Thanks — regarding intercepting an agent’s generated text before it goes to TTS: yes, you can. To do that, override the agent’s

llm_node

and stream the LLM output yourself. Inside your custom

llm_node

you can inspect, modify, filter, log, or replace text chunks as they arrive, then yield those (possibly modified) chunks so they continue to the TTS stage. Minimal example:

async def llm_node(self, chat_ctx, tools, model_settings=None):
    async def process_stream():
        async with self.llm.chat(chat_ctx=chat_ctx, tools=tools, tool_choice=None) as stream:
            async for chunk in stream:
                if chunk is None:
                    continue

                content = getattr(chunk.delta, 'content', None) if hasattr(chunk, 'delta') else str(chunk)
                if content is None:
                    yield chunk
                    continue

                # Intercept/modify text here
                processed = content.replace("undesirable text", "")

                # If chunk object holds content, update it; otherwise replace chunk
                if hasattr(chunk, 'delta') and hasattr(chunk.delta, 'content'):
                    chunk.delta.content = processed
                else:
                    chunk = processed

                yield chunk
    return process_stream()

If you want, tell me: • Which agent framework/version you’re using, • Whether you need synchronous blocking or streaming behavior, • Any specific modifications (filtering, censorship, markup, logging), and I’ll provide a tailored snippet or integration notes. Sources: Pipeline nodes and hooks | LiveKit docs | Modifying LLM output before TTS | LiveKit docs | LLM Output Replacement | LiveKit docs