Wiring a multi-agent system to Genie Spaces or a conversational API

Multi-agent systems built with Databricks (e.g., using Mosaic AI Agent Framework) can incorporate a Genie Space as a tool, allowing an orchestrator agent to route natural-language questions about structured data to Genie for SQL-based retrieval over governed tables. This lets agents combine conversational data exploration (via the Genie Conversation API) with other tools like vector search or function calling in a single multi-agent workflow.

1 · Learn the must-know

• 1Genie Spaces let business users and agents ask natural-language questions over curated Unity Catalog tables, and Genie translates these into SQL queries automatically.
• 2The Genie Conversation API exposes programmatic endpoints (start conversation, send message, poll/get message status, get query results) so an agent can invoke Genie as a callable tool rather than through the Genie UI.
• 3In a multi-agent architecture, a supervisor/orchestrator agent can register a Genie Space as one of several tools (alongside custom functions or retrievers) and route structured-data questions to it based on intent.
• 4Because Genie queries run against governed Unity Catalog data, existing table permissions and row/column-level security are respected, so the calling agent's identity or service principal must have appropriate access.
• 5Genie responses may include both a natural-language answer and the underlying SQL/result table, which the orchestrating agent can further process, summarize, or pass to another agent in the chain.
• 6Genie Spaces are best suited for structured/tabular data retrieval, so multi-agent designs typically pair them with a vector search or retrieval tool for unstructured data to cover both data types.

2 · Check your understanding