We implement rate limits on requests to Foundry APIs to safeguard the system from being overwhelmed by excessive data and to fairly allocate system resources.
Be advised that these restrictions and allocations may be subject to alterations.
Callers should not rely on counting requests, but should instead implement exponential backoff in response to 429 and 503 responses.
Concurrent requests over the limit will get throttled and receive a 429 error.
In addition to the mechanism described above, additional fallback concurrency limits exist that can result in clients receiving a 429 or 503 error.