Databricks Makes Genie Ontology Generally Available for Enterprise AI
Databricks has made Genie Ontology generally available, giving AI agents a shared layer of business context learned from enterprise work and governed by people. The release adds Genie Search, Ontology Manager and generally available Unity Catalog Domains and Pages.
- Genie Ontology is now generally available from Databricks.
- Genie Search adds access to enterprise documents, tickets and wikis.
- Ontology Manager and Unity Catalog Domains and Pages support curation and governance.
- Genie recorded a 91% pass rate in a Turing benchmark.
- Genie MCP makes the context available to ChatGPT, Claude and Microsoft Copilot.
A shared business context for agents
Databricks describes Genie Ontology as an agentic ontology because agents continuously build its knowledge for other agents to use, while people govern definitions that require precision. It combines learned and modeled knowledge, ranks that information by authority and delivers trusted context while respecting existing user permissions.
Most of the ontology's knowledge is learned from systems and artifacts that reflect how an organization operates. These include Databricks dashboards, SQL queries, notebooks and semantic models in tools such as Power BI. The information is captured as Ontology Snippets, reusable pieces of context covering calculations, business rules and data relationships. Databricks says the snippets are continuously updated as their underlying sources evolve.
People can define and govern business meaning through Unity Catalog semantics. This includes Metric Views for certified KPIs, Domains for organizing knowledge by business area and Pages for authoritative definitions. According to Databricks, organizations can use these tools for the relatively small set of definitions that must be exact, such as recognized revenue or ARR.
Ranking context across the enterprise
When an agent asks a question, OntoRank evaluates learned and modeled knowledge to identify the most trustworthy context. Databricks says it considers source authority, certification, the expertise and relevance of the people behind the knowledge, popularity and recency when resolving conflicting definitions.
The resulting context can be delivered to agents running outside Databricks through Genie MCP, including ChatGPT, Claude and Microsoft Copilot. Databricks says this allows teams to make the agents they already use more knowledgeable about their business without rebuilding them.
The release also introduces Genie Search, now generally available. It lets people and agents find enterprise knowledge in documents, tickets and wikis alongside governed Databricks assets, within their existing permissions. Genie can return synthesized, cited answers alongside dashboards, tables and Genie Agents. Databricks says indexed content also informs OntoRank, allowing documents and tickets to become evidence in decisions about which sources to trust.
Benchmark results and new data sources
Databricks worked with Turing on a benchmark that simulated a live Databricks workspace for a fictional UK energy supplier. The test compared Genie, powered by Genie Ontology, with two frontier agents identified as Agent A and Agent B. Each agent answered the same questions five times per task, with answers graded against a rubric set before the runs.
Genie had a 91% pass rate, compared with 45% for Agent A and 58% for Agent B. Its average wall time per run was 100 seconds, versus 435 seconds for Agent A and 307 seconds for Agent B. Genie used an average of 432K tokens per run, compared with 889K and 1.05M. Databricks says Genie passed 1.6 times as many runs as the stronger frontier agent, used less than half the tokens and finished three to four times faster.
Databricks attributes the difference to context. Genie used knowledge already learned from the workspace, while the other agents had to rediscover the relevant business rules on each run.
What the release means for enterprise users
Genie Ontology is aimed at organizations deploying multiple agents across business functions. Databricks argues that separate, agent-by-agent context creates partial views that can conflict, while a shared ontology can give agents a common, governed understanding of how the business works.
Genie Search supports Google Drive, Jira and Confluence, according to the announcement. Administrators connect sources and enable indexing through the admin console. For real-time tasks, such as checking or creating a Jira ticket, Genie Search works alongside MCP connectors, with Genie deciding whether to use the index, MCP or SQL.
The benchmark reflects an established deployment where teams had already accumulated business knowledge. Databricks says account teams can help customers run the same evaluation in their own workspaces, giving organizations a way to assess how an agentic ontology performs against their own context and workflows.
Databricks: Announcing GA of Genie Ontology
