Skip to main content
Turn detection determines when a user has finished speaking and it’s the agent’s turn to respond.

What is Turn Detection?

Turn detection analyzes conversation context to predict:
  • End-of-turn probability - Likelihood that the user has finished speaking
  • Dynamic endpointing - Adjusts delay based on probability
    • High probability → Shorter delay (0.5s)
    • Low probability → Longer delay (3.0s)

How Turn Detection Works

Turn detection processes conversation context:
Conversation History

Current Transcript

Turn Detector

End-of-Turn Probability

Dynamic Endpointing Delay

Turn Committed

Configuration

Basic Configuration

from kuralit.server.agent_session import AgentSession

# Using Multilingual Turn Detector
agent = AgentSession(
    turn_detection="multilingual/v1",  # Multilingual turn detector
    # ...
)

Available Providers

  • Multilingual - Supports multiple languages, no API keys needed
View turn detection providers →

Next Steps