A Generative AI Engineer has a provisioned throughput model serving endpoint as part of a RAG application and would like to monitor the serving endpoint's incoming requests and outgoing responses…

Vector Search is used to index and retrieve embeddings for RAG applications and does not log serving endpoint traffic.

Lakeview is Databricks' dashboard and BI visualization layer and is not designed to capture model serving logs.

DBSQL (Databricks SQL) is a query engine for running SQL analytics on data and does not intercept or store model serving endpoint traffic.

Inference Tables is a built-in Databricks feature that logs all incoming requests and outgoing responses from a model serving endpoint directly into a Delta Lake table. It requires no additional infrastructure or custom middleware, making it a simpler and more integrated solution than a separate microservice. This enables monitoring, debugging, and auditing of production LLM traffic natively within Databricks.