smolvm

smolvm runs Linux microVMs on your laptop with hardware isolation. Apache 2.0, sub-second cold start, single-binary CLI. Pull any OCI image (Docker Hub, ghcr.io, no Docker daemon required), boot it as a tiny VM with its own kernel, run code in it, throw it away. The pitch: containers share your kernel; this doesn't. Real boundary.

macOS uses Hypervisor.framework, Linux uses KVM. Network is off by default and you whitelist hosts explicitly. SSH agent forwarding works without leaking your keys into the guest. The 'pack into portable executable' feature creates a self-contained binary with the VM baked in, no install step on the target machine.

Anyone running untrusted code (AI agent outputs, user-submitted snippets, build steps that pull random dependencies): this gives you a real isolation boundary, not the container kind. Local development with reproducible Smolfile configs: also good. Production workloads at scale: you want Firecracker or libkrun directly.

It's local-only and the project is young. The 'pack into binary' workflow is the differentiator, and it depends on libkrunfw, a custom kernel. If that stops getting maintained, the killer feature stops working.