all docs
/ troubleshooting

503 overloaded with Retry-After

ACE shed the request on its own capacity, not the provider's. Back off for the Retry-After it sends, or send direct to the provider.

Symptom

Calls return HTTP 503 with x-ace-error: overloaded and a Retry-After header (overloaded_error in the body on the Anthropic surfaces).

Cause

The gateway refused the request on its own capacity — at the transport, at engine entry after the request queued too long, or at dispatch when every destination was at its concurrency limit. Neither your key nor the provider is involved.

Fix

Retry after the Retry-After interval (1–30 s, computed from what the process has drained lately), or send the call direct to the provider. GET /healthz reports the gauge under load.

Source