microsandbox
Use as the OSS self-hosted sandbox when you need to run agent code on your own infra — libkrun-based microVM isolation with no per-sandbox vendor cost, the escape hatch from a managed runtime at high volume.
- Score
- 99.1002 signals
- Evidence
- 8,436 stars · 454 forks
- Last commit
- as last read from GitHub; most reads are from 2 Sep 2026 or later
- Listed
Install
No one-command install. Set it up from its source.
Alternatives · Sandbox
- claude-managed-agents-selfhost3,876 stars · 836 forks98.932
What it is
An open-source, self-hostable sandbox using libkrun for microVM-level isolation. It runs agent-generated code on infrastructure you control, avoiding per-sandbox cloud fees. It is the OSS counterpart to a managed Firecracker sandbox for teams that want to own the runtime.
When to use it
When data residency, cost at high volume, or a no-vendor-lock-in requirement rules out a managed sandbox. The trigger is "we need our own sandbox infra, self-hosted." Past a break-even of roughly thousands of sandbox-hours a month, self-hosting wins on cost.
How to install / invoke
Self-host the sandbox per its deployment docs and point the agent's execution layer at it instead of a managed provider. The code path the agent uses stays largely the same.
Notes
Trade managed convenience for control and cost. Daytona is a comparable OSS alternative. Below the break-even volume, a managed sandbox is usually the simpler and cheaper choice.
Related
claude-managed-agents-selfhost
Use for enterprise hosted-control agents — the agent loop runs at the provider while execution happens on your own infrastructure — when you want a managed control plane but data and code must stay on your machines.