Developer mattt’s AnySpeechRecognizer and AnyDecisionModel are community packages for using open models in Swift applications. The former offers an API resembling Apple’s SpeechAnalyzer, with Core ML recognition and diarization for up to four speakers. Its Korean-capable Qwen3 ASR path has limitations including long segment processing and no word-level timestamps. The latter returns structured yes/no, choice or scale decisions and probabilities through multiple local backends rather than generating prose. Published tests ran on an M4 Max with macOS 27, not an iPhone, and substantial meeting-audio transcription errors remain. They establish integration options, not universal accuracy or mobile performance.