Agilix Rate Limits

Buzz throttles on three distinct axes and the provider is explicit that all three can answer 429 with a Retry-After header, so a client must read the other response headers to tell them apart. (1) Time limiting — a per-customer budget of server processing time, bucketed per endpoint hostname, which is what most legitimate 429s are. (2) Rate limiting — per-endpoint call counts in a five-minute window, applied to security-sensitive and expensive-AI commands. (3) Server-busy limiting — 503 when autoscaling has not caught up. Specific rate-limited commands and their numeric limits are deliberately NOT published ("subject to immediate change in response to security and availability issues"); the provider's stated substitute is that the live limit is always readable from the response headers.