Apple is developing a dedicated application called Harbor that can run AI models directly on Macs without relying on cloud servers. The core feature is support for linking multiple Macs and pooling memory resources so personal computers can be used as AI servers.

On Nov. 12, Japanese media outlet Gigazine reported that Apple unit Claris is currently testing Harbor as a private preview. Claris develops the FileMaker database software.

Harbor is designed to run open-weight AI models on Apple silicon Macs. AI computation and data processing take place on the device, and user information is not sent to external clouds. Claris said it can be set up in minutes without entering separate commands, and there are no cloud fees or per-token charges.

A key feature is clustering that links multiple Macs to work like a single AI server. When two or more Thunderbolt 5-supported Macs are connected, they can use the unified memory of each device together. This makes it possible to run large AI models that are difficult to handle on a single Mac or to process multiple requests at the same time.

Screens shared by Apple-related news source Aaron Perris (에런 페리스) showed a download of OpenAI's 'gpt-oss-20b' model. The images also showed a screen running an AI model on an M5-powered MacBook Pro and an interface managing three Macs as a single cluster.

Harbor is not currently available to general users. It was briefly listed on the Apple App Store and then removed, and the official release schedule, price and supported models have not been disclosed.

Apple previously released research results in October on Minions, a distributed AI system that uses idle GPU resources across multiple Macs. While Minions focuses on distributing GPU resources, Harbor focuses on easily connecting multiple Macs to build an AI inference server.

If Harbor is officially released, it is expected to become a new option for companies and developers to run AI models on their own hardware.