Consider a conversation capture mode. It would be nice to capture/synthesize a conversation or dialog, and apply Cleft processing to it. Similar to what Otter.ai voice notes does. I assume your cusom prompts would need to be adjusted to recognize multiple speakers. I tried this with the existing setup, and Cleft provided an interesting first person stream of consciousness account of the conversation. It wasn't true to the dialog, but blended the two speakers thoughts into a relatively coherent paragraph. It did assume genders and names incorrectly, but that was actually welcome, as it fostered seeing the thoughts from someone elses viewpoint. I did ultimately change the pronouns/genders manually. You might consider allowing a command to do that automatically, or generally allowing prompts/commands.