An application currently called “Harbor” will help turn multiple Macs into a private AI inference server, keeping data local while making the process of creating an AI server much easier.
The Mac has become a very useful tool for AI research, but it could be even more useful. Apple is working on an application that can make it easier to run large language models (LLMs) locally, provided you have enough processing power and available memory.
Formerly known as Fabric, Harbor is the application that creates a private AI inference server on a Mac, according to Aaronp613 on X. It is capable of running open weight models on the Mac itself and is apparently extremely user-friendly.
Apple is working on a new app called Harbor
Harbor turns your Mac into a private AI inference server. Run open models locally on Apple silicon—your data never leaves your hardware and nothing reaches the cloud. Connect multiple Mac computers via Thunderbolt 5 to pool unified memory, enabling larger models and higher concurrency than a single computer can handle on its own.
Installed in minutes, not hours, Harbor automatically detects your devices via Thunderbolt 5 to form a cluster. Plug in, boot up, and go: no manual setup, no command line, no infrastructure expertise required.
Download and run models directly Browse, download and load open weight models from within the app. Harbor handles model management so you can focus on work, not configuration.
Large-Scale Cluster Connect two or more Thunderbolt 5-enabled Mac computers and Harbor consolidates their unified memory into a single inference resource. Run models that exceed the capacity of any single computer or process more concurrent queries on your cluster.
Harbor is currently in private preview under the Claris brand.
— Aaron (@aaronp613) November 10, 2026
Installation takes just a few minutes, with seemingly no need to configure anything manually or use a command line. Harbor manages the model on behalf of the user, including directly managing downloads.
If there is more than one Mac available, they can be connected as a cluster via Thunderbolt 5. These are automatically detected by Harbor, which manages the inter-device connection and the now larger pool of unified memory.
End users shouldn’t expect to see the app anytime soon. Harbor is currently in “private preview” under Apple’s Claris brand.
Massive aid for the development of AI
As with almost all new technology trends, getting started with AI can be difficult. Especially if you don’t want to use cloud-based services.
Not everyone wants to simply hand over their data to OpenAI or Anthropic in exchange for their services. Creating a local LLM server eliminates this, but at a fairly high financial cost in new hardware.
But at the same time, there is also work to be done to get the server up and running. If you have multiple Macs, you can group them together to handle heavier workloads.
Apple has already provided software support, with MLX using Thunderbolt networking to group devices together for this purpose. Later, Apple made changes to macOS and MLX to include support for Thunderbolt 5 as well as introducing GPU-based Neural Accelerator support.
In October 2026, he also detailed a new Minions system that uses idle GPUs in Apple Silicon Macs and turns them into a dynamic virtual GPU cloud.
Harbor appears to be an attempt focused on the process of getting things up and running, rather than improving AI performance. While this doesn’t make queries any faster, it’s a change that should make the whole system much easier to get started in the first place.
Gn tech
