#99th of 14 in Sandboxes
smolvm
MicroVM runtime with live checkpoints and branching for agent sandboxes
- Stars
- 7.1k
- License
- Apache-2.0
- Last commit
- Nov 2026
- Last release
- Nov 2026
- Language
- Rust
2,347 on 21 Apr 20267,077 on 10 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
smolvm runs each workload in its own libkrun microVM with a separate guest kernel, using Hypervisor.framework, KVM or Windows Hypervisor Platform. It boots any OCI image without a Docker daemon. It can checkpoint a running machine (RAM, CPU, disks), branch it copy-on-write, and pack it into a single .smolmachine executable. SDKs for Node, Python, Rust and Go embed it in-process.
Who it is for: Engineers who need isolated, snapshot-able VM sandboxes for agents
Strengths
- Per-workload VM with own kernel; network off by default, egress limited to named hosts
- Live checkpoint and copy-on-write branching of running machines, including RAM and CPU state
- SDKs for Node, Python, Rust and Go run in-process with no daemon
- Runs on macOS Apple Silicon, Linux (KVM) and Windows; packs to one .smolmachine file
Weaknesses
- No published macOS Intel build; that platform is untested
- GPU acceleration is unavailable on Windows
- Linux hosts need KVM (/dev/kvm); Windows needs Windows Hypervisor Platform enabled
- SDK source and user docs live in separate repositories
What it needs
- GPU optional
- Needs KVM (Linux), Hypervisor.framework (macOS), Windows Hypervisor Platform (Windows), libkrun
Quick answers
Is smolvm open source?
Yes. Its code is on GitHub under the Apache-2.0 license.
Can you run smolvm with Docker?
Yes. Its repository has a Docker setup: a Dockerfile, a Compose file or a published image.
Does smolvm need a GPU?
No. It runs on a CPU and can use a GPU when you have one.
Also in Sandboxes
See all 14| Rank | Project | Score |
|---|---|---|
| 1 | LightpandaHeadless browser written in Zig for agents and scraping, with CDP and MCP | 93 out of 100 |
| 2 | OpenShellPolicy-enforced sandbox runtime for autonomous AI agents | 88 out of 100 |
| 3 | ObscuraRust headless browser with V8 and CDP for scraping and agents | 86 out of 100 |
| #4 | NemoClawReference stack for running OpenClaw, Hermes and Deep Agents in OpenShell sandboxes | 85 out of 100 |
| #5 | opensandboxSandbox API for running AI agent code, commands and browsers in isolation | 85 out of 100 |