quota_throttle_expert

HTTP 429 OperationLimitExceeded errors occur when your AOAI deployment TPM capacity is exhausted by token usage. This Skill diagnoses throttling by correlating App Insights token utilization with deployment capacity to identify if saturation is sustained or burst-driven.

To diagnose AOAI TPM throttling, query the gen_ai.client.token.usage metric for your deployment model dimension in App Insights. This Skill computes utilization peaks over the failing window and classifies the throttle pattern to recommend a mitigation step.

Fixing AOAI quota exceeded errors requires reconciling deployment sku.capacity with observed token usage. This Skill recommends scaling capacity, migrating to PTU, introducing rate limiting via APIM, or requesting a quota increase based on the classified bottleneck.

Yes, noisy-neighbor dominance causing TPM throttling on standard-SKU deployments can be detected. This Skill classifies throttle patterns including sustained saturation, short-lived bursts, noisy-neighbor dominance, and regional quota limits using App Insights utilization data.

No, this throttle diagnosis process does not execute changes automatically. It reads Azure deployment metadata for sku.capacity and produces one reviewed mitigation step, such as an Azure CLI action for remediation, without applying the changes.