Apple is reportedly testing Harbor to run local AI across linked Macs
The private-preview Claris app reportedly combines model downloads with automatic Thunderbolt clustering. Public availability, pricing and performance remain unsettled.
Loading page…
The private-preview Claris app reportedly combines model downloads with automatic Thunderbolt clustering. Public availability, pricing and performance remain unsettled.
Listen to this story
A private-preview app called Harbor could give Mac users a single interface for managing local open-weight models and combining multiple Apple silicon Macs into one inference resource. The reported business-facing app would use Thunderbolt 5 to pool unified memory, potentially letting a cluster run models too large for one Mac; no measured speed gains have been disclosed. Its proposed convenience bundles model management with Mac discovery and clustering, areas already served separately by tools such as LM Studio and Exo. Apple has not announced a launch, pricing, or final availability.
9to5Mac reported Harbor on November 10, citing leaker Aaron Perris; MacRumors says an early-access App Store listing was removed.
Harbor reportedly detects Thunderbolt 5-connected Macs and can browse, download, and load open-weight models without command-line setup.
The promised pooled memory and higher request capacity are product claims, not published performance benchmarks.
Apple is reportedly testing an app that brings running AI models and linking multiple Macs into one interface. On November 10, 9to5Mac reported that leaker Aaron Perris had discovered Harbor, a private-preview app under Apple’s Claris brand. Its reported promise: local AI without command-line setup, including models too large for a single computer.
Perris described Harbor as a private AI inference server: software that runs trained models on Apple silicon Macs. His November 10 post says it runs open-weight models locally, with data staying on the hardware rather than reaching the cloud. That privacy description is part of the disclosed product claims, not an independently verified finding.
Harbor is described as working on a single Apple silicon Mac, but its more distinctive feature is clustering: connecting multiple computers so they act as one inference resource. The disclosed design uses Thunderbolt 5 connections to pool the Macs’ unified memory. That combined capacity is intended to accommodate a model that exceeds any one computer’s memory.
The feature description also promises more simultaneous requests than one Mac can handle alone. Neither description supplies a measured speed gain, so the claim should not be read as a performance benchmark.
The reported workflow puts three jobs inside the app rather than asking users to assemble them separately:
Harbor would not introduce either local model running or Mac clustering as a new category. MacRumors points to existing free alternatives: LM Studio for running a local AI server on one Mac, and Exo, an open-source tool for clustering Macs. The reported distinction is putting a simple model interface and multi-computer clustering together.
There is also operating-system groundwork beneath the proposed interface. macOS Tahoe 26.2 added RDMA over Thunderbolt 5, a capability MacRumors says makes clustering faster.
The Claris branding offers a clue about the intended customer, but not a settled answer. Apple’s subsidiary is known for FileMaker, its business-focused database platform. MacRumors also observed that Harbor’s sign-up page asks for a company name. That suggests a business audience, rather than establishing who will ultimately be able to use it.
Pricing and marketing remain unclear, including whether Harbor will be free. 9to5Mac’s suggestion that a launch could arrive in the coming weeks or months is an expectation, not an Apple commitment.
Loading discussion...
Join the conversation
Explain which setup hassles would make convenience worth paying for.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.