The Genie One MCP server is generally available
Databricks made the Genie One MCP server generally available, enforcing Unity Catalog permissions on every call. The beta endpoint shuts down on October 31, 2026.
By Brandon Gross
Data + AI Summit 2026: Genie One, business semantics and controls for agents
At Data + AI Summit 2026 Databricks made Genie One, Genie Agents and Genie Code generally available, added domains and a business glossary to Unity Catalog, extended Agent Bricks and put guardrails, spend controls and tracing for agents into Unity AI Gateway. Here is everything announced, with release stages.
By Brandon Gross
Databricks previews Genie Ontology
Databricks put Genie Ontology into public preview. It builds a graph that links metrics, business terms, tables and teams, and respects existing permissions.
By Matt O'Donnell
Databricks makes metric views generally available
Databricks made Unity Catalog metric views generally available as part of Unity Catalog Business Semantics, contributed metric views to Apache Spark, backed the Open Semantic Interchange and made the Genie Conversation API generally available.
By Brandon Gross
Databricks adds a Genie research agent in beta
Databricks released a research agent in Genie as a beta, along with beta APIs to create and update Genie spaces.
By Brandon Gross
Databricks introduces managed MCP servers in beta
Databricks released managed MCP servers in beta for Genie spaces, Vector Search indexes and Unity Catalog functions, with on-behalf-of-user authentication.
By Brandon Gross
Data + AI Summit 2025: AI/BI Genie is generally available
Around Data + AI Summit 2025, Databricks made AI/BI Genie generally available, put Unity Catalog metric views into public preview and launched Agent Bricks in beta.
By Matt O'Donnell
Databricks opens the Genie Conversation API in public preview
Databricks put the Genie Conversation API into public preview, so applications and chat tools can hold multi-turn conversations with a Genie space.
By Matt O'Donnell
See it on your own data
Together we pick one business area, the one where a wrong answer costs the most. Two weeks later you can ask a real question and check the answer.