Skip to content

Conventions

The same rules apply to /v1 of every product. Whether an operation supports a rule, you see in the reference with that operation.

Every JSON response has six fields:

FieldMeaning
dataThe result
contextThe context of the call: organisation, filters, and on an error error
completenessHow complete it is: total, returned, truncated
notIncludedWhat this result does not tell, and which call does
provenanceWhere it comes from
nextLogical follow-up steps

Exceptions are in the reference. In the LEO family /models and /chat/completions keep the OpenAI shape, and /openapi.json returns the document itself. ORDO returns the envelope when you send Accept: application/vnd.prudai.task+json; files keep their own format.

completeness.total is the exact total, or null when unknown; never an estimate. The parameters are in the reference per operation: limit with after and next_cursor in the LEO family, limit with offset in IRMA and ORDO.

A write that supports Idempotency-Key does not run the same key with the same body twice. The same key with a different body gives 422 idempotency_key_reused.

With ?dryRun=true you see what a write would do. Nothing is stored and the approval is not touched. Not every operation has a dry run: that is in the reference.

A change that supports If-Match is sent with the ETag you read. If the object changed in the meantime you get 412. If the header is missing you get 428.

DELETE deletes softly: the object can be restored through restore. Permanent deletion is only possible in the app. In the LEO family a conversation stays restorable for 30 days.

/v1 only changes by adding. Responses can gain new fields; ignore what you do not know. In the LEO family requests are strict: an unknown field gives 400.

  • X-Request-Id is on every response in the LEO family. Quote it to support.
  • RateLimit-Limit, RateLimit-Remaining and RateLimit-Reset give the tightest limit that applies.
  • Times are RFC 3339 in UTC.