MacFleet

Open source · MIT License

Distributed compute across your Apple Silicon Macs

MacFleet turns a room of Macs into one training cluster. Zero-config discovery, ring AllReduce gradient sync, and HMAC-authenticated fleet isolation — so you can train models instead of wiring infrastructure.

Get it on PyPIRead the docs
$pip install macfleet

Capabilities

Built for the hard parts

The networking, discovery, and security primitives you need to run distributed ML on macOS.

Pool Apple Silicon

Unify M-series Macs into a single compute fabric and train larger models than any one machine allows.

Zero-config discovery

Machines find each other on the local network automatically. No manual host lists or static IPs.

Ring AllReduce sync

Efficient gradient synchronization keeps distributed PyTorch and MLX training fast and stable.

HMAC + TLS isolation

Every fleet is cryptographically isolated so your cluster and your data stay private.

Getting started

Start a fleet in three steps

  1. 01

    Install macfleet on every Mac with pip.

  2. 02

    Run macfleet init on one machine to create a fleet.

  3. 03

    Train with PyTorch or MLX using the fleet as one cluster.