
smolvm
Tool to build & run portable, lightweight, self-contained virtual machines.
The Lens
By Erik Loyd, SaaS CEO and former COO/CFO of an AWS Premier Partner.
Updated Aug 2026
smolvm boots a Linux microVM on your laptop in under a second and throws it away when you are done. Pull any OCI image from Docker Hub or ghcr.io with no Docker daemon involved, boot it as a real VM with its own kernel, run untrusted code inside it. The pitch is the boundary: containers share your kernel, this does not. Apache 2.0, single-binary CLI.
Platform coverage has grown past the original macOS and Linux story: Apple Silicon and Intel Macs, Linux on x86_64 and aarch64, and Windows through the Windows Hypervisor Platform. Network is off by default, with per-host egress allowlists once you enable it, which is the correct default for a sandbox. GPU work runs through virtio-gpu with Vulkan, plus CUDA remoting for NVIDIA. Configured machines pack into portable .smolmachine artifacts, and snapshots push to an OCI registry.
This is the tool for running agent-generated code, untrusted builds, or anything you would rather not hand your kernel. Solo and small teams: the sub-second start makes it practical per task rather than per session. Medium and large teams get the same thing with one caveat: there is no orchestration story. It is a local CLI, not a fleet manager.
The catch is uneven platform parity. macOS needs code signing, Windows has neither GPU support nor snapshots, and volumes mount directories only. The docs also reference a "smolmachines cloud control plane" with nothing behind it yet, which usually signals a hosted tier coming. Nothing is gated today.
Free vs Self-Hosted vs Paid
fully freeWhat's Free
Everything shipped today. Apache 2.0. The CLI, OCI image pulling, microVM boot, snapshots, .smolmachine artifacts, egress allowlists, GPU passthrough, and CUDA remoting. No account, no license key, no gated feature.
Deployment Cost
$0. One-line install script or a binary download. It runs on hardware you already own and uses Hypervisor.framework on macOS, KVM on Linux, and the Windows Hypervisor Platform on Windows. No daemon, no registry account required for local images.
The Watch Item
A "smolmachines cloud control plane" is referenced in the documentation without a pricing page behind it. That is the usual shape of an open core play arriving later: local CLI free, fleet management paid. Nothing is restricted right now, and there is no announced tier, but it is worth knowing before you build tooling that assumes free forever.
vs Alternatives
- Docker: free for small teams, shares your kernel. Different security posture entirely.
- Firecracker: free, Apache 2.0, and the microVM engine behind AWS Lambda. Far more work to drive directly; smolvm is the ergonomic layer.
- Cloud sandboxes (E2B, Modal, Daytona): metered per second and remove the local hardware requirement. Pay when you need isolation somewhere other than your laptop.
Free and Apache 2.0 with nothing gated today. A cloud control plane is referenced in the docs but not shipped, so treat a future paid tier as possible rather than announced.
What to do by team size
- Solo
- free; the sub-second boot makes per-task sandboxing realistic
- Small team
- free; ideal for running agent-generated code without trusting it
- Medium team
- free; no orchestration layer, so expect to script around it
- Large team
- free locally, but you will still need a real sandbox fleet somewhere else
Get tools like this every Wednesday
One featured tool, three on the radar. No fluff.
A low score is not a verdict on quality. Young and niche tools start low by design. How we calculate scores
Trust Signals
License: Apache License 2.0
Use freely. Patent grant included.
Commercial use: ✓ Yes
About
- Owner
- smol machines (Organization)
- Stars
- 6,390
- Forks
- 311