Apple's Harbor app aims to make local AI servers on Macs simple

What Harbor is reported to do

Apple is working on an app, currently called Harbor, that turns a Mac into a private AI inference server, according to AppleInsider. An inference server runs large language models, the systems behind chatbots and text generation, on local hardware instead of in the cloud.

The report credits the discovery to a post by Aaronp613 on X. Harbor was previously known under the name Fabric, and it is designed to run open-weight models, meaning models whose weights are publicly available, directly on Apple silicon.

According to the description quoted by AppleInsider, data processed through Harbor never leaves the hardware and nothing reaches the cloud. The app is described as very user-friendly, aimed at people who want local AI without technical setup.

Clustering Macs over Thunderbolt 5

Harbor's main feature for larger workloads is clustering. Two or more Macs with Thunderbolt 5 can be connected so that their unified memory, the shared pool used by Apple silicon processors, is combined into a single inference resource.

This allows users to run models that exceed the capacity of any single machine, or to serve more simultaneous requests across the cluster. Harbor automatically detects connected devices over Thunderbolt 5 and handles the connections between them.

Setup is described as taking minutes rather than hours. The report says there is no manual configuration, no command line work and no need for infrastructure skills. Users can browse, download and load models from within the app itself.

Details about cluster limits remain unclear. AppleInsider's readers discussed whether clusters are limited by the number of Thunderbolt 5 ports on a host Mac, or whether machines can be daisy chained into larger arrangements. One reader noted that full mesh connections between Mac Studios, using two cables per link, could support a four-machine cluster. These figures come from reader discussion and are not confirmed.

Availability and the Claris brand

Harbor is currently in a private preview under Apple's Claris brand, the name Apple uses for its low-code development tools business. Readers should not expect a public release soon, as AppleInsider notes the app is unlikely to appear in the near term.

Because the information comes from a leaked or spotted preview rather than an official announcement, Apple has not confirmed the app, its features or any release plans. Details could change before any public launch.

How Harbor fits Apple's wider AI work

Harbor is not Apple's first step towards local AI on the Mac. The company already offers MLX, its machine learning framework, which can cluster devices over Thunderbolt networking. Apple later updated macOS and MLX to add Thunderbolt 5 support and GPU-based neural accelerator features.

In October 2026, Apple also detailed a system called Minions, which uses idle GPUs in Apple silicon Macs to form a dynamic virtual GPU cloud.

According to AppleInsider, Harbor appears focused on ease of use rather than raw performance. It may not make individual queries faster, but it should make the whole process of setting up a local AI server far simpler.

Why it matters

Running AI models locally appeals to people who do not want to send their data to cloud providers such as OpenAI or Anthropic. A local server keeps information on the user's own hardware.

Until now, that route required significant investment in hardware and considerable technical effort to configure. If Harbor works as described, it could lower both barriers, letting Mac owners with several machines pool their memory for demanding models.

For developers and small teams already invested in Apple silicon, the app could make private AI deployment practical without cloud costs or specialist knowledge. Whether it ships widely, and when, remains an open question.