databricks-model-serving

You can deploy MLflow models to Databricks Model Serving by logging them into Unity Catalog via autolog or custom pyfunc, then running a job-based deployment to create and query the endpoint.

Yes, you can serve ResponsesAgent or LangGraph agents on Databricks by logging them to Unity Catalog, deploying them as scalable endpoints, and querying them using OpenAI-compatible message formats.

Yes, Databricks Model Serving supports tool calling by integrating Unity Catalog Functions and Vector Search with your deployed GenAI agents for actionable inference.

You package custom preprocessing logic for MLflow model serving by using MLflow pyfunc to bundle your inference logic and signatures alongside the model before registering it in Unity Catalog.

The correct input format for querying a Databricks serving endpoint depends on the model type: GenAI agents use OpenAI-compatible messages payloads, while classical ML models require correctly formatted inputs matching their signatures.

Querying a Databricks model endpoint fails immediately after deployment because the endpoint requires readiness handling before inference, meaning you must check endpoint status to ensure it is fully provisioned before querying.