At the WWDC26 conference, Apple introduced new capabilities for creating and running Agentic AI directly on Mac devices using the MLX framework. This allows developers to build autonomous systems capable of planning and tool use, operating entirely on Apple Silicon.

image

What Happened

Apple presented capabilities for implementing reasoning loops locally on Mac via the MLX framework. Models can independently plan actions, use available tools, and analyze results. To work with large models, such as DeepSeek, support for distributed inference has been implemented via Thunderbolt or Ethernet using RDMA technology, providing up to a 3x speedup on clusters of multiple Macs.

Context

Traditionally, Agentic AI requires powerful cloud infrastructures, which creates dependency on providers and raises privacy concerns. MLX technology, combined with Apple Silicon architecture, allows these computations to be moved to personal devices, turning them into efficient nodes for working with LLMs.

Why It Matters for the Industry

For the industry, this means reduced dependency on cloud APIs and the ability to create more private and faster corporate automation systems. Support for distributed inference allows local computations to scale, making Macs competitive nodes for working with heavy models within Edge AI clustering.

Why It Matters for Users

Developers gain the ability to run powerful AI agents without subscriptions or a constant internet connection. Integration with Xcode and support for tools like OpenCode allow for using local AI to write and debug code in real-time, using only the resources of their own hardware.

Sources

Author

Look at AI, Editorial Team