databricks-agent-bricks

To build a multi-agent system on Databricks, you use a Supervisor Agent (MAS) to orchestrate and route queries across Knowledge Assistants, Genie Spaces, and model endpoints. This Skill provisions and connects these components to assemble conversational applications.

You create a document-backed knowledge assistant by provisioning a Knowledge Assistant (KA) tile. The KA leverages retrieval-augmented generation across documents stored in Unity Catalog volumes to provide document Q&A capabilities.

Yes, you can route queries across both formats using a Supervisor Agent. The Supervisor Agent handles cross-agent routing, directing SQL exploration to Genie Spaces and document questions to Knowledge Assistants.

The best way to set up SQL-based exploration is by creating a Genie Space. This Skill provisions Genie Spaces via dedicated actions, enabling natural language SQL exploration directly over your Unity Catalog data.

Yes, Unity Catalog volumes are required for Knowledge Assistants. The KA component uses these volumes to store and retrieve documents for retrieval-augmented generation, making them essential for document Q&A workflows.

Cross-agent routing works through a Supervisor Agent (MAS) that orchestrates complex queries. The MAS routes user requests to the appropriate Knowledge Assistant or Genie Space based on whether the query targets documents or SQL data.