CoreAIKit
A Swift SDK that downloads, caches, and runs Apple's Core AI models for chat, vision, and speech in an iPhone or Mac app, without a Python conversion step or bundled weights.
CoreAIKit is a Swift package that lets an iOS or macOS app download a tested Core AI model and run it, handling model selection, download, and caching. It ships with chat, vision, and speech examples, using models such as Qwen3 0.6B, downloaded from Hugging Face on first use and cached locally; no Python, model conversion, API key, or bundled weights are required.
It is an unofficial community package, not affiliated with Apple, and requires Xcode 27 with macOS 27 or iOS 27 to actually run models on a real device, since the Core AI framework is not available in the iOS Simulator; an app can still deploy to macOS 26 or iOS 26 and check availability before using these features at runtime. Each release records the specific Mac, OS and SDK builds, and model revisions it was validated against.
It installs as a Swift Package Manager dependency pinned to an exact version, and streams chat responses through an async API.