Thanks — you're asking whether data hooks are available for realtime models like OpenAI Realtime or Gemini Realtime. Yes — data hooks are available. What you can use • Lifecycle hooks on the Agent:

on_enter()

,

on_exit()

, and

on_user_turn_completed()

to run custom logic at key moments in a session (greeting, cleanup, modify context before reply). • Realtime model node:

realtime_audio_output_node()

to inspect or modify audio frames before they’re sent to the user. • Transcription node:

transcription_node()

to access or alter STT / TTS-aligned transcriptions. • STT / LLM / TTS pipeline nodes (

stt_node()

,

llm_node()

,

tts_node()

) if you use separate components alongside a realtime model. These hooks/nodes work for OpenAI and Gemini realtime integrations and can be implemented in Python or Node.js by extending the Agent class. Would you like a short code example for a specific hook (e.g.,

on_user_turn_completed()

or

realtime_audio_output_node()

) and which language do you prefer? Sources: Pipeline nodes and hooks | LiveKit docs | Azure OpenAI Realtime API and LiveKit | LiveKit docs