The

llm_node

function in the

Agent

class is invoked as part of the agent's processing pipeline. Specifically, it is called by the

_llm_inference_task

function, which is triggered during the execution of the

_pipeline_reply_task

. This task coordinates various operations, including transcription, LLM inference, and TTS. In this flow, the

_pipeline_reply_task

calls

perform_llm_inference

, which starts the

_llm_inference_task

. The

_llm_inference_task

then executes the

llm_node

method, passing the

chat_ctx

(chat context) along with other parameters like tools and model settings to process the context and generate responses. This ensures that the

llm_node

is dynamically integrated into the agent's pipeline to handle LLM interactions effectively. Sources: livekit-agents/livekit/agents/voice/agent_activity.py