In LiveKit agent sessions, the terms Publishers and Subscribers refer to the roles participants take when exchanging media tracks: Publishers: • Publishers send media tracks to the room. These tracks can include: ◦ Audio tracks: From microphones, background music, or synthesized speech from AI agents. ◦ Video tracks: From webcams, screen sharing, or synthetic video/visual effects from AI agents. Subscribers: • Subscribers receive and consume media tracks published by others in the room. For example: ◦ Audio tracks are played on subscribers' speakers. ◦ Video tracks are displayed on subscribers' screens. ◦ AI agents can subscribe to tracks for tasks like speech-to-text or vision-based processing. Participants can act as both publishers and subscribers simultaneously, depending on the application’s needs. LiveKit uses separate peer connections for publishing and subscribing to manage these data flows efficiently. Sources: Realtime media | LiveKit Docs | livekit/src/rtc_engine/rtc_session.rs | OpenAI Realtime API and LiveKit | LiveKit Docs