Apple Harbor, a new app from Apple subsidiary Claris, turns Mac computers into private AI servers that run open-weight models without cloud fees. The app is in a private preview, and Apple has already pulled its unlisted App Store entry.
The pitch targets AI users who pay per-token fees to cloud providers. However, free tools already cluster Macs, so Harbor mainly shifts the cost of AI from monthly bills to Apple hardware.
Sponsored
Sponsored
Why Apple Harbor Moves AI Costs From the Cloud to the Desk
The app, previously called Fabric, surfaced on X, where a user shared its description and later reported the takedown.
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
— Aaron (@aaronp613) November 10, 2026
Claris, best known for its FileMaker business database, says a single Mac can serve as an AI endpoint within minutes. The team also promises no per-token billing.
Data and prompts stay strictly within your physical Mac hardware.
Claris
For larger models, the Apple Harbor app links two or more Thunderbolt 5 Macs and pools their unified memory. As a result, models too big for one machine can run across the cluster.
Promotional images show a MacBook Pro with an M5 Max chip linked to two Mac minis. Another shows the app downloading OpenAIs open-weight gpt-oss-20b model from Hugging Face, a public model library.
Apple Harbor Cluster Preview. Source: X
Sponsored
Sponsored
That shifts spending toward devices. Ethereum co-founder Vitalik Buterin has noted that the cost of AI inference sits almost entirely in hardware.
For Apple shareholders, this is a hardware story. After Apple beat earnings expectations on record iPhone sales, Harbor gives the company a new pitch for high-memory Macs.
Apple (AAPL) Stock Price. Source: TradingViewCan a One-Click Setup Beat Free Open-Source Rivals？
Harbor enters a crowded field. Developers already cluster Macs with Exo and run models on one Mac with LM Studio, and neither costs a cent.
Apple Harbor targets setup rather than speed, bundling both jobs into one app. It detects connected Macs on its own and handles model downloads, with no command line needed.
Claris has not named a price, but the sign-up form asks for a company name, which points to business customers first.
Running models locally could also ease the chatbot lock-in worries that Shark Tank investor Mark Cuban raised, since user data stays off third-party servers.
Meanwhile, open-weight models keep improving. Bloomberg Intelligence estimates Chinese models have narrowed the US AI lead to about 3%.
If that trend holds, more AI users could run strong models at home, and Apple Harbor would put Macs at the center. A steep price or a business-only rollout would weaken that case.