databricks-model-serving

To deploy MLflow models to Databricks model-serving endpoints, you register the model and apply it to dedicated endpoints. This workflow supports traditional ML models, custom PyFunc models with signatures, and artifacts for scalable inference.

Yes, you can deploy GenAI agents like ResponsesAgent or LangGraph to Databricks endpoints. This process integrates Unity Catalog functions and Vector Search as tools, allowing you to build and query tool-enabled agents via REST or SDK.

Before serving models on Databricks, you need a Unity Catalog-enabled workspace with model-serving enabled. You also need access to defined foundation model endpoints to successfully deploy and query your ML or GenAI agents.

You can query endpoints and check statuses in Databricks from a single workflow. This allows you to manage deployments, send inference requests to traditional ML models or GenAI agents, and monitor their operational state continuously.

Yes, this approach supports deploying both traditional ML models using MLflow autolog and custom PyFunc models. You can register both types with their respective signatures and artifacts to dedicated Databricks model-serving endpoints.

The best way to host ML models and agent-based workflows in production is deploying them to scalable Databricks model-serving endpoints. This orchestrates training, logging, deployment, and runtime querying for both traditional ML and GenAI agents.