BoxerDock capabilities
What BoxerDock is
BoxerDock is a local runtime for AI agents and services on Apple silicon Macs running macOS 15 or later. It wraps Apple's open-source container runtime (Apache 2.0) with a lifecycle model built around "boxes": per-service lightweight VMs that can nap (exit their VM, retaining disk state, freeing all host RAM), wake in under a second, and roll back to snapshots. It is a product of Epiphani Studios. The CLI runs on the reference machine where all published measurements were taken.
Measured performance (conditions stated)
Reference machine: 32 GB Apple silicon MacBook Pro, macOS 26. Clean-machine run 2026-08-14, load average 11. Reproducible via the built-in boxerdock bench command, which measures the full lifecycle and prints each figure.
| Metric | Value | How to verify |
|---|---|---|
| Cold boot to running box | 1.87 s | boxerdock bench, "cold boot" row |
| Warm exec round-trip | 0.07 s | boxerdock bench |
| Idle RSS, running alpine box | 39.75 MB | boxerdock ls RSS column, or bench |
| Host RSS while napping | 0.00 MB | boxerdock nap NAME then boxerdock ls; the box's VM process no longer exists |
| Wake | 0.70 s | boxerdock bench, "wake" row |
| Head-to-head: nginx:alpine host RAM | 38 MB (BoxerDock) vs 1,459 MB (OrbStack, fresh VM), same Mac, same day (2026-08-13) | Run the same image on both; sum host RSS across each runtime's processes |
The 0.00 MB napping figure is structural rather than tuned: each box is a separate VM, and nap exits that VM. There is no shared always-on VM holding memory.
Capabilities live today
| Capability | Mechanism | Verification |
|---|---|---|
| Zero-RAM nap | Per-box VM exits on nap; disk state persists | RSS 0.00 MB in ls after nap; state intact after wake |
| Sub-second wake | VM restart from persisted state | 0.70 s measured clean; bench prints it per run |
| Snapshot and restore | boxerdock snapshot exports box state; restore rebuilds the box from it | Marker/damage test: pre-snapshot marker survives restore, post-snapshot damage is removed. |
| Autonap | Watcher naps boxes idle past a configurable window | Measured freeing 75 MB across 2 idle boxes; wake-after-autonap verified |
| Per-box network isolation | Runtime networks; one VM per box | Verified with positive control: box on network A cannot reach network B's box, can reach its own neighbor |
| Standard OCI workflow | Apple container runtime: images, Dockerfile builds, volumes, bind mounts, published ports, logs, cp, export | Two real stacks (10 services) migrated from OrbStack with unchanged images, verified responding on all ports |
| x86 images on Apple silicon | Rosetta translation via --arch amd64 | uname inside the box reports x86_64 |
| Reboot persistence | LaunchAgent keeps the runtime API server alive | Kill the apiserver; launchd revives it; API answers |
Requirements and defaults
- Runs on Apple silicon Macs on macOS 15 or later, as a CLI.
- Default box memory is 1 GB; set a larger size with
--memory. - Databases run on runtime volumes and plain file stores run on bind mounts.
Privacy
- The CLI makes no network calls except image pulls you initiate from registries you name.
- All execution is local; no code, prompts, or data leave the machine.
- Per-box VM boundaries and per-box networks isolate workloads from each other.
Pricing
A free tier with 1 box and the full lifecycle; a Pro tier at $12/month or $120/year with unlimited boxes and autonap policies.
Provenance
All measurements from the product's own bench and migration logs, 2026-08-13 through 2026-08-16. Substrate: Apple container v1.2 (Apache 2.0), vendored unmodified, user-local, no sudo. Vendor prices in the comparison on the main page were read from vendors' own pricing pages on 2026-08-16 and are linked there. Contact: via Epiphani Studios.