Claris Harbor Turns Macs Into a Private AI Server--Apple's Software Subsidiary Has a Private Preview of a Mac App That Runs Open-Weight Models With No Cloud Fees and Pools Unified Memory Over Thunderbolt 5, Its App Store Listing Has Been Pulled, and the Sign-Up Form Asks for a Company

A man with short brown hair, a beard and tortoiseshell glasses, in a faded denim jacket, sits in a tan leather chair at a light wooden desk in a sunlit room, one hand on a white keyboard, looking at an Apple Studio Display that shows a magazine spread in a page-layout app; a silver Mac Studio stands beside the display, a potted succulent and printed magazine pages lie on the desk, and rooftops show through the windows behind him.
Photo: Claris · A man with short brown hair, a beard and tortoiseshell glasses, in a faded denim jacket, sits in a tan leather chair at a light wooden desk in a sunlit room, one hand on a white keyboard, looking at an Apple Studio Display that shows a magazine spread in a page-layout app; a silver Mac Studio stands beside the display, a potted succulent and printed magazine pages lie on the desk, and rooftops show through the windows behind him.

Claris, the Apple subsidiary that sells FileMaker, has a page up for Claris Harbor, "a new private AI server from Claris, an Apple company," marked Private Preview. Claris's Harbor page says the app "turns the Apple silicon computers you own into a private AI server," runs open-weight models "with zero cloud fees, no per-token billing," and, when you want larger models, connects "multiple machines over Thunderbolt 5 to pool unified memory on hardware you control." MacRumors and 9to5Mac reported it on November 10; MacRumors says there was an App Store listing, since removed, and 9to5Mac credits Aaron Perris with finding it. It is an app from inside Apple for the Mac clusters Apple has advertised since August, which Apple's own pages so far leave to the command line and a third-party tool.

What Claris says Harbor does

One Mac is enough: "Set up an endpoint in minutes on a single Mac." Two tiles promise "Strictly private," with "Data and prompts stay strictly within your physical Mac hardware," and "Thunderbolt clustering," with "Pool unified memory across machines for demanding model sizes." The listing copy MacRumors quotes goes further: Harbor "automatically detects your devices over Thunderbolt 5 to form a cluster," with "no manual configuration, no command line, no infrastructure expertise required," lets you "Browse, download, and load open-weight models from within the app," and with "two or more Thunderbolt 5-capable Mac computers" it "pools their unified memory into a single inference resource." The page names no macOS version, no Mac model, no price and no date, and has no download. Its form asks for a name, an email address, a company, a phone number and a country; MacRumors reads the company field as "more of a business focus." We have not been invited to the preview and have not run it.

The cluster Apple built the Mac Studio for

Harbor sits on something Apple shipped first. Apple's August 25 Mac Studio press release says teams can join several Mac Studios into a cluster "using the built-in support for Thunderbolt 5 and RDMA (remote direct memory access)" and that "A cluster of four Mac Studio systems delivers up to a remarkable 3x faster AI inference than a single system"; the M5 Ultra model starts at $5,499 and takes up to 512GB of unified memory. Apple's WWDC26 session on distributed inference with MLX names the version, "Starting in macOS 26.2, Remote Direct Memory Access protocol, shortly RDMA, is supported over Thunderbolt 5," and the switch in System Settings: search for RDMA, "click on 'Enable RDMA over Thunderbolt', enable RDMA, and reboot." Apple's October 15 research post ran Muse Spark Open on four Mac Studios at "over 36 tokens per second." The cluster needs Thunderbolt 5 on every Mac: Apple's Mac mini press release gives the $1,699 M5 Pro model three Thunderbolt 5 ports and the clustering line, and the $899 M6 model Thunderbolt 4, so the $899 Mac mini could run Harbor on its own but not join a cluster.

The two tools Apple's own page recommends

Apple's Mac Studio page tells buyers to "Run LLMs locally using LM Studio Bionic" and to "use exo to cluster multiple AI workstations." exo on GitHub is Apache-licensed, needs "a Mac with Apple silicon and macOS Sequoia 15.1 or later," and says "RDMA over Thunderbolt needs macOS Tahoe 26.2 or later"; its caveats are the ones "no manual configuration" is aimed at. Every Mac in an RDMA cluster "must be connected to all other devices in the cluster," "On a Mac Studio, you cannot use the Thunderbolt 5 port next to the Ethernet port," and the macOS versions must "match exactly (even beta version numbers)." MacRumors' reading is that Harbor "appears to combine LM Studio's simple interface with clustering." It follows the Minions paper of October, in which Apple's researchers harvest the idle GPUs of Macs already on a network.

Then and now

Claris "began as an Apple subsidiary" in 1986, by Claris's August 6, 2019 announcement of its return to the name, which also records that "In 1998, the company was renamed as FileMaker, Inc., focusing on its flagship product." On January 28, 1998, our note on Apple's press release said Claris would become FileMaker, Inc., "focusing only on its highly successful and cross-platform FileMaker Pro database software," with ClarisWorks and the rest "re-absorbed into Apple." And on January 8, 2004 our Xgrid note quoted Apple, "Xgrid turns a group of Macs into a supercomputer, so they can work on problems greater than each individually could solve." Apple put RDMA in macOS, its researchers wrote the cluster paper, and its product page sends buyers to a third-party tool, so we would rather the app came from Apple itself, in macOS and free, than from a subsidiary whose sign-up form wants a company name.

Sources

  1. Claris, “Claris Harbor | Claris, an Apple company” accessed Nov 11, 2026The page's badge reads "Private Preview"; the heading "Run AI models locally on your Mac."; the subheading "Early access to a new private AI server from Claris, an Apple company."; "Claris Harbor turns the Apple silicon computers you own into a private AI server."; "Run open-weight models with zero cloud fees, no per-token billing, and the assurance that your data never leaves your environment."; "Set up an endpoint in minutes on a single Mac."; "When you want to run larger models, connect multiple machines over Thunderbolt 5 to pool unified memory on hardware you control."; the tiles "Strictly private" ("Data and prompts stay strictly within your physical Mac hardware.") and "Thunderbolt clustering" ("Pool unified memory across machines for demanding model sizes."); the form fields First Name, Last Name, Email, Company, Phone, Country and Comments; no macOS version, Mac model, price, date or download on the page. The header picture claris-local_ai_server-header.jpg is the item's image. Accessed November 11, 2026.
  2. MacRumors, “Apple's Claris Testing 'Harbor' App for Local AI on Macs - MacRumors” accessed Nov 11, 2026Juli Clover, Tuesday November 10, 2026, 3:34 pm PST: "Apple's Claris subsidiary is testing an app called Harbor, which turns Apple silicon Macs into a private AI server."; the quoted listing copy "Harbor automatically detects your devices over Thunderbolt 5 to form a cluster.", "no manual configuration, no command line, no infrastructure expertise required.", "Browse, download, and load open-weight models from within the app.", "Connect two or more Thunderbolt 5-capable Mac computers and Harbor pools their unified memory into a single inference resource."; "Harbor is in early access and is not publicly available."; "There was an App Store listing, but it has now been removed."; "The sign-up page for Harbor asks for a company name, which suggests more of a business focus."; "Apple's Harbor app appears to combine LM Studio's simple interface with clustering." Accessed November 11, 2026.
  3. 9to5Mac, “Apple has new Mac app coming to make running local AI models easy - 9to5Mac” accessed Nov 11, 2026Ryan Christoffel, November 10, 2026, 11:15 am PT: "Leaker Aaron Perris has discovered evidence of a new Mac app in the works at Apple called Harbor, which makes it easy to run local AI models across one or more Macs."; "Harbor is currently in a private preview under Apple's Claris brand". Accessed November 11, 2026.
  4. Apple Newsroom, “Apple introduces new Mac Studio with M5 Max and M5 Ultra - Apple” accessed Nov 11, 2026Press release of August 25, 2026: users and teams who want to pool AI compute can join several Mac Studio systems into a cluster "using the built-in support for Thunderbolt 5 and RDMA (remote direct memory access)"; "A cluster of four Mac Studio systems delivers up to a remarkable 3x faster AI inference than a single system."; "up to 512GB of unified memory" on M5 Ultra; "Mac Studio with M5 Ultra starts at $5,499 (U.S.)". Accessed November 11, 2026.
  5. Apple Developer, “Explore distributed inference and training with MLX - WWDC26 - Videos - Apple Developer” accessed Nov 11, 2026Session 233 of WWDC26; the transcript: "Starting in macOS 26.2, Remote Direct Memory Access protocol, shortly RDMA, is supported over Thunderbolt 5."; "Open settings on the machine, search for \"RDMA\", click on \"Enable RDMA over Thunderbolt\", enable RDMA, and reboot." The item puts single quotation marks around the two settings names inside the quotation. Accessed November 11, 2026.
  6. Apple, “Local and Distributed Large LLM Inference with MLX and RDMA on Mac Studio - Apple Machine Learning Research” accessed Nov 11, 2026Dated October 15, 2026: "macOS 26.2 introduced a low-latency RDMA driver over Thunderbolt 5"; "On a cluster of four Mac Studios, Muse Spark Open generates over 36 tokens per second". Accessed November 11, 2026.
  7. Apple Newsroom, “Apple unveils a more powerful Mac mini featuring the all-new M6 and M5 Pro - Apple” accessed Nov 11, 2026Press release of August 25, 2026: "three Thunderbolt 4 ports on Mac mini with M6, and three Thunderbolt 5 ports on Mac mini with M5 Pro"; "Thunderbolt 5 also allows users to cluster multiple Mac mini systems together to run large AI models entirely on device."; "Mac mini with M6 starts at $899 (U.S.)"; "Mac mini with M5 Pro starts at $1,699 (U.S.)". Accessed November 11, 2026.
  8. Apple, “Mac Studio - Apple” accessed Nov 11, 2026The AI apps panel: "Run LLMs locally using LM Studio Bionic." and "And use exo to cluster multiple AI workstations." Accessed November 11, 2026.
  9. GitHub, “GitHub - exo-explore/exo: Run frontier AI locally.” accessed Nov 11, 2026The README: license Apache-2.0; "The app needs a Mac with Apple silicon and macOS Sequoia 15.1 or later. RDMA over Thunderbolt needs macOS Tahoe 26.2 or later."; the caveats "Devices that wish to be part of an RDMA cluster must be connected to all other devices in the cluster.", "On a Mac Studio, you cannot use the Thunderbolt 5 port next to the Ethernet port." and "Please ensure that OS versions match exactly (even beta version numbers) on all devices." Accessed November 11, 2026.
  10. Claris, “FileMaker reborn as Claris.” accessed Nov 11, 2026Press release dated August 6, 2019, Santa Clara: FileMaker, Inc. "unveiled the start of a new chapter in the company's history as Claris International Inc." at its 24th annual DevCon; "In 1986, Claris began as an Apple subsidiary."; "In 1998, the company was renamed as FileMaker, Inc., focusing on its flagship product."; "Claris International Inc., is an Apple subsidiary." Accessed November 11, 2026.
Saw something? Send a tip or a correction

The archive ran on reader tips. What did you see, where, and do you want the credit? Something wrong on this page? Say so and it gets fixed.

Read by the editor. You get a copy by email. Never published without your say.

More in Software · This month in the archive