Abstract
Apple's Foundation Models framework, announced at WWDC 2025, gives iOS developers direct access to a 3-billion parameter on-device language model — no cloud, no API costs, and no user data leaving the device. This tutorial introduces the framework through a practical banking app example: categorising transactions using structured AI output. It covers availability checking, the @Generable macro for type-safe responses, @Guide for constrained values, system instructions, streaming output, and tool use for letting the AI query live app data. It also addresses production concerns such as input validation and graceful error handling. Developers building apps in sensitive domains — banking, health, personal finance — will find this a practical starting point for integrating on-device AI without privacy trade-offs.