ArcBox is a fast, open-source container and VM runtime for macOS that runs AI agents on isolated machines with their own kernel, filesystem, and network.
ArcBox addresses the need for a performant, open-source alternative to closed-source container runtimes on macOS. It solves this by providing a unified runtime written from scratch in Rust, implementing its own virtual machine monitor, VirtIO devices, filesystem sharing, and network datapath. The tool consolidates four workload types—containers with Docker compatibility, disposable sandboxes for AI agents and untrusted code, full Linux VMs, and macOS guest VMs—under a single daemon and CLI. Containers expose a Docker-compatible socket proxying to a guest dockerd, supporting images for both amd64 and arm64 architectures, with x86-64 translation handled transparently via FEX emulation. The runtime includes native Kubernetes through a managed k3s cluster and offers container data browsable in Finder via NFSv4 over vsock, eliminating the need for docker cp commands.
Developers migrating from Docker Desktop or OrbStack will find ArcBox particularly relevant if they value open-source tooling and want to avoid vendor lock-in. The project suits teams building AI agent infrastructure locally before scaling to cloud deployments, since the same sandbox primitives run both locally and on ArcBox Platform. The tool is positioned as matching OrbStack's performance and overhead characteristics while remaining open source under MIT and Apache-2.0 licenses. Migration from existing Docker Desktop or OrbStack installations is supported, transferring images with all tags, named volumes, user-defined networks, and container state, with running containers automatically restarted unless explicitly prevented.
The project is in public beta with active community engagement through Discord and GitHub issues. Development activity shows ongoing work across the full stack of container runtime components, with particular focus on compatibility features like Docker CLI support, BuildKit integration, and Kubernetes integration. The codebase maintains comprehensive command-line tooling through abctl with diagnostic capabilities like abctl doctor and resource monitoring via abctl top.
