Voice assistants are getting easier to deploy, but they’re still generic. This talk follows my attempt to build a mostly-local Home Assistant voice system that not only controls the house, but has a personality. We’ll explore the trade-offs between local and cloud services, improving recognition for regional accents, creating custom Piper voices, and adding character to an assistant that would otherwise sound like every other synthetic voice. Along the way we’ll discover where cloud services are still better, where local solutions shine, and why a mildly judgemental anime tsundere can be more enjoyable to live with than a perfectly efficient assistant. Hmpf.