Press the mic to start. The agent greets you first, then listens — speech in, on-device model, speech out.

• 1. Agent speaks a short greeting
• 2. You hold to talk
• 3. Model replies on-device, then speaks