AI/BI Genie is now Generally Available

Last June, we announced Databricks AI/BI, our entry into the Business Intelligence category, built around AI that deeply understands your data, semantics and usage patterns, and architected from the ground up to leverage the Data + AI Platform.  AI/BI not only offers a conventional BI experience through Dashboards, it also features Genie, a conversational analytics experience that enables everyone to answer their own data questions through natural language.  Today, we’re thrilled to share that AI/BI Genie will now join AI/BI Dashboards as Generally Available on all clouds.

What is Genie?

At its core, AI/BI Genie allows users to ask questions in natural language and get instant insights from data. Business users can ask questions like “How is my sales pipeline?” and Genie will answer in a combination of text summaries, tabular data and visualizations.  With every response, Genie also provides an explanation of how it arrived at the answer to give users confidence in that output.

Our goal with Genie is to deliver a conversational analytics experience that actually performs as advertised in the real world—trustworthy responses that are accurate and relevant within your organization's unique context. This is a challenge even for the latest and greatest LLMs, as they don’t understand the unique context or semantics specific to each organization’s data and business.

AI/BI Genie solves this problem by enabling analysts to package data (such as tables and views) along with semantics (like metric definitions, sample queries, text instructions, and certified assets) into a Genie space focused on a specific topic—for example, the sales pipeline.

The Genie space acts as a local knowledge store, marrying semantics extracted from existing assets on the Data + AI Platform with curation and feedback provided by expert analysts.  In essence, this system enables data analysts to scale their impact by creating a knowledgeable agent that can answer data questions just as they would. It empowers business users to self-serve questions that aren’t covered by existing dashboards and reports, freeing analysts to focus on higher-value work.

What updates has Genie received since its June announcement?

Since we announced AI/BI Genie last June, we’ve been working hard to introduce more capabilities, trustworthy answer quality and refined user experiences based on the feedback from our preview customers.  Here are a few highlights:

• Best-in-class accuracy:   Response quality is the top priority for the Genie product team, and our focus has been on addressing real-world issues surfaced by thousands of active customers. For example, we added the concept of value dictionaries, which—as shown in the example below—allows Genie to semantically match a request for “first level group” to the literal value “1st line support group” based on what is in the underlying data. Genie is also now able to handle much larger datasets (zero column count restrictions), complex instructions, and has gone through multiple cycles of model updates. 

• Unity Catalog Metric Views: Genie is built natively on the Data + AI Platform, and we continue to expand on that foundation. For example, we’ve added support for Unity Catalog Metrics, allowing customers to use governed, trusted data models to ensure accuracy for critical use cases. With AI/BI Genie, Unity Catalog serves as the single source of truth for both data and semantics—eliminating the need for redundant modeling.
• Support for evaluation and benchmarks: Skilled engineers will tell you that any AI effort should start with an evaluation phase.  Failure to do so means failure in production. Conversational analytics experiences are no different - when you create a new Genie space, you are really creating a new AI system for a specific domain. This is why we’ve added features such as Benchmarks and Ask for Review. Genie Benchmarks are curated test questions with expected SQL answers that allow authors to systematically evaluate and improve the accuracy of Genie’s responses over time, while ‘Ask for Review’ enables end-users to flag specific responses for verification or correction by space admins, fostering trust and collaboration between business users and data teams.

• Ad-hoc analysis with Genie File Upload: One common analysis pattern that we discovered during the preview period was the need to blend a local spreadsheet (like a list of leads from a campaign) with a managed dataset in Unity Catalog (such as Salesforce accounts and contacts) to complete an ad-hoc task.  With Genie File Upload, you can simply drag and drop Excel or CSV files into Genie and complete any comprehensive analysis through natural language.

• Genie everywhere: Dashboards and Genie work better together—which is why every AI/BI Dashboard includes an integrated Genie space. Dashboards show what data is available for exploration, while Genie lets users go beyond the limits of predefined slices, filters, and combinations—all within the same experience. Additionally, with the new Genie Conversation APIs, developers can similarly embed Genie into collaboration tools like Slack, Microsoft Teams, or custom internal applications. This opens up new ways for business users to access conversational analytics in the flow of work — without switching contexts or logging into a separate interface.

We have also delivered a range of other enhancements—from improved visualizations to a redesigned user interface to guide end users through their analysis.  We plan to continue this high level of feature velocity post-GA; please check out the AI/BI release notes and stay tuned to for more blogs as we continue to add features over time.

Over 4,000 customers adopted AI/BI Genie during its preview

We’ve been thrilled by the rapid adoption of Genie; over 4000 customers adopted AI/BI Genie during its preview to democratize data and to empower their non-technical users.  Here’s a sample of what some of our preview customers have shared with us about Genie:

Genie is changing how we empower teams at HP. With its intuitive natural language interface, we’re enabling users to instantly access critical data, no coding required, making data-driven insights faster and more efficiently.—Bruce Hillsberg, VP, Data Engineering & Insights, Technology & Innovation, HP Inc

Users were constantly pinging analysts to ask- Hey, what did sales look like for the southwest region compared with the previous year?- The idea of being able to just ask Genie, rather than hunt for the right analyst and hope they get the answer right, has been very exciting for the business.—Shahmeer Mirza, Senior Director of Data, AI/ML, and R&D, 7-Eleven

Executives and other LOB leaders in marketing and finance can use Genie to access the data they need just using plain English. Absolutely incredible.—Felix Baker, Head of Data Services, SEGA Europe

We have many other customers who have seen tremendous success with Genie including Premier Inc, Magic Orange, The AA, and Casas Bahia to name a few.  We’re excited about enabling more organizations to join them by making Genie generally available. 

What's new in Genie's next-generation knowledge store?

The heart of Genie is its knowledge store—a living semantic model that Genie’s AI system consults each time it answers a question. We are now releasing the next generation of our knowledge store, which allows authors to curate knowledge at much finer granularity with much less effort. Authors can add knowledge directly to data assets, including column‑level synonyms and sampled values, as well as table‑level context such as primary/foreign‑key joins or certified metrics and more. Genie’s compound AI system can retrieve the right pieces of knowledge for each question to provide accurate answers to users’ questions. 

Even with these enhancements, curating semantics is still a chore. To make this process significantly easier, Genie now performs knowledge mining and extraction. 

Knowledge mining (above) allows Genie to sift through Unity Catalog’s lineage and the query history, while preserving governance guarantees of Unity Catalog, to suggest new instructions or highlight recurring patterns that merit promotion to the knowledge store, slashing manual curation for space owners. 

We also have many customers leveraging Genie for net-new use cases where there is no prior reference to bootstrap the knowledge store.  To handle these cases, we’ve enhanced Genie’s knowledge extraction so that curation occurs as a part of regular usage. During every chat session, Genie extracts semantic information, proposes concise knowledge snippets, and—once the user approves—commits them to the store. 

Paired with the built‑in activity monitoring and feedback loops, day‑to‑day usage itself becomes a continuous‑learning pipeline that keeps Genie’s semantic model accurate without periodic overhauls.

What is Genie Deep Research?

While the GA of Genie is a significant milestone, we are not slowing down on our journey to reimagine BI with an AI-first approach.   We’re excited to announce the preview of Genie Deep Research – a new mode in Genie designed to address more complex, multi-step questions. Until now, Genie has been optimized for “what happened” questions (e.g., “What is our sales pipeline?”). Answering “why” and “how” questions (e.g., “Why are sales down?” or “How can we increase our pipeline?”) required users to form their own hypotheses and ask Genie to test each one individually.  With Deep Research, Genie will leverage the latest advances in LLM research to tackle these types of questions by producing a research plan and analyzing multiple hypotheses in parallel before summarizing the results.

As with the rest of Genie, Deep Research will build upon and extend Genie’s knowledge store to deliver accurate and trustworthy results. Genie will also explicitly provide citations for the conclusions it drew and provide easy one-click experiences for scrutinizing any work done along the way.   Deep Research will be made available as an experimental feature to Genie customers later this summer. 

How does Databricks One give business teams access to Genie?

One of the few concerns we’ve heard from customers about rolling out Genie broadly is that accessing AI/BI has to date required navigating the technical Databricks workspace—an experience that can feel intimidating for non-technical business users. As part of our Data and AI Summit announcements, we’re also introducing Databricks One — a new experience that gives business teams simple and safe access to AI/BI Dashboards and Genie spaces, as well as Databricks Apps. Just like AI/BI, Databricks One is freely available to all Databricks SQL customers.  

Where can you learn more about AI/BI Genie?

AI/BI Genie and Dashboards are now available to Databricks SQL customers with no additional licenses required.  If you want to learn more, you can check out the AI/BI Genie web page and get started by exploring the following content:

1. Documentation: Dive deeper into the details with our documentation.
2. Demos: Watch our demo videos, take product tours and get hands-on tutorials to see these AI/BI in action.
3. Training: Get started with free product training through Databricks Academy
4. eBook: Download the Business Intelligence Meets AI eBook

If you’re not an existing Databricks customer, you can also test-drive Genie and other Databricks products by signing up for a free trial.

---

Frequently Asked Questions

What exactly is a Genie space, and how do you set one up?

A Genie space is a curated container where analysts package specific data assets—like tables and views—together with semantics such as metric definitions, sample queries, text instructions, and certified assets, all focused on a particular topic (for example, the sales pipeline). It functions as a local knowledge store that blends semantics pulled from the Data + AI Platform with curation and feedback from expert analysts. Setting one up means selecting the relevant data and layering in these instructions so Genie can answer questions in that domain as knowledgeably as the analyst who built it.

What are value dictionaries in Genie, and how do they improve accuracy?

Value dictionaries let Genie semantically match the wording in a user's question to the actual literal values stored in the underlying data. For instance, a user asking about the "first level group" can be matched to the literal value "1st line support group" even though the phrasing differs. This mapping is one of the accuracy improvements Databricks added based on real-world issues surfaced by active customers during the Genie preview.

What's the difference between Genie Benchmarks and Ask for Review?

Genie Benchmarks are curated test questions paired with expected SQL answers that let space authors systematically evaluate and improve Genie's accuracy over time, while Ask for Review is a separate feature that lets end-users flag specific responses for verification or correction by space admins. Together they were introduced because creating a Genie space is effectively creating a new AI system for a domain, and skilled teams treat evaluation as a required starting point. Ask for Review specifically fosters trust and collaboration between business users and the data teams who maintain a space.

Do you need an additional license to use AI/BI Genie?

No—AI/BI Genie and AI/BI Dashboards are available to Databricks SQL customers with no additional licenses required. This applies to the generally available Genie experience as well as the integrated Genie spaces embedded in every AI/BI Dashboard. Databricks One, the new business-team access layer for Genie and Dashboards, is likewise freely available to all Databricks SQL customers.

Is there a training course for learning how to build a Genie space?

Yes, Databricks offers a course called "Building Reliable Conversational Agents with Genie" that teaches you to design, build, and maintain a Genie Space from end to end. It covers understanding source data, defining benchmarks, and configuring the full set of Knowledge Store curation tools—including metadata, synonyms, prompt matching, SQL logic, example queries, and text instructions—as well as sharing spaces and using monitoring and feedback to improve quality over time. The course expects general familiarity with the Databricks workspace, Unity Catalog concepts, and intermediate SQL proficiency, plus access to a workspace with the Databricks SQL entitlement.