Totally get it I’ve faced that too. Deepgram sends the transcript but nothing goes to the LLM, so the agent just hangs. It’s usually a timing or connection issue between the transcription and model call. I can share what’s helped me stabilize it if you want.