Apple could be working on a Mac app called Harbor that turns the machine into a private AI inference server, according to a leak from Aaron Perris that 9to5Mac reported on November 10, 2026. The Harbor app is in a private preview under Apple’s Claris brand, and Apple has not announced it publicly.
The idea is local first: open-weight models run on Apple silicon, the data stays on the hardware, and nothing is sent to the cloud.
What the Harbor app does
Setup is meant to take minutes rather than hours, because the app detects Thunderbolt 5 devices on its own and forms a cluster without manual configuration or a command line.
Models are browsed, downloaded and loaded inside the app, with Harbor handling model management.
And the cluster is the part that matters for large models: two or more Thunderbolt 5-capable Macs pool their unified memory into one inference resource, which lets the setup run models bigger than any single machine could hold, or serve more concurrent requests.
Why Apple would build it
9to5Mac places the app against the Mac’s year: the lineup has grown more popular, both through budget options like the MacBook Neo and as hardware for running AI models locally on a Mac Studio or Mac mini. Harbor fits that second use, packaging cluster setup and model downloads into an app instead of a set of commands.
What is unconfirmed
Apple has said nothing about Harbor, no release date has surfaced, and the private preview is the only confirmed stage.
Source: 9to5Mac
Frequently Asked Questions (FAQ)
An unannounced Apple Mac app that runs open-weight AI models locally and turns the machine into a private inference server.
No. The leak describes inference on the Mac itself, with data staying on the hardware.
Yes. Two or more Thunderbolt 5-capable Macs can pool their unified memory into a single inference resource.
Not publicly. It is in a private preview under Apple’s Claris brand.
Two or more Thunderbolt 5-capable Macs can pool their unified memory into one inference resource.

