databricks-model-serving

Deploy MLflow models to Databricks Model Serving endpoints by logging the model with Unity Catalog resources, specifying pip_requirements, and creating an asynchronous deployment job to publish a scalable REST endpoint for production inference.

Serve custom Python logic on Databricks by packaging preprocessing, postprocessing, and complex logic as mlflow.pyfunc models with defined signatures and pip_requirements, then registering and deploying them via Model Serving endpoints.

Databricks Model Serving supports deploying GenAI agents like ResponsesAgent by logging the agent to MLflow, registering Unity Catalog functions and Vector Search resources, and deploying asynchronously via jobs for scalable inference.

Integrate Unity Catalog functions and Vector Search with deployed LLM agents by registering them as callable tools during MLflow logging, enabling automatic resource registration for authentication passthrough and seamless querying.

Query a deployed Databricks serving endpoint using the MLflow Deployments SDK, Databricks SDK, or REST API to interact with and manage your serving endpoints for both classical ML models and GenAI agents.

Unity Catalog is required to manage authentication passthrough for deployed models and agents, automatically registering integrated functions and Vector Search retrievers as callable tools to ensure secure resource access during inference.