smol machines 2026: shipping a whole Linux VM that boots in under 200ms

smol machines builds smolvm, an open-source runtime that runs each Linux workload in its own hardware-virtualized microVM with a separate guest kernel on Hypervisor.framework (macOS), KVM (Linux), or the Windows Hypervisor Platform. The README advertises sub-second cold starts, elastic memory via virtio balloon, and the ability to pack a stateful machine into a single .smolmachine file that boots anywhere the host architecture matches.

BinBin He started the project from the classic 'works on my machine' gap: containers share the host kernel and isolate weakly, while QEMU-style VMs take tens of seconds to boot. He posted smol machines as a Show HN on 2026-04-17, arguing that a full VM could become as convenient as a container if the hypervisor were a library inside one binary and OCI images booted without a Docker daemon.

The post drew 499 points and 152 comments, with the discussion focusing on sandboxing untrusted code, AI agents, and Docker or Kubernetes workloads inside microVMs. The company joined Y Combinator's Spring 2026 batch, and engineering posts through mid-2026 showed a real k3d/k3s cluster running inside one machine, plus GPU and CUDA-remoting support.

As of 2026-09-02 smol machines is active: the Apache-2.0 runtime is public on GitHub and the same .smolmachine artifact can deploy to a hosted smol cloud or self-hosted infrastructure. The company frames itself for agent sandboxes and automated untrusted-code execution rather than as a general replacement for Docker.