Apple Utvikler Harbor for Å Kjøre AI-modeller Lokalt På Flere Mac-er

Connecting Macs Over Thunderbolt 5

Apple is developing a new application called Harbor that enables users to link multiple Mac computers together to run large artificial intelligence models locally, ITavisen reported.

Mac leakster Aaron first revealed the macOS platform, stating that the software is currently undergoing closed testing, according to ITavisen. The system utilizes Thunderbolt 5 to pool unified memory across separate devices, allowing two machines with 64 gigabytes of memory each to combine for 128 gigabytes of total capacity.

ITavisen reported that the application automatically detects connected Mac hardware without requiring manual configuration or command-line inputs. The development builds on Apple’s ongoing hardware investments in Apple silicon architecture, where the central processing unit and graphics processing unit share the same memory pool.

Running Open Models Locally Without Cloud Servers

Harbor includes a built-in interface for discovering, downloading, and executing open-weight AI models directly on user hardware, ITavisen reported. This design keeps all personal data on the local machine rather than routing information to external cloud servers operated by external providers.

ITavisen noted that the software positions Apple alongside existing local AI tools such as LM Studio and Ollama. Project testing is currently taking place using the Claris brand name, a historical Apple subsidiary originally known as FileMaker Inc.

Common Questions About Harbor

What hardware is required to run Harbor?

The application is designed for Apple silicon computers connected via Thunderbolt 5 ports. It pools unified memory and neural processing unit capabilities across multiple devices, such as pairing a MacBook Pro with a Mac Studio.

Are user data and prompts sent to the cloud?

According to ITavisen, models run entirely on local hardware, ensuring data does not leave the user’s device. Specific details regarding pricing, hybrid cloud support, and model compatibility remain unannounced.