Thanks — I’m responding to your question about how to get live transcription from an agent via API. You can receive real‑time (interim + final) transcriptions from an agent by subscribing to the agent session’s transcription events on the backend and/or by reading the

lk.transcription

text stream on the client. Below are the key pieces and example code to get you started. Backend — listen for transcription events • Subscribe to the

user_input_transcribed

event on the AgentSession. • The event payload includes: ◦

transcript

(the text) ◦

is_final

(boolean: false for interim updates, true when complete) • Handle interim updates differently from final results if you need partial display vs. storage. Python example:

@session.on("user_input_transcribed")
def on_transcript(transcript):
    if transcript.is_final:
        print(f"Final transcript: {transcript.transcript}")
    else:
        print(f"Interim transcript: {transcript.transcript}")

Frontend — receive text stream messages • Agent text responses and transcriptions are sent over a TextStream topic named

lk.transcription

. • Register a

TextStreamHandler

for that topic and read messages. Use the

lk.transcribed_track_id

attribute to tell user transcriptions apart from agent responses. JavaScript example:

room.registerTextStreamHandler('lk.transcription', async (reader, participantInfo) => {
    const message = await reader.readAll();
    if (reader.info.attributes['lk.transcribed_track_id']) {
        console.log(`User transcription: ${message}`);
    } else {
        console.log(`Agent response: ${message}`);
    }
});

Quick notes and next steps • Interim transcripts arrive while the speaker is still talking; final transcripts arrive when speech ends. • If you want a full working example, check the Pi Zero Transcriber recipe which demonstrates logging real‑time final transcriptions to a file. • If you’d like, tell me: ◦ Which language/runtime you’re using on the backend (Python, Node, etc.) ◦ Whether you need interim transcripts shown in a UI or only final transcripts saved ◦ If you plan to stream agent text-only responses or also want audio forwarded to an external STT provider I can then provide a tailored code sample or an integration checklist for your stack. Sources: Pi Zero Transcriber | LiveKit Documentation | How to build an agent with speech-to-text input and text-only output | Live... | Text and transcriptions | LiveKit Documentation