Genie One

Genie One — the natural-language assistant in Databricks One — answers business questions over your data without anyone writing SQL. Connect the Nimble MCP server and Genie can also reach the live web: ask a question in plain English and Genie calls Nimble’s search, extraction, and Web Search Agent tools, then answers with cited sources. No notebook, no SQL, no leaving Databricks.

​

What you get

• The live web in natural language. Business users ask Genie a question and get an answer grounded in current web data — no SQL, no tool names, no API keys.
• Reuses the governed Nimble connection. Genie calls the same Unity Catalog MCP connection installed from the Marketplace — so every web call inherits the same Unity AI Gateway audit, access control, and governance.
• Cited, inspectable answers. Genie shows each tool call (request and response) inline and cites the source URLs in its final answer.
• Stays inside Databricks. Only authorized outbound calls hit Nimble; the connection and its credentials live in Unity Catalog.

​

Prerequisites

• The Nimble MCP server installed as a Unity Catalog connection — follow Marketplace & MCP first. Genie connects to that connection (for example nimble-mcp-marketplace); it does not create its own.
• Access to Genie One in your Databricks One workspace.
• USE CONNECTION on the Nimble connection (granted on the Marketplace & MCP page).

​

Connect Nimble in Genie One

​

See it work

Ask Genie a question that the warehouse alone can’t answer — something that depends on what’s live on the web right now:

Which noise-cancelling headphones are top-rated this year, and how do their current prices compare? Use Nimble and cite sources.

Genie routes it to the Nimble MCP server and shows the call inline:

• Calling tool → Running Nimble Search. Genie expands a thinking trace as it works.
• Request / Response. Each completed call shows the exact request Genie sent ({"query": "best noise-cancelling headphones price comparison", "max_results": 8, "search_depth": "deep", "focus": "shopping"}) and the raw response — full transparency into what was fetched.
• Cited answer. Genie synthesizes the results into a ranked comparison with prices, ratings, and source citations.

Because Genie calls Nimble through the governed Unity Catalog connection, every one of these tool calls is auditable in the Unity AI Gateway audit table — same as any other proxied Nimble request.

​

Try these questions

Paste any of these into Genie One once the Nimble connection is enabled. Each forces a different Nimble tool.

​

Roll out across your organization

• Install once, share the connection. The Nimble MCP connection is a single Unity Catalog object. Grant USE CONNECTION to the business-user groups who should query the web in Genie; they never see or manage the API key.
• Govern through Unity AI Gateway. Every Genie-initiated Nimble call routes through the managed proxy and lands in system.access.audit. Use the AI Gateway Usage Analytics dashboard to track web-tool usage by user — see Marketplace & MCP.
• Mind the Beta limits. Custom MCP connections in Genie don’t yet gate individual read/write actions behind approval. Until that ships, scope USE CONNECTION deliberately and prefer it for read-oriented research use cases.
• Prefer SQL for batch? When the job is enriching many rows rather than answering one question, the same Nimble capabilities are available as Unity Catalog table functions — see Data Enrichment.

​

Resources