Apple develops Harbor app to turn Mac computers into private AI servers
November 13, 2026 · 2 min read
Also published in Português (Brasil)
Apple is completing a new application called Harbor that would allow Mac computers to run large artificial intelligence models locally without sending user data to cloud services, according to a leak from a person known as Aaron who reports on Apple products. The technology enables multiple Mac computers to be connected together to create a powerful AI server using their combined memory and processing power.
Harbor is currently in closed testing with Claris, a company Apple acquired in 1987. The application would automatically detect Mac computers connected via Thunderbolt 5 and combine their memory resources. Two Mac computers with 64 gigabytes of unified memory each could theoretically provide 128 gigabytes of combined memory for running AI models, along with combined processing power from their neural processing units.
The application would include a built-in interface for finding, downloading, and running open artificial intelligence models directly on the user's hardware. This approach differs from services like ChatGPT and Claude, where the AI model runs on the provider's servers. Harbor would keep all data processing local to the user's machines.
Apple's approach addresses a limitation of individual Mac computers: 64 gigabytes of memory, or even double that amount, falls short of what is needed to run large language models comparable to cloud-based versions. By connecting multiple machines with high memory bandwidth through Thunderbolt 5, users could access significantly more processing capacity without purchasing expensive specialized hardware.
The application requires no command line interface, manual configuration, or extensive technical expertise to set up connected machines. This positions Harbor as a competitor to existing tools like LM Studio and Ollama, which are popular applications for running AI models locally. It remains unclear which AI models Harbor will support, whether Apple plans to charge for the application, or whether hybrid solutions that partially use cloud services will be available.