Skip to main content

​
What you saw

The HTTP status is 429, which most SDKs treat as retryable and back off from on their own.

​
Why it happened

The gateway caps each user at 200 requests per minute. The cap is there mostly to keep runaway loops from melting the proxy. Your agent crossed it inside a 60-second window.

​
How to fix it

  1. Add a small delay or exponential backoff on retry. Most SDKs handle this automatically when they see a 429.
  2. If parallel workers are hitting the gateway, throttle them so they share the budget.
  3. On Cloud, the 200 requests/minute cap is fixed. Lean on the backoff in step 1. Self-hosted, raise it with MANIFEST_RATE_MAX_REQUESTS in your .env and restart. See Environment variables.