Isolated, forkable computers for your AI agents.
Millisecond microVM sandbox forking on Kubernetes.

Website  ·  Core repo  ·  Quickstart  ·  Discussions

import mitos

sb = mitos.create("python")              # Ready microVM sandbox (~27 ms warm-claim)
print(sb.exec("echo hello").stdout)      # hello

# Fork into independent siblings to try two approaches at once.
a, b = sb.fork(2)
a.exec("echo conservative > /workspace/plan.txt")
b.exec("echo aggressive  > /workspace/plan.txt")

sb.terminate()

Set MITOS_API_KEY from mitos.run. The base URL defaults to the hosted endpoint, so no Kubernetes required. The same code runs against a self-hosted cluster by setting MITOS_BASE_URL.

• ⑂ Live-fork a running VM. N-way copy-on-write fork of a live microVM: daughters share the parent's memory pages until they write, so each fork lands in a warm, ready environment. Branch one agent into many parallel attempts.
• ⚡ ~27 ms warm-claim activate. Firecracker microVMs restore from a memory snapshot in the tens-of-milliseconds class. P50 ~27 ms on the bare-metal reference node, reproducible from a script in the repo.
• ◎ Open-source, self-hostable, Kubernetes-native. As far as we know, the only runtime that does all three. Run it hosted, or on your own KVM cluster where your data never leaves your infrastructure.

pip install mitos-run          # the import stays `import mitos`
export MITOS_API_KEY=sk-...

P50 ~27 ms warm-claim activate (snapshot load + fork-correctness handshake + guest-ready) on the bare-metal reference node, reproducible from bench/husk-activate-latency.sh. Full methodology in BENCHMARKS.md.

• mitos-run/mitos: the core runtime, SDKs (Python · TypeScript · Go), and mitos.run CRDs.
• Quickstart: from pip install to a forked swarm of subagents.
• Migrating from E2B: drop-in path for existing agent harnesses.
• Roadmap · Architecture

We build in the open. Stars and issues genuinely help us prioritize.

⭐ Star the repo · Good first issues · Contributing · Security

_{Apache-2.0 · Mitos™ · trademarks.}