1

Keys

  • The key lives in an environment variable or secrets manager, not in source code.
  • It holds only the scopes the integration uses.
  • You know how to rotate it.
2

Customers

  • You send your own user id as externalId.
  • You store each customer’s Cerium id so you do not call POST /customers on every message.
  • You handle 409 on POST /customers by fixing the data, not by retrying.
3

Retries

  • Every POST /conversations and POST /…/messages carries an idempotencyKey that is stable across retries.
  • You retry only network errors, 429 and 5xx, with backoff and jitter.
  • You honour Retry-After.
4

Timeouts

  • Your HTTP client has a timeout (10 seconds is a sensible start).
  • A slow Cerium call does not block your user’s request. Send messages from a queue or background job where you can.
5

Errors

  • You match on code, not on message.
  • You log the status and code of failures, and never log the key.
6

Volume

  • Lists use limit=100.
  • You do not poll faster than you need to.
  • You have talked to support if you expect heavy traffic.
7

Privacy