roadmap¶
Phases, done-lines, and the platform track are in Roadmap. Architectural forks are recorded in Decisions.