NATIVE CONTAINERS FOR APPLE SILICON

The Docker your Mac was waiting for.

Build Apple apps and run AI on the GPU of a Mac mini M4 — no virtual machine, no emulation, real macOS.

🔥

Native GPUPyTorch (MPS) and MLX using the chip’s GPU and unified memory.

🛠️

Xcode in a containerUpload your project .zip and get back a .dmg or .ipa signed with your certificate.

⚡

Container in ~130 msEach job starts from an APFS clone of the image.

🔒

Isolated by defaultmacOS sandbox, a dedicated service user and no network while the job runs.

Sign in to MacDocker

Use the same account you signed up with. No account yet? Create one.

Loading…

Create account

Choose how you want to start. Sign-in is through GitHub, Google or Apple: we do not create passwords.

Beta tester MANUAL APPROVAL 90 days with the Individual plan quota. Your request joins the queue and is approved by our team. Trial INSTANT ACCESS 7 days and 15 minutes of usage. The account ends when the period ends, even if minutes are left. Plans WAITING LIST Individual or Corporate. Subscriptions are not open yet: you join the waiting list and are notified when they open.

Plan

Name (optional)

Company or project (optional)

What do you want to run

How did you hear about us

Tell us what you plan to test (optional, helps with approval) I have read and accept the Terms of Service and the Privacy Policy.

🧪Development mode. No OAuth provider is configured: the buttons create a local test session.

Uploaded code runs isolated and is deleted at the end of the retention period. Signing certificates are destroyed at the end of every build.

NATIVE GPU

Your M4 GPU, inside the container.

Docker on the Mac runs a Linux VM — and the VM cannot see Metal. Here PyTorch (MPS) and MLX talk directly to the GPU and unified memory.

DOCKER ON MACCPU in a Linux VM.
No Metal, no MPS.

MACDOCKERApple Silicon GPU in the job,
inside the sandbox.

app.macdocker.com:8000/app/#/containers/mnist-gpu-training/stats

Stats tab of a GPU training container: charts for CPU, memory (including the GPU), processes and elapsed time of a completed run.

gpu_training.py · python-torch:3 · container output
PyTorch 2.14.0 | device: mps | MPS available: True

PyTorch and MLX on the Apple Silicon GPU. Create account →

XCODE IN A CONTAINER

Build Apple apps in a container. Yes, you can now.

Upload your project .zip and get back a .dmg or .ipa signed with your certificate. Linux can’t run Xcode — MacDocker runs real macOS.

✓ Aurora.dmg ready in 18 s · sample app

app.macdocker.com:8000/app/#/builds

Builds screen of the dashboard: a completed Aurora build with the xcode:26.5 image, producing an Aurora.dmg package available for 30 days.

Sandboxed build, no networkSwift packages resolved in an isolated phaseYour certificate, your app

macOS and iOS builds signed with your certificate. Create account →

ISOLATION AND LIMITS

Truly isolated. Truly fast.

app.macdocker.com:8000/app/#/containers/mnist-gpu-training/inspect

Inspect tab of a container: ID, name, image, script, status and exit code, plus the limits (maximum time, 8 GB of memory including the GPU, 64 processes, network blocked) and the run’s peak usage.

Sandbox

macOS Seatbelt on every job, with its own service user

No network

Network blocked while running

Watchdog

Memory (including the GPU), processes and time are capped

~130 ms

The container starts from a copy-on-write APFS clone

Each customer in their own service account, everything audited in a chained log — and without the cost of a VM per job.

Every job isolated, without the cost of a VM. Create account →

ROADMAP

The best is yet to come.

⚙️
Runner for GitHub ActionsCI in MacDocker containers
🖥️
Many Macs, one dashboardMulti-node scaling
🎬
ffmpeg + VideoToolboxRender video with hardware
🧠
python-mlx-lmLocal LLMs on the GPU, with MLX
🔏
Automatic notarizationYour app ready for Gatekeeper
📦
Set up your own nodeSetup assistant + signed images

app.macdocker.com:8000/app

Dashboard Hub with the available images (python-torch, python-mlx and xcode) and those coming soon: python-transformers, xcode 26.4 and ffmpeg-videotoolbox.

COMING SOON COMING SOON COMING SOON

Beta testers help choose what comes first. Request beta access →

PLANS

Choose how to get started

Beta tester and trial are free. Security does not change between plans: everything runs isolated, with no network during the job, and signing credentials are erased at the end of every build.

Swipe the table sideways to see Trial, Individual and Corporate →

Loading…

* The Individual and Corporate plans are not on sale yet: they will only be sold after the beta period ends, and the prices above are estimates that may change before launch. If you want to be notified, join the launch waiting list.

† Weighted tokens: an output token is worth the model's number of billions of parameters, and an input token a quarter of that (Qwen3-4B uses 4 per generated token; Qwen2.5-0.5B, 0.5). Separate from usage minutes.