rudrankriyam/core-ai-framework-lab

This repository is an early exploration of a new Apple framework called CoreAI, which Apple introduced in the Xcode 27 beta. Xcode is Apple's tool for building apps for iPhone, Mac, and other Apple devices. A framework is a library of pre-built code that developers can use. CoreAI is Apple's lower-level system for loading, specializing, and running AI models directly on a device's hardware, including the Neural Engine chip found in Apple Silicon processors. The project is not a finished app. It is a lab that maps out what the CoreAI framework actually looks like at this early stage, because Apple's beta documentation is still minimal. The author compiled the framework, inspected its public API, and wrote small code examples demonstrating each area: loading a model from a file, checking if it is valid, choosing which hardware to run it on (CPU, GPU, or Neural Engine), loading specific functions from the model, and running inference. These examples are organized as a small SwiftUI app so you can browse them in Xcode. An important note: this is not the same as Apple's FoundationModels framework, which is the high-level API for using Apple's on-device language model for things like text generation. CoreAI sits one level lower and is about loading arbitrary trained model files and running them efficiently on device, rather than accessing a specific built-in model. The project currently has no bundled AI model to actually run, because compatible CoreAI model packages are not yet publicly available in this beta period. The examples show the code structure and API calls that would be needed, but a complete inference demo requires a model file that does not yet exist in a compatible format. The API itself may also change between Xcode beta releases. The project requires Xcode 27 beta and targets iOS 27 or macOS 27. It is open source under the MIT license and welcomes contributions from other developers who are also exploring this framework early.