Running a big AI model on your own Mac usually means a terminal window, a pile of configuration files, and a lot of patience. Apple appears to be building an app to make all of that go away.
Leaker Aaron Perris (@aaronp613 on X) found a Mac app called Harbor that, in Apple’s own description, “turns your Mac into a private AI inference server.” 9to5Mac and AppleInsider both reported the find today. To be clear up front: Harbor is not a launched product. It’s in a private preview under Claris, the Apple-owned software company best known for FileMaker, and Apple hasn’t announced it.
What Harbor does
According to the description Perris posted, Harbor lets you “run open-weight models locally on Apple silicon,” and it makes a privacy promise: “your data never leaves your hardware, and nothing reaches the cloud.”
The listing highlights three things:
- Setup with no command line. “Harbor automatically detects your devices over Thunderbolt 5 to form a cluster,” the description says, with “no manual configuration, no command line, no infrastructure expertise required.”
- Models from inside the app. You can “browse, download, and load open-weight models from within the app,” and Harbor handles model management for you.
- Clustering. Connect “two or more Thunderbolt 5-capable Mac computers” and Harbor “pools their unified memory into a single inference resource.” The pitch is that you can run models too big for any one Mac, or serve more requests at the same time.
AppleInsider adds one more detail: the app was formerly known as Fabric. The description doesn’t say which models Harbor will support, whether it will cost anything, or when it will ship. 9to5Mac thinks a public launch could come “in the weeks and months ahead.” AppleInsider is more cautious and says end users “shouldn’t expect to see the app anytime soon.”
Why Apple would build this
Macs have quietly become popular machines for running AI models locally. 9to5Mac notes that the Mac has been getting more popular this year, partly as a tool for running models on the Mac Studio and Mac mini. The reason is unified memory: on Apple silicon, the CPU and GPU share one pool of memory, so a Mac with a lot of RAM can hold a model that wouldn’t fit on a typical graphics card.
Apple has also already built the plumbing for linking Macs together. Starting with macOS 26.2, Macs with Apple silicon and Thunderbolt 5 support RDMA over Thunderbolt, according to Apple’s developer documentation. It lets connected Macs pass data directly between their memory with very low latency. Apple’s open-source MLX framework uses it for distributed inference, and Apple’s machine learning research team has shown large models running split across two Mac Studios.
The catch today is that setting it up is very much a developer job. You enable RDMA from macOS Recovery, write host files, and launch jobs from the terminal. Harbor’s description reads like a direct answer to that. Nothing in the leak says what Harbor runs on under the hood, but the hardware requirement lines up exactly with the Thunderbolt 5 clustering Apple already ships.
What it means for you
Who it’s for. At least at first, Harbor looks aimed at people and small teams who want a shared AI model on hardware they own: developers, researchers, and small businesses that can’t send sensitive documents to a cloud chatbot. Putting it under Claris, a brand that sells to businesses, points the same way.
The privacy pitch. Your prompts and files stay on your Macs and never go to OpenAI, Google, or even Apple’s servers. That’s the whole point, and it’s the main reason to choose this over ChatGPT or Claude.
The hardware. Clustering requires Thunderbolt 5, so older Macs need not apply. Big models also need a lot of unified memory. If you already own a recent high-memory Mac Studio or Mac mini, Harbor could make it far more useful. If you don’t, this is an expensive hobby.
The competition. Apps like LM Studio and Ollama already run local models on Macs. What Apple could offer is the part they don’t make easy: plugging several Macs together and having it just work.
For now, Harbor is a preview most people can’t get. But it’s a clear sign that Apple sees the Mac as more than a client for cloud AI. It wants the Mac to be the server too.
Sources: 9to5Mac (Nov. 10), AppleInsider (Nov. 10), Aaron Perris (@aaronp613) on X, Apple Developer TN3205, and Apple Machine Learning Research.
Reporting reference: 9to5Mac ↗