Apple just published documentation for Core AI, a new native framework that gives iOS, iPadOS, and macOS developers a direct path to on-device AI features without relying on external cloud APIs. The framework is now live on Apple's developer portal.
The framework gives app developers structured access to Apple's Neural Engine - the dedicated AI chip present in every iPhone since the A12 Bionic in 2018 and in every Apple Silicon Mac. On-device inference means generating AI outputs locally on the device hardware rather than sending a request to a remote server and waiting for a response. It works offline, has no latency from a round-trip to an external service, and keeps user data on the device.
What This Means for App Developers
Until now, building AI features into an Apple app meant choosing between calling a third-party cloud API like OpenAI or Anthropic, wrestling with Apple's older Core ML framework to deploy custom models, or using the MLX library that Apple's research team released without integrating it into standard development workflows. None of those were clean options for shipping production apps.
Core AI changes that. A unified, officially supported framework means stable APIs with proper versioning and documentation - not piecing together unofficial approaches that could break with an OS update. Apple can also enforce on-device privacy guarantees in a way that's impossible with cloud APIs, which is a credible selling point for apps in health, finance, and personal productivity where users increasingly question what happens to their data.
What the Docs Don't Yet Answer
The documentation doesn't currently specify which foundation models Apple bundles with Core AI, whether developers can bring their own fine-tuned models (models further trained on custom data), or how Core AI integrates with Apple Intelligence features. Those details determine whether this is genuinely useful for the things developers actually want to build, or primarily a wrapper around Apple's own narrowly scoped capabilities.
Apple's hardware is legitimately strong - Neural Engine chips in current devices outperform dedicated AI accelerators from a few years ago. The bottleneck has always been software access. Core AI looks like Apple's serious attempt to open that up to third-party developers. The documentation suggests this is still early-stage, but the framework showing up officially in Apple's developer portal is a meaningful signal about where Apple plans to take on-device AI for apps.