Machine limitations

This page keeps the mvmctl machine happy path honest. If a capability is not listed as supported here, treat it as backend-specific, experimental, or future work rather than a default guarantee.

Host Nix

Image-backed machine runs do not need host Nix. The normal path is a signed mvmctl binary plus mvmctl machine run --image ....

Workflows that build Linux images, evaluate flakes, or use source-checkout Nix recipes still use mvm’s builder boundary. On macOS that means the project builder VM owns Linux/Nix work; on Linux the command may use the native Linux path or the builder boundary depending on the workflow.

Network Protocol Scope

Machine networking is default-deny. --net enables dev-tier egress, and --allow-host HOST[:PORT] narrows the allowed destinations.

Current common-path policy is host/port-oriented TCP egress. Do not assume ICMP, raw sockets, multicast, arbitrary L2 behavior, inbound services, or transparent :80/:443 interception are available on every backend. When a guide depends on a backend-specific network behavior, it must name that backend.

Volume Shapes

Volumes are explicit host shares. They default to read-only unless a workflow spells out writable access.

Do not treat arbitrary host paths, device nodes, nested mounts, special files, or symlink traversal as portable volume features. Production inputs should be declared host-side and promoted into builds or artifacts explicitly; mutable dev state inside a persistent machine is not automatically a production input.

No SSH, On Any Tier

There is no SSH capability of any kind in a machine, on any profile — including dev/permissive. Private key files, ~/.ssh, known-hosts files, SSH servers, SSH clients, SSH config, and host ssh-agent forwarding are never copied, mounted, or bridged into a guest. TCP/22 is blocked on the default machine path. The console PTY-over-vsock transport (machine exec -it, machine shell) is the only interactive path into a dev-tier machine.

macOS Signing And Entitlements

Apple Silicon macOS uses the supported macOS virtualization path. Some behavior depends on host OS version, Apple virtualization framework availability, and the local signed binary/entitlement posture.

If a command fails only on macOS, include mvmctl doctor, the macOS version, CPU architecture, and backend in the report. Do not assume Intel macOS is a supported runtime target.

GPU Status

GPU passthrough or virtual GPU acceleration is not part of the default mvmctl machine surface. Do not depend on GPU access from a machine unless a future capability page documents the backend, admission, audit, and security semantics for that path.

Host And Guest Architecture Support

Supported host/runtime combinations are listed in Platform support. The common supported hosts are Linux with KVM and Apple Silicon macOS.

Guest image architecture must match the backend’s supported guest target. A portable artifact or OCI image that verifies on one architecture is not automatically runnable on another; machine check-artifact refuses host-arch mismatches.

Portable Artifact Preview

machine check-artifact verifies artifact signature, hash, format, and host architecture before deriving admission posture. That is not the same as a complete machine pack / machine run <artifact> product workflow.

Until the pack/run workflow lands, keep transfer, cleanup, and lower-level artifact operations in advanced docs and do not present .mvm artifacts as self-executing bypass blobs.