Announcing GA of Genie Ontology

Databricks announced the General Availability of Genie Ontology, an agentic ontology layer that agents build and use while people govern its business definitions, with expanded knowledge sources and new curation tools. Databricks said thousands of customers already run Genie Agents in production, including 7-Eleven, Virgin Atlantic and General Motors, and it shared benchmark results developed with Turing showing agents work more accurately when drawing on the ontology. Genie Ontology learns business context from Databricks dashboards, SQL queries, notebooks and Power BI semantic models as reusable Ontology Snippets, complemented by modeled knowledge governed through Unity Catalog semantics and delivered to agents while respecting existing user permissions.

The agentic ontology, built by agents, governed by people, and designed for every agent, is now Generally Available with expanded knowledge sources and new tools to curate and refine it. by Ken Wong https://www.databricks.com/blog/author/ken-wong , Paul Roome https://www.databricks.com/blog/author/paul-roome and Richard Tomlinson https://www.databricks.com/blog/author/richard-tomlinson AI agents are only as useful as their understanding of the business in which they work. Within a given domain, teams can curate that understanding by hand. Thousands of customers now run Genie Agents https://www.databricks.com/product/genie/agents in production this way, including 7-Eleven, Virgin Atlantic and General Motors. However, enterprises are no longer deploying a single agent for a single domain. They are deploying many agents across every function, and each needs the same accurate and complete understanding of how the business works. The enterprise context that agents need is far larger than any team can write down or model by hand. It lives in the logic behind thousands of production dashboards, SQL queries, notebooks and BI semantic models, as well as in the documents and applications where teams record definitions and maintain a working knowledge of who does what. It also changes every week as products launch, campaigns run, and teams reorganize. This is why ontologies have historically struggled in the enterprise. Built by specialists and maintained manually, they captured only a narrow slice of the business and quickly became outdated as the business evolved. Agents change both sides of that problem. They are now the primary consumers of enterprise context, and they can also produce it by learning from the work teams already do. The result is a new kind of agentic ontology, built by agents, used by agents and governed by the people who know the business. We believe every organization will need a single agentic ontology that connects to all of its data, learns continuously from how the business operates and gives every agent the same governed understanding, whether it runs on Databricks or elsewhere. Building context, agent by agent, produces separate and partial views of the business, and those views inevitably conflict. An enterprise ontology should therefore be shared infrastructure that IT and data leaders own in the same way they own the data platform beneath it. At Data + AI Summit, we introduced Genie Ontology https://www.databricks.com/blog/introducing-genie-one-genie-ontology-and-genie-agents as that layer. Today, we're announcing its General Availability, with new sources for agents to learn from and new tools for people to curate and govern what they learn. We're also sharing results from a benchmark developed with Turing that show how much more accurately agents work when they draw on it. Genie Ontology is agentic because agents continuously build its knowledge for other agents to use, while people govern the business definitions that require precision. It combines learned and modeled knowledge, ranks it by authority, and delivers the most trusted context to agents while respecting existing user permissions. Most of Genie Ontology's knowledge is automatically learned from the systems and artifacts that reflect how an organization operates. Agents extract business context from the assets teams already work with, including Databricks dashboards, SQL queries, notebooks and semantic models in tools like Power BI. This knowledge is captured as Ontology Snippets, reusable pieces of business context such as calculations, business rules, and data relationships, which are continuously updated as the underlying sources evolve. By learning directly from everyday work, Genie Ontology grows more useful over time and captures the long tail of team-specific knowledge that would be impractical to model and maintain manually. Learned knowledge is complemented by modeled knowledge , which is the business meaning teams define explicitly and govern through Unity Catalog semantics https://docs.databricks.com/en/uc-semantics/ . This includes Metric Views https://docs.databricks.com/en/uc-semantics/metric-views/ that hold certified KPIs, Domains https://docs.databricks.com/en/uc-semantics/domains that organize knowledge by business area and Pages https://docs.databricks.com/en/uc-semantics/pages that record authoritative definitions of business concepts. Organizations model the comparatively small set of definitions that must be exact, such as recognized revenue or ARR, and Genie Ontology treats those definitions as the source of truth wherever they apply. When an agent asks a question, OntoRank, Genie Ontology's authority-aware ranking algorithm, evaluates learned and modeled knowledge to identify the most trustworthy context. It weighs signals such as source authority, certification, the expertise and relevance of the people behind the knowledge, popularity, and recency to resolve conflicting definitions and select the best context for each user and question. Genie Ontology then delivers that context directly to the agent, which can reason from the correct definitions immediately rather than spending time and tokens on open-ended exploration of the data. Through Genie MCP https://docs.databricks.com/en/agents/mcp-tools/genie-mcp , the same context is available to agents outside Databricks, such as ChatGPT, Claude and Microsoft Copilot, so the agents teams already use become smarter about the business without being rebuilt. The clearest test of an agentic ontology is whether agents that draw on it outperform those that must find context on their own. Existing data agent benchmarks focus mostly on tables, and frontier agents now score similarly on them, even though real enterprise questions often depend on business context that is fragmented across notebooks, dashboards and documents. To better approximate the real world, we worked with Turing https://www.turing.com/ to build a new benchmark simulating a live, realistic Databricks workspace for a fictional UK energy supplier. It compares Genie, powered by Genie Ontology, with two leading frontier agents, labeled Agent A and Agent B. Every agent answered the same questions five times per task, and each answer was graded against a rubric defined before the runs began. | Agent | Pass Rate | Avg wall time per run | Avg tokens per run | | Genie | 91% | 100 seconds | 432K | | Agent A | 45% | 435 seconds | 889K | | Agent B | 58% | 307 seconds | 1.05M | Genie answered correctly in 91% of runs, 1.6 times the rate of the stronger frontier agent, while using less than half the tokens and finishing three to four times faster. Because all three agents write SQL well, the difference comes from context. Genie drew on knowledge the Ontology had already learned from the workspace and went directly to the business rules each question required, while the frontier agents had to rediscover that knowledge on every run and frequently returned answers that failed to meet those rules. The results reflect an established deployment where teams have accumulated that knowledge, and the benchmark research blog http://www.databricks.com/blog/genie-genie-ontology-delivers-16-accuracy-less-half-tokens covers the methodology and task walkthroughs. Your account team can help you run the same evaluation in your own workspace. An agentic ontology can learn only from data its agents can reach, so a context layer for the whole enterprise depends on connecting to a wide variety of sources. Structured data either enters Unity Catalog via Lakeflow Connect https://www.databricks.com/product/data-engineering/lakeflow-connect or is federated in place, so Genie can compute answers directly from live operational data. Much of what a business knows, however, lives outside the lakehouse in the documents, tickets and wikis where teams record decisions and explain how things work. With this release, we are significantly expanding the systems the Ontology connects to. Genie Search, now Generally Available , lets people and agents find and use enterprise knowledge in documents, tickets and wikis alongside their governed Databricks assets, within the permissions they already have. Users can find documents anywhere in the enterprise or ask questions that span many of them at once. Genie returns synthesized, cited answers alongside the dashboards, tables and Genie Agents they already use. Because indexed content also informs OntoRank, the knowledge in those documents and tickets becomes part of the evidence Genie Ontology weighs when deciding which sources to trust. To configure Genie Search http://docs.databricks.com/genie/genie-ontology/search-index , through the admin console, an administrator connects each source and enables indexing. Genie can then search the full index in a single pass rather than querying each source separately, making answers more complete and enabling agents to find the right information with fewer tool calls and tokens. For real-time work, such as checking a Jira ticket status or creating a new one, Genie Search works alongside MCP connectors. Genie decides for each task how much to draw on the index, MCP or SQL. Genie Search supports Google Drive, Jira, Confluence and Workday today, with SharePoint in Beta and many more connectors coming soon. Recreating and maintaining source permissions is one of the most difficult parts of deploying enterprise search. Genie Search ingests each source's access controls alongside its content through Unity Catalog connections and enforces them before returning any result, so people and agents see only what they could already open in the original application. Because Genie Search runs as a managed service on Databricks, organizations do not need to maintain a separate permission model or procure separate search infrastructure. Genie Search is available today in Genie and for asset search on the Discover page https://docs.databricks.com/en/discover/discover-page . Learned knowledge is the largest part of Genie Ontology, and with this release it extends beyond the workspace into external BI tools and the organizational relationships that determine whose knowledge is most relevant to each person. Ontology Snippets https://docs.databricks.com/genie/genie-ontology how-do-i-see-or-edit-the-snippets-genie-learned , now Generally Available , are the units that make up Genie Ontology's learned knowledge. Genie builds them automatically and continuously from the dashboards, SQL queries and notebooks people already use and refreshes them as that work changes, so the Ontology keeps pace with the business without anyone maintaining it by hand. Each snippet stays tied to the asset it came from and inherits that asset's permissions, which means Genie draws only on context the user is already allowed to see. When Genie cites a snippet, users can see where the definition came from, who authored it and why it was trusted. Much of an organization's business logic already lives in the semantic models of BI tools such as Power BI, Tableau and Sigma, and rebuilding it for every AI initiative duplicates work and creates context drift. External knowledge sources, now in Beta, bring that logic into Genie Ontology by connecting to external BI platforms, identifying semantic models, measures, calculated columns and relationships, and generating Ontology Snippets from them. Power BI is supported today. Tableau and Sigma support is coming soon, and lineage support will follow. To get started, review the product documentation http://docs.databricks.com/genie/genie-ontology/external-knowledge-sources . Business knowledge is only half the picture. To make it useful, agents also need to understand the people behind it and how they relate to the person asking. Organizational directories http://docs.databricks.com/genie/genie-ontology/connect-people-source , now Generally Available , bring this context into Genie Ontology. Starting with Workday, they import people, teams and reporting relationships, making them accessible through a people browser in Genie. Genie Ontology combines these formal organizational relationships with activity signals that reveal who actually works together through the assets they create, share and use. OntoRank uses both to give greater weight to knowledge from teammates and close collaborators, helping agents deliver answers that are more relevant to each user. Genie Ontology continuously learns and improves through usage, but the quality of an agent's answers still depends on the accuracy of the knowledge it relies on. That's why organizations need the ability to review, refine, and govern their ontology. Databricks provides these capabilities within the same platform that governs the underlying data, ensuring agents operate on trusted business context. Governance builds on the controls organizations already use across Databricks. Unity Catalog https://www.databricks.com/product/unity-catalog governs the underlying knowledge, managing permissions, lineage, and definitions to ensure users and agents access only what they're authorized to see. Unity Gateway https://www.databricks.com/product/artificial-intelligence/unity-gateway governs how agents interact with that knowledge, controlling connections, usage limits, guardrails, and auditing. Together, they provide consistent governance as Genie Ontology extends to more agents and systems, without introducing a separate control plane. Domains and subdomains https://docs.databricks.com/en/uc-semantics/domains , now Generally Available , give Genie Ontology its organizational structure. They group data and AI assets by their business purpose, so when the same term means different things to different teams, Genie can apply the definition that best fits the question based on its associated domain. Domains also enable people to browse and find data on the Discover page https://docs.databricks.com/en/discover/discover-page , based on how their business is organized. Curation and management can be delegated at the domain or subdomain level, so Finance and Marketing can each maintain their own definitions and assets without relying on a central team. “Making data easier to discover and understand is an important part of how we're evolving our analytics capabilities across the Airline. Databricks Domains gives us a clearer, more intuitive way to organize our data, helping our teams navigate an increasingly complex data landscape with greater confidence.”- Tom Barber, Head of BI & Advanced Analytics, Virgin Atlantic “Domains and subdomains make it simple for business users to locate and gain access to data assets needed in their business role."- Michael Poremba, Senior Manager, Data Architecture, McAfee Unity Catalog Pages https://docs.databricks.com/en/uc-semantics/pages , now Generally Available , provide your authoritative definitions to Genie Ontology. Each Page is a governed explanation of a business term, entity or acronym, with a named owner and a home in a domain or subdomain. When a question involves a defined concept, Genie treats the Page as the source of truth, preferring it over learned context, and cites it so users can verify the answer. Agents can help create Pages as well as consume them. Teams can use the "Bulk import Pages" prompt in Genie Code to draft Pages from MCP-connected tools such as Confluence, Slack, Google Docs and GitHub, or use the public API to sync with existing knowledge platforms. Pages link directly to Unity Catalog assets, dashboards and other Pages, and each one remains under the control of its named owner. "What used to take me two weeks at a previous job, I can now do in five minutes. With Unity Catalog Pages, Genie understands how Agrifirm defines active and inactive customers and calculates inactivity windows, making complex historical analyses faster and more transparent. By showing the methodology behind the numbers, Genie helps us build trust in its answers upfront." - Henri van Dijk, Data Governance and Quality Assurance Officer, Agrifirm "With Unity Catalog Pages, we can give our Genie Agents a trusted source of business knowledge that is prioritized by Ontology and results in much better answers to very specific questions. And because we're using Pages to define industry terms and acronyms, each agent's instructions can be much more streamlined, resulting in better working memory - and again - better answers."- Ian Smith, Databricks Platform Manager, Waste Connections As agents add to the Ontology, the people responsible for it need a way to see what it knows, find where it falls short and curate the knowledge that matters most. Ontology Manager, coming soon in Beta, will be the first visual interface for this work. Data stewards and context engineers will assess the overall health of the Ontology, identify the most significant gaps, set the top-level domain structure and configure external knowledge sources. Domain experts will organize and curate knowledge within their business area, and business users will be able to see the context Genie draws on within their existing permissions. Ontology Manager will also surface high-impact opportunities, such as domains where coverage is thin or frequently asked concepts that lack governed definitions, and stewards will be able to promote what agents have learned into authoritative Pages directly from a review queue. An interactive graph will let users explore the entities and relationships within Genie Ontology and create new knowledge, while a hierarchical view will support day-to-day curation within a specific domain. To learn more about Ontology Manager, watch the product demo https://www.databricks.com/resources/demos/videos/genie-ontology . "We value that Genie Ontology makes the context behind answers more transparent and inspectable; citations let people trace answers to the underlying sources and business definitions, which helps build trust and improve the experience for our stakeholders.”- Jehan Saga, Manager of Data Engineering, Superhuman "Genie Ontology is exactly what we need to scale AI without losing the business context that makes it useful. It's a force multiplier for our data teams." - Noah Cepela, SVP, Analytics and Insights, Mode Global "Databricks Ontology with Genie Code has been a game changer for us. We're using it to build dashboards, apps, AI agents, and more. Power BI reports that used to take days to create can now be built in hours with Genie Code, and our analysts are so excited by the experience that they're already looking to move more of their reporting to Databricks Dashboards. The impact goes well beyond reporting: analysts who previously worked primarily in Excel and SQL are now using Genie Code to build Databricks Apps that support our operational teams. By putting that capability directly in the hands of analysts, we've been able to deliver solutions much faster and save our operational teams over 1000 manhours every month."- Kyle Burke, VP, Data Platform, BHG Financial “Business semantics have made our Genie experiences more accurate and context-aware across Genie, Agent and Code. Instead of embedding business knowledge into individual agents through skills and instructions, we can establish it once and make it available across solutions and domains.”- Kristoffer Absalonsen, Enterprise Architect, Odfjell Technology “At Zepto, we want AI to understand how our business works, not just the data it processes. Genie Ontology helps us do this by bringing business definitions and context together. For instance, metrics like order fulfillment time, delivery TAT and fill rate are meaningful only when understood in the right business context. This helps AI go beyond reporting metrics to interpreting them and turning business data into actionable insights.”- Deepak Siroya, VP of Analytics, Zepto Genie Ontology is Generally Available today . Organizations can start with the context they already have, governing the definitions that matter most in Unity Catalog and letting Genie Ontology learn the rest from the work their teams do every day. To learn more: This is one of many announcements during our Week of Genie . Over the coming days, we will share what is new across Genie, Genie Agents and Genie MCP, all of which are built on the agentic ontology introduced here. We can't wait to see how Genie Ontology helps you deploy reliable AI agents across your entire enterprise. Subscribe to our blog and get the latest posts delivered to your inbox.