Assistants API | New 'role' to differentiate final message while streaming

[hayescode]

I am loving the new Assistants API and the constant improvements. One item I'm struggling with is determining if the thread.message is the final answer (it has everything it needs and is responding) or if it's trying to fix an issue with a tool call. This seems limited to streaming, since non-streaming the last message is the final answer. My UI is Chainlit and looks something like this, where tool calls are nested/collapsible and the final response is displayed.

In this example, if there is an issue with any of the tools it "talks out loud" and tries to fix the issue. This is great but in the API this looks identical to the final response. It would be great to be able to differentiate this while streaming.

To reproduce, ask your assistant to Create a visualization of a sine wave using Plotly. It will fail because plotly isn't installed and will print messages followed by more tool calls.

Suggestion: Either a new role besides assistant/user or a new field for "intention" with values like "troubleshoot"/"tool_call_id" or "final answer".

class MessageDelta(BaseModel):
    content: Optional[List[MessageContentDelta]] = None
    """The content of the message in array of text and/or images."""

    role: Optional[Literal["user", "assistant"]] = None
    """The entity that produced the message. One of `user` or `assistant`."""

[rattrayalex]

Thanks for the suggestion, I've passed it along!