Limits by plan
The limit applies per organization, not per API key. Every key you issue draws on the same budget, so a Growth organization gets 2,000 requests per minute in total no matter how many keys it holds. Issuing another key does not buy more throughput.
Reading the headers
Every response carries your current position, so you never have to guess:Being limited
Over the limit, you get429 with a Retry-After header:
Handling limits well
Retry with exponential backoff
RespectRetry-After when it is there, and back off exponentially when it is not.
Slow down before you are stopped
X-RateLimit-Remaining lets you ease off rather than hit the wall:
Prefer bulk endpoints
Creating 500 links one at a time is 500 requests.bulk_create_shortlinks does it in one.
get_org_metrics_batch returns several analytics metrics in one call instead of one request per metric.
Ask for larger pages
per_page=100 walks a 1,000-record list in 10 requests rather than 100.
Failing open
If the rate limiter itself is unavailable, requests are served rather than refused. When that happens you will seeX-RateLimit-Remaining: ? — the limit is published, but that request was not counted against it. Treat it as informational.