Question

A GenAI engineer is deploying a RAG chain to a Databricks Model Serving endpoint. The chain calls an external third-party API that requires an API key. The engineer must ensure the key is available to the endpoint at inference time without hardcoding it into the logged model artifact or exposing it in plaintext. What is the recommended approach?

Accepted Answer

Store the key in a Databricks secret scope and reference it in the endpoint configuration using an environment variable that resolves the secret at runtime

Answer

Include the API key as a default parameter value in the MLflow model signature so it is packaged with the model

Answer

Add the API key to a config.py file inside the chain code before logging the model with MLflow

Answer

Pass the API key as part of every inference request payload from client applications