Apple subsidiary Claris is testing a new app called Harbor that turns Mac computers into private AI servers that run open-weight models without cloud fees.
The application surfaced briefly on the App Store before Apple pulled its unlisted entry. The app was previously called Fabric before surfacing on X.
The app turns your Mac into a private AI inference server by letting users run open-weight models locally on Apple silicon. Data and prompts stay strictly within the physical hardware, and nothing reaches the cloud.
Harbor connects two or more Macs with Thunderbolt 5 and pools their unified memory into a single inference resource for larger models that exceed a single computer’s capacity. Promotional images show a MacBook Pro with an M5 Max chip linked to two Mac minis.
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… pic.twitter.com/UYJBQeX4Nf
The clustering relies on macOS Tahoe 26.2, which added Remote Direct Memory Access over Thunderbolt. RDMA allows one Mac to read another’s memory directly with very low latency. Configurations of this nature are supported for Thunderbolt 5-equipped devices like the M4 Pro Mac mini, M4 Max Mac Studio, M4 Max MacBook Pro, and M3 Ultra Mac Studio, according to the open-source exo project.
Harbor targets setup rather than raw speed, bundling device detection and model downloading into a single interface. According to the app description shared on X, Harbor automatically detects your devices over Thunderbolt 5 to form a cluster.
Claris claims that a single Mac is capable of operating as an AI endpoint in just a few minutes. The company is behind the FileMaker business database.
Harbor replaces recurring monthly AI subscription fees with the upfront cost of Apple equipment. Existing free utilities already enable Mac clustering.
Claris recommends a 32GB memory floor for those deploying large models, with 64GB to 96GB or higher advised.
Moving spending from monthly cloud subscriptions to local hardware involves high device costs. Apple explicitly marketed its new chips for running large AI models directly on the device, with pricing listed as follows: a Mac mini with an M6 chip and up to 32 GB of memory starts from $899, a Mac mini with an M5 Pro chip and up to 64 GB starts from $1,699, and a Mac Studio with an M5 Ultra chip and up to 512 GB starts from $5,499.
Claris, an Apple subsidiary, is testing Harbor, an app that turns Apple silicon Macs into private AI servers running open weight models locally, keeping data off the cloud. It works on one Mac or clusters several over Thunderbolt 5, pooling unified memory for larger models and… pic.twitter.com/UTV2kylnmW
Macs have become a preferred choice for local AI tasks because language models require their entire data set to be stored in memory during execution, and Apple’s chips use a unified memory architecture for both the processor and graphics unit.
Bloomberg Intelligence estimates Chinese models have narrowed the US AI lead to about 3%. Apple Harbor could position Macs as the primary hub for home-based users running powerful AI models if that trajectory continues.
More on Claris
Apple's Claris is testing Harbor, an app that turns a Mac into a private AI server and pools memory across Macs over Thunderbolt 5. Private preview only: no price, no date, no independent tests yet.https://t.co/ANTE9UUHXU#AI #Apple #LocalAI
Apple just surfaced a new Mac app called Harbor today. Here's what you need to know. Harbor turns a Mac into a private AI inference server. It runs open-weight models locally on Apple silicon. Data never leaves your hardware and nothing reaches the cloud. It can also cluster… https://t.co/Yfen9XizXK pic.twitter.com/pxTGJz0W3s
