Language models can do a lot, but they're only a small part of what machine learning can do. Today we're introducing two new Swift packages for speech and decisions, with open models from 🤗 Hub that run on device:

"Mom, can we get GPT?" "No, we have GPT at home." Introducing preowned: a local server that puts the models built into macOS behind an OpenAI-compatible API. Your Mac drove off the line with a language model, an image maker, a transcriber, a translator, and a trunk full of voices. Point any OpenAI client at preowned, and chat (with streaming and structured output), embeddings, transcription, translation, and speech run right there on your Mac. Ask for gpt-4o or whisper-1, and you get Apple's model for the job. Don't worry about what's under the hood. Any API key works — no credit check required! Sold as-is: no tool calling, no Responses API, and on macOS 27, Image Playground wants a person to click Done (or a hack that clicks it for you). Requires macOS 26 or later on Apple silicon, with Apple Intelligence turned on for chat and images. Take it for a test drive: brew install mattt/tap/preowned preowned serve Come on down! 🚗

iMCP 1.8.0 is out, and it packs in so many new features that we had to add a second row to fit them all! Mail, Notes, Voice Memos, and Music join the lineup.