#C088ZNU7QQ5 i am using realtime model of google. i want to collect the user complelted turn to record the data like how many questions did the user ask. is there any way so that i can collect those?

Thanks for reaching out, Karthik. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

i am using python. what i want is when user ask questions and the agent replies, then i want this to count as 1 so that based on this i can keep track of the questions asked.

Thanks — I’ve got what you want: using Python, count one interaction when a user asks a question and the agent finishes replying. Below is a simple, reliable pattern you can use with LiveKit Agents. Summary of the approach • Listen for the

user_input_transcribed

event and only react to final transcripts. • When a final transcript arrives, generate the agent reply and wait for the reply to finish playing. • Increment the interaction counter only after the agent reply completes. Minimal Python example

import asyncio
from livekit.agents import AgentSession

# Create your session (example placeholders for STT/LLM/TTS/VAD)
session = AgentSession(
    # supply your configured stt/llm/tts/vad plugins here
)

interaction_count = 0  # counts user question -> agent reply interactions

async def handle_user_transcript(event):
    global interaction_count

    # Only count when transcript is final (user finished speaking)
    if not event.is_final:
        return

    user_text = event.transcript

    # Ask the agent to generate a reply and get a handle that resolves when playback completes
    # Depending on your SDK, this may be session.generate_reply or session.reply; adapt accordingly
    handle = await session.generate_reply(user_input=user_text)

    # Wait for the agent's reply (speech/playback) to finish
    await handle

    # Now count this as one completed interaction
    interaction_count += 1
    print(f"Completed interactions: {interaction_count}")

# Register the event listener
@session.on("user_input_transcribed")
def on_user_input_transcribed(event):
    # Run async handler without blocking the event loop
    asyncio.create_task(handle_user_transcript(event))

Implementation notes and suggestions • Ensure transcription is enabled so

user_input_transcribed

fires (transcription is enabled by default). • Decide what “question” means for you: ◦ Simple: count every final user turn like the snippet above. ◦ Strict: run a lightweight rule (ends with “?” or contains question words) or a small intent classifier and only call generate_reply/count when it’s a question. • Persistence: ◦ For short sessions, in-memory

interaction_count

may be fine. ◦ For long-term analytics or cross-session totals, write each increment (or the full event record: transcript, timestamp, session_id, user_id) to a database or analytics service inside the handler. • Concurrency: ◦ Using

asyncio.create_task

avoids blocking the event loop so multiple user turns or other events remain responsive. • Edge cases: ◦ If your agent can interrupt or produce multiple reply segments, ensure your chosen API handle resolves only after the whole reply is complete. ◦ If a user asks multiple short questions in one turn, decide whether that counts as 1 or multiple — you’d need extra NLP logic to split/count questions inside the transcript. If you’d like, I can: • Provide a version that only counts transcripts recognized as questions (with a short heuristic). • Show how to persist counts to a specific database (Postgres, Redis, etc.) — tell me which you use. • Adapt the example to match the exact LiveKit Python SDK method names you’re using if you share those. Sources: Interrupt User | LiveKit Documentation