Claris previews Harbor for private AI across Apple silicon Macs

Claris describes Harbor as early access to a private AI server from Claris, which it identifies as an Apple company.[2] The company says Harbor turns Apple silicon computers into a private AI server and runs open-weight models.[2] Claris also says the service has zero cloud fees and no per-token billing.[2] For teams considering local model use, those details frame Harbor as an option to evaluate on hardware they control, rather than a cloud-billed service; the description does not establish what models or workloads will perform well.

Claris says users can set up an endpoint in minutes on a single Mac.[2] For larger setups, its description says users can connect multiple machines over Thunderbolt 5 and pool unified memory on hardware they control.[2] Aaron Perris’s description adds that Harbor automatically detects devices connected over Thunderbolt 5 to form a cluster, and that users can browse for, download and load open-weight models within the app.[1] Perris also says Harbor can run models that exceed the capacity of any single computer and that a cluster can serve more concurrent requests across its Macs.[1]

Claris labels Harbor a private preview, while 9to5Mac reported that leaker Aaron Perris had found evidence of an Apple Mac app in development.[2][1] Perris’s post and 9to5Mac’s report were both dated November 10, 2026.[1] Perris’s description says getting started requires no manual configuration, command line or infrastructure expertise.[1] MacRumors says Apple added RDMA over Thunderbolt 5 with macOS Tahoe 26.2, making clustering faster.[3] Together, the descriptions point to a local setup designed to be simpler to assemble, but they do not establish general availability.