Voice settings
Stored only in this browser (localStorage) — there's no backend in this project, so keep this key personal.
Auto tries Gemini's native realtime voice first and silently switches to the standard speech pipeline if it can't connect.
Used for both text mode and the fallback voice pipeline. Personality still comes from personality.md in the project root.