Apple Harbor App Leak: Secret Mac Tool Would Pool Multiple Macs Into a Private AI Server, No Cloud Needed

CUPERTINO, Calif. — Apple is testing a new Mac app called Harbor that would turn Apple computers into private artificial intelligence servers. The app could also link several Macs together to run AI models too large for a single machine, according to reports this week.

The app was first revealed by leaker Aaron Perris, who posted details of its description on X. It was later reported by 9to5Mac, MacRumors and AppleInsider. Harbor is being developed by Claris, an Apple subsidiary best known for its FileMaker database software, and is currently in private preview.

"Harbor turns your Mac into a private AI inference server," the app's description reads. It continues: "Run open-weight models locally on Apple silicon—your data never leaves your hardware, and nothing reaches the cloud."

Apple has not publicly announced Harbor. According to MacRumors, an App Store listing for the app briefly appeared before it was removed. AppleInsider reported that the app was previously known as Fabric.

Clustering Macs for bigger models

Harbor's most notable feature is its ability to combine the power of multiple Macs.

According to the app's description, users can connect two or more Macs over Thunderbolt 5, Apple's fastest wired connection standard. Harbor pools their unified memory into a single resource for running AI models. That would allow users to run larger models and handle more requests at once than a single computer could manage alone.

"Harbor automatically detects your devices over Thunderbolt 5 to form a cluster," the app's description says.

Memory is one of the biggest limits on running large AI models locally. By pooling memory across machines, Harbor could let users run models that would otherwise require expensive server hardware. Two high-end Mac Studios, each configured with 512 gigabytes of memory, would together offer roughly a terabyte.

Apple laid some of the groundwork for clustering in macOS Tahoe 26.2. That update added support for a faster data-transfer technology, known as RDMA, over Thunderbolt 5.

No command line required

Apple is pitching Harbor as easy to use. The app's description promises "no manual configuration, no command line, no infrastructure expertise required." It says users can set up in minutes rather than hours.

Users would be able to browse, download and load open-weight AI models directly within the app. Open-weight models are AI systems whose underlying parameters are publicly released, allowing anyone to run them on their own hardware.

Running AI locally appeals to people and businesses that do not want to send sensitive data to cloud services, as well as to developers who want to avoid paying for cloud computing.

Competing with existing tools

Harbor would enter a market already served by free tools. LM Studio and Ollama let Mac users run AI models on a single machine. Exo, an open-source tool, can already cluster multiple Macs.

MacRumors said Harbor appears to combine LM Studio's simple interface with clustering capabilities. Reports say pooling memory across multiple Macs is something Ollama and LM Studio do not currently offer.

Analysts said routing the app through Claris rather than Apple's main Apple Intelligence brand suggests Apple may be targeting professionals and developers first. It also gives Apple a hardware selling point: customers who want to run larger models would need to buy more Macs.

Many people already use Mac Studio and Mac mini computers to run AI models, thanks to Apple silicon's unified memory design. Harbor would make that use case far more accessible.

There is no announced release date or pricing for Harbor. It is unclear which AI models the app will support at launch, whether Apple will charge for it, or whether it will ever be offered to ordinary Mac users beyond the Claris preview.

Apple did not publicly comment on the reports.