Process an operator's recorded audio for a ticket with one
multimodal model call. dictate returns a verbatim transcript; smart
treats the audio as spoken keywords and drafts a fluent reply using the
ticket + conversation as context (no knowledge-base retrieval). The client
uploads the audio via POST /uploads first and passes the resulting URL.
Bearer <token>.Public URL of the uploaded audio clip (from POST /uploads).
dictate returns the transcript; smart returns a drafted reply.
Ok