By Hamza Faisal

Apple is quietly testing an app that turns a Mac into a private AI server. Called Harbor, the software runs open-weight AI models entirely on Apple silicon hardware, keeping user data on the device and sending nothing to the cloud, according to a leak reported by 9to5Mac on November 10.

The app is in a private preview under Apple's Claris subsidiary, the enterprise software unit best known for the FileMaker database platform. An App Store listing for Harbor briefly appeared before Apple pulled it from view, and a sign-up page for the early preview asks applicants for a company name — a hint that Harbor is aimed at businesses first.

What Harbor does

Harbor's pitch is simple: your Mac becomes your own AI inference server. Users can browse, download and load open-weight models directly inside the app, which handles model file management for them. There is no command line, no manual configuration and no cloud account standing between the user and the model.

Because everything runs locally, prompts, documents and other data never leave the machine. That matters for teams working with sensitive material — legal files, medical records, unreleased code — where sending data to a cloud provider is either a security risk or a compliance problem.

Thunderbolt 5 turns Macs into a cluster

Harbor's standout feature is clustering. A single Mac can run lighter models on its own, but Harbor can also link two or more Thunderbolt 5-capable Macs into one inference resource, pooling their unified memory to host models too large for any single machine or to handle more requests at once. The app automatically detects nearby hardware over Thunderbolt 5 and forms a cluster in minutes.

The feature leans on macOS Tahoe 26.2, which added Remote Direct Memory Access over Thunderbolt 5 to speed data transfers between connected Macs. It is a notable use of hardware Apple already ships: the Mac Studio and Mac mini have become popular machines for running models on premises, and Harbor would give that trend a first-party, graphical front end — something like LM Studio's simplicity combined with the multi-machine clustering of open-source tools such as Exo.

What Apple hasn't said yet

Apple has said nothing about Harbor publicly, and neither the company nor Claris has announced pricing or a release date. 9to5Mac reported that a public launch seems likely in the weeks and months ahead, while AppleInsider cautioned that end users should not expect the app anytime soon. Commercial details remain undetermined, and it is unclear whether Harbor will be free or sold as a Claris product.

Even so, the leak fits a pattern. Apple chief executive John Ternus warned investors on November 3 that hardware expenses are accelerating as technology companies compete with AI data centres for memory supplies. A tool that lets businesses run capable models on Macs they already own — instead of renting cloud inference — would be a timely answer to that squeeze. For now, Harbor remains a private preview, but the brief appearance of its App Store listing suggests Apple is closer to showing it off than to keeping it secret.