mint rain
I'm hoping someone out there has a write up of the higher level model/thinking/design of the Realtime API messages. I've figured out there are responses, items, and content parts, and that user messages, assistant text modality messages, and assistant voice drive messages all juggle these concepts with event type messages.
But where I just got tripped up was related to the content array and content_index fields. Apparently for the same response item (by id) and same content index, the object might be an type:audio with a transcript or a type:text with a text field. I don't understand what this is modeling if that isn't consistent.
(PS - Suggestions for tags in this discussion might include "Docs" and each of the high level API types: "Completion", "Assistant", "Realtime", etc...)