Apple Harbor app enables private local AI inference on Mac

Apple is preparing Harbor, a new application designed to streamline local AI model execution on Mac hardware. The app transforms a Mac into a private inference server running open-weight models directly on Apple silicon, keeping user data on-device and avoiding cloud transmission entirely.

Harbor’s core innovation is resource pooling across multiple machines. Macs connected via Thunderbolt 5 automatically cluster together, their unified memory aggregated into a shared inference resource. This allows users to run models exceeding any single machine’s capacity or serve multiple concurrent requests across pooled hardware. Setup requires minimal effort—the app handles device discovery, model management, and deployment without command-line knowledge or infrastructure expertise.

Currently in private preview through Apple’s Claris division, Harbor represents Apple’s push to make on-device AI computation accessible as the Mac gains ground as a platform for local AI workloads.