smolvm is an open-source runtime for branchable microVMs built for AI agents: it embeds lightweight virtual machines directly into your software so agents and untrusted code can execute inside a hardware-isolated Linux VM that boots in under 200 milliseconds. Written in Rust on top of libkrun, it is portable across macOS, Linux and Windows, and the same machine artifact runs locally, self-hosted, or in the smol cloud. The project has about 6,800 stars on GitHub and is under very active development in 2026.
Agents that run code need isolation stronger than a shared container kernel, but traditional virtual machines are slow to start and heavy to operate. smolvm links the virtual machine monitor into the binary itself, with no host daemon, so each workload gets its own guest kernel and hardware isolation through KVM, the macOS Hypervisor framework or the Windows Hypervisor Platform. Machines can be branched: fork a running machine into a batch of copies instead of booting from scratch, which makes it practical to give every reinforcement-learning rollout or agent task its own sandbox. Files, packages and workspace state persist across sessions, memory is elastic through virtio ballooning, and machines are packaged into portable .smolmachine artifacts so a setup built on a laptop runs identically in the cloud.
The smolvm runtime is free and open source under the Apache-2.0 license for local and self-hosted use. The hosted smol cloud offers one-call deployment with an account and free usage credit to start, on paid usage beyond that, which makes the overall offering freemium.
smolvm is for AI platform engineers and agent builders who need to execute model-generated code safely, run large numbers of isolated RL or evaluation environments, or ship portable development environments without the overhead of full virtual machines.