session.say() and session.generate_reply() serve distinct purposes in LiveKit: • session.say() ◦ Makes the agent speak predefined text you specify. ◦ Requires a TTS plugin for speech synthesis. ◦ Supports optional pre-synthesized audio for lower latency. ◦ Chat history inclusion is controlled via an

add_to_chat_ctx

parameter (default

True

). ◦ Use when you want exact spoken responses (greetings, confirmations, scripted text). • session.generate_reply() ◦ Prompts an LLM to dynamically generate a response based on your instructions or user input. ◦ Requires an LLM (Language Model) to be configured. ◦ Automatically adds relevant context to chat history (either instructions or user input based). ◦ Use when you want the agent to produce context-aware, AI-generated responses. Both methods return a

SpeechHandle

for tracking, support the

allow_interruptions

parameter, and trigger

speech_created

events. In short, use

say()

for fixed text and

generate_reply()

for dynamic, AI-driven responses. Sources: Agent speech and audio | LiveKit Docs | examples/voice_agents/multi_agent.py