{ "instructions": "Classify as: BILLING, SUPPORT, SALES. Reply with the label only.", "input": "I was charged twice this month and I want a refund."}BILLINGClassify incoming messages
Define the categories inside `instructions` and ask for the label alone. No training, no labelled data, no infrastructure.
For harder cases, ask for the reasoning before the label and read only the last line — more accurate, more tokens.
The model returns text. Compare it against your category list and handle anything that does not match.