Continuously perceives live audio and video while deciding whether to respond, interrupt, or keep listening. It couples interaction reasoning with audiovisual generation for real-time, full-duplex exchanges.