Structured classification in a single forward pass. Give it a piece of text and a field schema (enums / booleans); one request judges every field at once and returns each value with a calibrated confidence score. No API key needed here — just try it.
No result yet — enter text on the left and hit Classify.
This endpoint is request/response-identical to the upstream API — it just doesn't need an
Authorization header. Point your base URL here and it works.
| Field | Meaning |
|---|---|
context | Required. The text to classify, ≤8000 characters. |
schema | Optional. Omit it to use the server default: fraud_risk / block_account / language / sentiment. |
schema.<field>.type | "enum" or "boolean" |
schema.<field>.choices | Enum only. 1–64 options, each ≤80 characters. |
schema.<field>.description | The judging criteria, ≤500 characters — writing this well is what determines accuracy. |
One hard-won tip: in the description, spell out when each option applies, and give an explicit positive fallback rule ("code with no side effects is always true"). Ask an open-ended question like "is this safe?" and the model skews systematically toward the negative answer — and it will be confidently wrong, so the confidence score won't save you.