I’m here to help you prevent the LLM from generating a confirmation after using

session.say

in a tool call, so it can move smoothly to the next steps. Here are some friendly methods you can try: • Return No Value from the Tool Function: Simply not returning a value from your tool function will keep the LLM from giving a verbal response. It’s an easy way to run a tool quietly. • Cancel Tool Reply: You can use

ev.cancel_tool_reply()

inside the

function_tools_executed

event. This is especially handy when you have multiple tools running at the same time and want to avoid extra replies. • Interrupt or Skip Generation: By raising

StopResponse()

in the

on_user_turn_completed

hook, you can stop the response generation altogether. • Custom LLM Node: If you want more control, you can override the

llm_node

to detect tool calls and either suppress or tailor the agent’s verbal output. • Defer Response: For Realtime API-based models, setting

defer_response=True

helps prevent automatic replies during interruptions. If you’d like, I’m happy to walk you through detailed examples or assist you further with any of these approaches. Just let me know! Sources: Pipeline nodes and hooks | LiveKit Docs | agent.py | livekit-plugins/livekit-plugins-ultravox/livekit/plugins/ultravox/realtime/...