Apple’s Harbor App Turns a Mac Into a Private AI Server

The Harbor app from Apple subsidiary Claris turns a Mac into a private AI inference server that runs open-weight models on Apple silicon. Two or more Macs with Thunderbolt 5 can pool their unified memory, so a cluster can load models that are too big for a single Mac. Aaron Perris posted Harbor’s App Store description on X, and Apple has since taken the listing down. The app is in private preview under Claris, the brand that also makes FileMaker, and it has no announced price or release date.

The Harbor App Pools Mac Memory

Harbor detects your Macs over Thunderbolt 5 and forms the cluster on its own, with no manual setup and no command line. You connect two or more Thunderbolt 5 Macs, and Harbor pools their unified memory into one inference resource. You can then run models larger than any single Mac can hold, or serve more concurrent requests across the cluster. The amount of unified memory in your Mac sets how large a model you can load, and even the Mac Studio with M5 Ultra tops out at 512GB. Harbor’s cluster view shows a MacBook Pro with M5 Max linked to two M4 Pro Mac minis, with the two Mac minis linked to each other. Clustering Macs already works without Harbor, because the Mac Studio supports Thunderbolt 5 and RDMA clustering. The manual route means cabling the Macs in a mesh, where every Mac connects to every other, or in a ring, where each Mac connects to two neighbors. After the cabling, you turn on RDMA in System Settings, restart, and generate a configuration with the mlx.distributed_config command-line tool. Harbor aims to replace those steps with plugging the Macs in and pressing start.

Apple is working on a new app called Harbor Harbor turns your Mac into a private AI inference server. Run open-weight models locally on Apple silicon—your data never leaves your hardware, and nothing reaches the cloud. Connect multiple Mac computers over Thunderbolt 5 to pool unified memory, enabling large models and higher concurrency that a single computer can handle alone. Set up in minutes, not hours Harbor automatically detects your devices over Thunderbolt 5 to form a cluster. Plug in, start, and you’re running—no manual configuration, 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 the work, not the setup. Cluster for scale Connect two or more Thunderbolt 5-capable Mac computers and Harbor pools their unified memory into a single inference resource. Run models that exceed any single computer’s capacity, or serve more concurrent requests across your cluster. Harbor is currently in a private preview under the Claris brand.

— Aaron (@aaronp613) November 10, 2026

Harbor Runs Models Locally

Harbor runs open-weight models on Apple silicon, so your data stays on your Mac and nothing goes to the cloud. You can find and download models inside the app, and Harbor handles loading and managing them. Harbor’s server panel has a copy button for an API endpoint and a Switch Model button for the loaded model. Its hardware section shows Thunderbolt and RDMA settings. Harbor joins local AI options that already run on the Mac. Perplexity Computer now runs fully on the device for Pro, Max and Enterprise subscribers, and Meta says its Muse Spark Open model runs locally on two M5 Ultra Mac Studios.