astonishing-kangaroo-12980
12/19/2025, 9:36 AMtall-airline-16756
12/19/2025, 9:36 AMtall-airline-16756
12/19/2025, 9:37 AMastonishing-kangaroo-12980
12/19/2025, 9:38 AMtall-airline-16756
12/19/2025, 9:39 AMastonishing-kangaroo-12980
12/19/2025, 9:40 AMtall-airline-16756
12/19/2025, 9:40 AMlk.transcription) for UI display or client-side handling. This does not trigger agent logic.
• Run a minimal LiveKit agent backend that uses the Soniox plugin—this is the supported way to keep the full agent workflow (STT → LLM → TTS) intact.
• Keep everything JS-only and use a JS-supported STT provider (Deepgram or OpenAI) so you can both transcribe and integrate with LiveKit without a Python backend.
Which option do you want to proceed with? If you choose the first, I can show a Node.js example to send transcripts into the room; if the second, I can outline the minimal agent setup; if the third, I can provide a LiveKit + Deepgram/OpenAI JS example.
Sources: Soniox STT plugin guide | LiveKit docs | Transcriber | LiveKit docs | Text and transcriptions | LiveKit docs | Sending text | LiveKit docs