models-macos

Answer Mac apps’ calls to Apple’s models from prebuilt .axxmodel files in the app’s axx-models folder (in its ./ files, its home’s Library/Application Support): Vision (image classification, text recognition, human and animal rectangles, person segmentation), Speech (SFSpeechRecognizer) and Foundation Models text sessions, of the system model and of Apple’s server model (PrivateCloudComputeLanguageModel; a session of a model of the app’s own is left as it is), as models-ios does in the iOS Simulator, with the same contracts (macos.* APIs). desktop-macos loads Axx’s library into the app as it starts, from its environment (DYLD_INSERT_LIBRARIES): macOS lets it into an app signed without the hardened runtime, or with the com.apple.security.cs.allow-dyld-environment-variables and com.apple.security.cs.disable-library-validation entitlements (the app’s test build’s), which Axx checks before it starts the app. Speech uses log-mel matching for app-provided audio and the microphone step’s precomputed matches for queued recordings; microphone scenarios feed AVAudioEngine input taps and connections, AVAudioRecorder files and AVCaptureSession audio data outputs, then silence (a capture session still opens the Mac’s audio capture device, which needs one, and the user’s permission to use it). A Mac has no test’s grant of speech recognition or the microphone: a stubbed recognizer, and the microphone Axx feeds, are authorized. A stub whose contract gives format.details.availability (deviceNotEligible, appleIntelligenceNotEnabled or modelNotReady) is a Mac without the system model, which reports it unavailable. Foundation Models stubs need macOS 27 or later; the rest, macOS 15. Copy stubs before launching the app; restarting reloads them.

Add it to a project with axx pack add models-macos; it builds on models-core and desktop-macos and app-core, which comes with it.