Super Rad's Microsandbox Gives AI Agents Hardware-Level Isolation at Docker Speed

Microsandbox is an open-source, local-first microVM runtime built specifically for running untrusted workloads , AI agent code, user-submitted scripts, plugins, CI jobs , with hardware-level isolation and a developer experience that feels like Docker. Built in Rust by Super Rad Company (a YC-backed startup), it has crossed 6,700 GitHub stars and is shipping releases at a fast clip, currently at v0.6.1.

The container escape problem nobody wants to talk about

The thesis behind microsandbox is that containers are insufficient for running truly untrusted AI-generated code , shared kernels mean sophisticated attacks can escape. This is not a theoretical concern. Containers share a fatal flaw: the host kernel is a shared attack surface. One container escape compromises everything. As AI agents gain the ability to write and execute arbitrary code, this attack surface becomes a live production risk.

The emergence of microVM technologies like Firecracker and libkrun has rendered the old dichotomy of "slow, secure VMs versus fast, insecure containers" largely obsolete. They offer a compelling "best of both worlds" approach that has become the new standard for high-security, ephemeral code execution. Microsandbox is betting that this technology should be as easy to use as docker run.

What's under the hood

Microsandbox uses libkrun, a lightweight virtualization library from the containers project that leverages KVM (Linux) or Hypervisor.framework (macOS) for hardware isolation. Unlike Firecracker (used by E2B and AWS Lambda), libkrun includes its own Virtual Machine Monitor and uses Transparent Socket Impersonation (TSI) for networking , avoiding the overhead of virtual network interfaces.

Average boot times are under 100 milliseconds. It boots faster than Docker on bare-metal Linux, with full hardware isolation. Each sandbox gets its own Linux kernel, filesystem, and network boundary , not just namespaces on a shared host kernel.