Thanks — I see you're hitting the error: AgentSession.*init*() got an unexpected keyword argument 'before_tts_cb' This means the

AgentSession

constructor does not accept a parameter named

before_tts_cb

. The

AgentSession

options include things like

stt

,

llm

,

tts

,

vad

, and

turn_detection

, but

before_tts_cb

is not a supported init argument. What you can do instead • Override the TTS node in a custom

Agent

subclass to run preprocessing before text is sent to the TTS engine. This is the supported way to inject logic before TTS runs. • Example pattern to override `tts_node`:

async def tts_node(self, text: AsyncIterable[str], model_settings: ModelSettings) -> AsyncIterable[rtc.AudioFrame]:
    # custom text preprocessing can go here
    async for frame in Agent.default.tts_node(self, text, model_settings):
        # custom audio postprocessing can go here
        yield frame

Next steps for me to help further • If you want, paste the code where you construct

AgentSession

(the init call) and the

Agent

subclass you’re using. • Tell me whether you intended

before_tts_cb

to modify the text before TTS, or to change audio after TTS — that will determine whether to put logic in

tts_node

(pre/post) or another node. If you share the snippet, I can show exactly how to move your callback logic into

tts_node

or another appropriate hook. Sources: Agent session | LiveKit Documentation | Pipeline nodes and hooks | LiveKit Documentation