Apple is developing a new app called Harbor (formerly known internally as Fabric) designed to transform one or more Macs into a private AI inference server. This tool is intended to make it much easier to run large language models (LLMs) locally, without needing the power of cloud-based services. Harbor prioritizes keeping all data on Apple devices, offering a more secure path for AI processing on-device.
What is Harbor and how will it work?
Harbor aims to simplify the setup of AI inference servers for users who have compatible hardware. The app is focused on open-weight models and manages the entire workflow: downloading models, serving them locally, and scaling across machines. The UI is meant to be intuitive, avoiding the complexities of terminal commands or manual infrastructure configuration.
When multiple Macs are available, Harbor leverages Thunderbolt 5 to connect them into a cluster. It auto-detects Macs over Thunderbolt 5, aggregates their unified memory, and treats them like one larger inference resource. This allows LLMs that demand more capacity than a single Mac can offer to run effectively, and more concurrent tasks to be handled.
Status and background
At present, Harbor is in a private preview under Apple’s Claris brand — not yet available publicly. It follows Apple’s prior work in clustering Macs for AI workloads, including support in macOS and the MLX framework for Thunderbolt networking and neural accelerators. A related technology, Minions, was introduced in October 2026 to utilize idle GPUs in Mac machines to create virtual GPU clouds.
Harbor is thus positioned as less about boosting raw AI performance and more about dramatically lowering the barrier to running your own local LLMs. While performance at scale remains limited by the hardware, the ease of deployment is expected to be a significant shift.
For users concerned about privacy or cloud dependency, Harbor promises a compelling solution: everything happens on device or within your cluster, with zero reliance on remote servers. For those with multiple Macs, Harbor could turn what was once a tricky DIY project into a setup that’s plug-and-play.
Though Harbor is still under wraps, its development signals Apple’s increasing investment in local AI. Between Thunderbolt 5 clusters, MLX enhancements, and the Minions project, Harbor represents a broader push to make Macs legitimate AI workhorses without sending data off-premises.
What to watch: Whether Harbor will support external AI model marketplaces, how it handles security of model updates, and how well it scales across heterogeneous Mac hardware. Also key will be the timeline for release: Claris-private preview now, but no word yet on public availability. This could redefine the local AI landscape—if done right.
