self.session.say()

self.session.generate_reply()

llm_node

async def llm_node(self, chat_ctx, tools, model_settings):
        """
        LangGraph entrypoint for LLM response generation.

        Extracts latest user message and full chat history,
        loads or initializes LangGraph state, and invokes the agent graph.

        Args:
            chat_ctx (ChatContext): Conversation history context
            tools (list): Optional tool definitions
            model_settings (ModelSettings): LLM settings

        Returns:
            str: AI response string
        """
        from asgiref.sync import sync_to_async

        if not hasattr(self, "graph"):
            session = await sync_to_async(AgentSession.objects.get)(session_id=self.thread_id)

            agent = await sync_to_async(lambda: session.agent.enabled_agents.first())()
            self.agent_obj = agent

            self.graph = await sync_to_async(lambda: ActiveAgentBuilder(agent).build())()
            self.agent_obj = agent

        user_messages = [
            item.content[0] if isinstance(item.content, list) else item.content
            for item in chat_ctx.items
            if item.role == "user"
        ]

        latest_user_message = user_messages[-1] if user_messages else ""

        state = await sync_to_async(load_or_bootstrap_state)(self.thread_id, self.agent_obj)
        state["last_user_message"] = latest_user_message

        state = await sync_to_async(
            lambda: self.graph.invoke(input=state, config={"configurable": {"thread_id": self.thread_id}})
        )()

        last_ai_message = state.get("last_ai_message")

        if state.get("end_call"):
            print(f"END CALL INVOKED")
            await self.session.say("Thank you for the conversation. Goodbye!")
            await asyncio.sleep(6)
            await self.session.drain()
            await self.session.aclose()

        return f"{last_ai_message}"