Skip to main content
The Management API serves account data from the control plane. It never runs inference: model requests go to the Inference API, with an inference key.

Base URL

Authentication

Send a key as a bearer token on every request:
  • GET /credits and GET /activity need a management key (sk-prl-mgmt-). Org admins create them on the Management Keys page; the full key is shown once. A management key reads data for its whole organization and cannot call models.
  • GET /key takes any API key, inference keys (sk-prl-infapi-) included, and describes the key that calls it. An inference key on any other endpoint gets a 403.
  • A revoked, deleted, or expired key stops working right away and gets a 401.

Endpoints

Example
Response

Conventions

  • A successful response is JSON with the result under data.
  • Money is USD as a JSON number, rounded to 9 decimal places, so fractions of a cent still show. Sum large sets of figures with a decimal type if you need exact totals.
  • Dates are UTC calendar days in YYYY-MM-DD format, and timestamps are ISO 8601 in UTC.
  • prompt_tokens counts every input token, including the ones served from the prompt cache; cached_tokens is the cached part of it.
  • Usage figures match the Analytics page. Recent usage can take a few minutes to appear, usage that billing hasn’t priced yet counts as 0 until it is, and costs can differ slightly from your final invoice.

Errors

Every error response is JSON with the HTTP status as code and a message saying what went wrong:
Retry 429, 500, and 503 with backoff. A 400, 401, or 403 fails the same way until the request or the key changes.