arcboxlabs/arcbox

Run AI agents on real and isolated machines — own kernel, filesystem, and network — with <100ms boot. Local first, OCI compatible, pure Rust.

What it solves

ArcBox is an open-source, high-performance container and VM runtime for macOS, designed as an alternative to Docker Desktop and OrbStack. It provides a unified runtime for Docker containers, Kubernetes clusters, Linux VMs, and macOS guests, while minimizing overhead and maximizing speed.

How it works

Built from scratch in Rust, ArcBox implements its own Virtual Machine Monitor (VMM), VirtIO devices, and a userspace network datapath. It supports two hypervisor backends (Hypervisor.framework and Virtualization.framework). For Apple Silicon, it uses FEX to translate x86-64 images to run on arm64. It exposes a Docker-compatible socket to act as a drop-in replacement for the Docker engine.

Who it’s for

Developers on macOS who need a fast, open-source alternative to Docker and VM management tools, as well as developers building AI agents that require isolated, disposable microVM sandboxes.

Highlights

  • Multi-workload support: Runs Docker containers, native Kubernetes (k3s), full Linux VMs, and macOS guests on a single runtime.
  • Agent Sandboxes: Provides disposable microVMs (via Firecracker) for AI agents like Claude Code, ensuring isolation from the host.
  • High Performance: Custom network datapath and VMM for low cold-boot times and high throughput.
  • Finder Integration: Guest Docker data is mounted read-only on the host at ~/ArcBox for easy browsing.
  • Cross-Architecture: Runs linux/amd64 images on Apple Silicon via FEX translation.

Related

  • Project
  • Project
  • Project
  • Project
  • Project