PRODUCTION SCENARIO
A Generative AI Engineer must report monthly token consumption per requesting identity across all 12 governed foundation model endpoints. Finance wants one query that covers every endpoint rather than a separate extract per endpoint.
Which table should the engineer query?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: The system.ai_gateway.usage system table
Gateway usage tracking writes system.ai_gateway.usage, whose schema carries requester, the ID of the user or service principal that made the request, alongside input and output token counts and the endpoint name, so one query spans every endpoint. Billing usage records DBUs by SKU rather than tokens per identity, an inference table logs request and response payloads for one endpoint at a time, and evaluation metrics measure answer quality rather than consumption.
Official Source: https://docs.databricks.com/aws/en/ai-gateway/usage-tracking ↗