Apple's Claris Is Testing Harbor, a Local AI App That Links Macs Over Thunderbolt 5
Claris, an Apple subsidiary, is developing Harbor, an app for running open-weight AI models locally on Apple silicon Macs.
2 sources 19h ago
Reporting from 2 sources: ASCII.jp, GIGAZINE.
Why it matters
Harbor and Minions both treat a pile of Macs as shared compute, which puts Apple's local AI pitch on hardware people already own rather than on cloud capacity.
Key Facts
- Claris, an Apple subsidiary, is developing Harbor, an app for running open-weight AI models locally on Apple silicon Macs.
- Claris lists Harbor as able to connect multiple Macs via Thunderbolt 5 to share memory.
- Harbor is in private preview, and the timing of its official release is unknown.
- Screenshots posted by Aaron Perris show Harbor running Llama 3.6-2-Instruct-4bit on a MacBook Pro with an M5 chip.
- Apple is also developing Minions, an application that connects multiple Macs to build a virtual GPU cloud.
A Claris page lists setup in minutes, memory sharing across multiple Macs connected by Thunderbolt 5, and data stored on the Mac rather than exposed externally. Harbor is in private preview, and the release timing is unknown. Apple is also developing Minions, which links Macs into a virtual GPU cloud.
Screenshots posted by Aaron Perris show the app downloading an MLX-compatible build of gpt-oss-20b, running Llama 3.6-2-Instruct-4bit on a MacBook Pro with an M5 chip, and managing three Macs joined over Thunderbolt 5 as one cluster. Claris lists the app as private preview, with no release date given.
The memory sharing is the part worth watching. A single Mac's unified memory caps the model size it can hold, and pooling several machines over a cable raises that ceiling without a datacenter. Harbor is not the only project on that idea: Apple is also developing Minions, which gathers unused Macs into a virtual GPU cloud.
Written with AI by Yomimono from the 2 cited sources below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google