Apple ships Core AI: a unified on-device framework for model execution

Apple has published documentation for Core AI, a new framework positioned as the foundation layer for running machine learning and generative models across its platforms. The framework appears to consolidate what was previously fragmented across Core ML, the Neural Engine APIs, and the various Apple Intelligence entry points into a single developer surface for model loading, inference, and hardware acceleration.

The move signals Apple’s intent to make on-device AI a first-class platform primitive rather than a bolt-on. By centralizing the runtime, Apple can route workloads across CPU, GPU, and Neural Engine without forcing developers to target each backend individually, and it gives Apple a chokepoint to enforce its privacy posture — keeping inference local rather than shipping prompts to third-party clouds.

For developers, the practical questions are which model formats Core AI accepts, how it interoperates with existing Core ML pipelines, and whether it exposes the lower-level controls (quantization, KV cache management, streaming) that production generative workloads require. The documentation landing page is sparse so far, suggesting this is an early reveal ahead of a fuller WWDC-style rollout.