Each API key is limited to about 1,200 requests per minute, with room for short bursts. Requests over the limit are refused with 429:
Wait for the number of seconds in Retry-After, then try again.

Staying within the limit

Use limit=100 on lists rather than many small pages.
Checking a conversation every second for a reply uses 60 requests a minute for one user. Poll slowly, and stop when the conversation is closed or the screen is not in use.
Store each customer’s Cerium id after you create them. Do not call POST /customers before every message.
Do not retry in a tight loop. Use exponential backoff with jitter, as in the retry pattern.
The limit applies per key, so a batch job on its own key cannot slow your live traffic.
If you expect sustained traffic above this limit, contact support before launch.
Requests may also be refused if the body is larger than 1 MB (413).