Hi, the conversation API is great but I don’t understand what’t the use case/when do we need to create an item:
https://platform.openai.com/docs/api-reference/conversations/create-items
My flow is to create a conversation, then add the conversation id in the response API.
There are a few possible uses of new message addition feature, besides that you can add messages when creating the conversation ID itself:
Preloading a conversation with messages: a developer message instead of using “instructions”, multi-shot examples of how to respond as training. Note that the ‘conversations’ mechanism can auto-drop old turns, so it is not a good place for a permanent message if you use “truncation”…“auto”.
Applying the new input directly to the conversation object instead of as “input” of an API call, so retries don’t need to re-send input against an uncertain state of “conversation”. Requiring more latency of another API call to do so.
Externally-run products such as knowledge search that don’t involve the conversations AI itself can be placed as an understandable context, instead of a multiturn addition as “input”.
Or: copying out all the objects to make a new conversation or branch, for example, to limit the length being run, to strip old reasoning and tools, to overcome the poor implementation and the unbounded costs.
“Great” is highly subjective, it seems.