databricks-agent-bricks

You can orchestrate multiple Databricks agents by using a Supervisor Agent (MAS) to route queries across specialized Knowledge Assistants, Genie Spaces, and model endpoints. This provides structured integration and multi-agent routing within your workspace.

Building a RAG knowledge assistant in Databricks involves using the Knowledge Assistant (KA) component to perform document-based Q&A. You provision a KA endpoint using text and PDFs stored in Unity Catalog Volumes.

Yes, you can use natural language to query SQL data in Unity Catalog by creating a Genie Space. It translates natural language queries into SQL, enabling data exploration and analytics directly over your catalog data.

To provision a multi-agent routing system, use the manage_ka and manage_mas workflows to assemble the components. This codifies the required prerequisites, data preparation, and example-driven evaluation for production deployment.

Deploying multi-brick AI apps requires data preparation in Unity Catalog Volumes, structured agent descriptions, and explicit routing instructions. Following the provided provisioning timelines and best practices ensures a scalable workflow.

Use a Supervisor Agent when you need to orchestrate multiple specialized agents like Knowledge Assistants, Genie Spaces, and external MCPs together. It handles query routing, whereas individual components only perform isolated retrieval or analytics tasks.