Apple's new Harbor app turns Macs into private local AI servers

What Harbor is

Apple appears to be working on a Mac app called Harbor, according to 9to5 Mac. Leaker Aaron Perris found evidence of the app, which is described as turning a Mac into a private AI inference server. Inference is the process of running a trained AI model to generate responses.

Harbor is designed to run open-weight models, which are AI models whose weights are publicly available, directly on Apple silicon. Apple's own description states that data never leaves the hardware and nothing reaches the cloud.

The app is currently in a private preview under Apple's Claris brand. 9to5 Mac reports that a public launch in the weeks and months ahead seems likely, though Apple has not confirmed any release plans.

Clustering Macs over Thunderbolt 5

The most notable feature is clustering. Harbor can connect two or more Mac computers that support Thunderbolt 5 and pool their unified memory into a single inference resource.

According to the description Perris surfaced, this allows users to run models that exceed the capacity of any single machine, or to serve more concurrent requests across the cluster. It effectively lets several Macs behave as one larger AI server.

Setup is meant to be simple. Apple's materials say Harbor automatically detects devices over Thunderbolt 5 to form a cluster, with no manual configuration, no command line and no infrastructure expertise required.

Model management inside the app

Harbor also handles the model side of the workflow. Users can browse, download and load open-weight models from within the app itself.

Apple's description says the app manages the models so users can focus on their work rather than setup. This suggests a consumer-friendly alternative to command-line tools that are commonly used today to run local models on Macs.

No specific models, supported Macs or pricing details have been confirmed. Those points remain unverified.

Why it matters

9to5 Mac notes that the Mac has grown more popular this year, both because of budget-friendly options and because machines such as the Mac Studio and Mac mini are used to run AI models locally.

Harbor would see Apple lean harder into that use case. A supported, first-party tool for pooling Macs into an inference cluster could appeal to developers and small teams who want private AI without cloud services.

Since the app is only known from leaked evidence and a private preview, readers should treat the details as unconfirmed until Apple makes an official announcement.