Smol machines – Sub-second cold start and portable virtual machines

smolvm is a CLI-based virtual machine management tool that runs on macOS and Linux. It's as lightweight to use as a container, but runs each workload as a microVM with its own independent kernel, providing hardware-level isolation.

The core goals are threefold.

• Sub-second cold start: under 1 second, with the introduction also mentioning boot times under 200ms.
• Portability: VM state is packaged into a single .smolmachine file, allowing it to be re-run without dependencies on platforms with the same architecture.
• Operational convenience: boots OCI images directly without a Docker daemon, and manages environments declaratively via Smolfile.

Its execution model runs a libkrun-based VMM and custom kernel (libkrunfw) on top of Hypervisor.framework (macOS) or KVM (Linux). The default configuration is 4 vCPU, 8GiB RAM, adjustable via --cpus and --mem.

Security and development workflows are also addressed.

• Networking disabled by default, controlled via --net or --allow-host
• SSH agent forwarding supported, with private keys never copied to the guest
• Smolfile (TOML) declares image, network, init commands, volumes, and auth options
• Packaged VMs are designed to be runnable instantly without installation

The comparison table emphasizes that smolvm is an alternative aiming to combine the usability of containers with the isolation of VMs. In particular, native macOS support, built-in SDK, and portable artifact aspects are presented as differentiators.

The limitations are also clear.

• Networking is off by default
• Only TCP/UDP supported, no ICMP support
• Volume mounts are directory-level only
• On macOS, only binaries signed for Hypervisor.framework can run
• Using --ssh-agent requires an SSH agent on the host

In summary, rather than being a Docker replacement, it's closer to a portable microVM that starts quickly for development, sandboxing, and deployment, proposing a different direction from traditional container ecosystems like Docker Compose or nested VMs.