Databricks Genie
Databricks · Databricks Genie: Governed Data Agent That Answers Business Questions and Runs Multi-Step Analysis on Your Lakehouse
Databricks Genie is the natural-language data agent built into Databricks: business users ask questions in plain English and Genie writes and runs read-only SQL against Unity Catalog data under their own permissions, returning answers, charts, cited reports, documents and scheduled briefings. It suits companies whose data already lives in Databricks and that want analysts to curate trusted Genie Agents for finance, sales, marketing and operations teams.
Best for
Companies that already run their data on Databricks with Unity Catalog and want business teams to ask questions, run deeper analysis and get scheduled briefings on governed data, with analysts curating trusted Genie Agents and per-user row and column security enforced automatically
Not ideal for
Teams whose data is not in Databricks, small teams without anyone to curate tables and definitions, buyers who need predictable flat pricing after the free promotion ends in January 2027, and anyone expecting the always-on autonomous Genie to be available today
Who it's for
Business users in finance, sales, marketing and operations, and the data analysts who curate Genie Agents for them, at companies whose data runs on Databricks
Capabilities
- Natural-language questions converted into read-only SQL that runs on a pro or serverless SQL warehouse, returning the query, a results table and visualizations
- Genie Agents (formerly Genie Spaces): analyst-curated domain workspaces built from up to 50 Unity Catalog tables, views or metric views, with instructions, example SQL, synonyms, join relationships and trusted parameterized queries and functions
- Agent mode for complex questions: plans the research, runs multiple SQL queries, iterates on the results and returns a report with citations, visualizations and supporting tables; Flash (Beta) is a faster variant for simple questions
- Benchmarks for measuring a Genie Agent's accuracy on test questions, plus Inspect (Public Preview), which checks and rewrites generated SQL for filters, date ranges and joins
- Genie One chat that routes each question to a relevant Genie Agent or searches the dashboards, queries, metric views and tables you can access, and cites the sources it used
- Scheduled tasks created in plain language that run on a schedule, post results to a chat thread and email them with visualizations and a PDF
- MCP connectors run by Databricks for Google Drive, Gmail, Google Calendar, Microsoft 365, Jira and Confluence, Asana, Slack and GitHub, used with each user's own OAuth credentials to look things up or take actions such as drafting email or updating a Jira issue
- Genie Ontology: a shared context layer combining modeled knowledge (metric views, domains, Pages, certification) with knowledge Genie learns from dashboards, SQL queries, Genie Agents and dbt runs
- Per-user governance: data access is checked against the asking user's Unity Catalog privileges, row filters and column masks, and each query is attributed to that user
- Documents, slide decks and other artifacts (Beta) drafted from a conversation, plus reusable skills that can be shared with a team (Beta)
- Genie Agents API with Chat mode, Agent mode and Management endpoints, a Genie One API, embedding in external apps, and the Genie One MCP server for any MCP client
- Access from the web, iOS and Android apps, a macOS and Windows desktop app (Beta), the Databricks Genie apps for Slack and Microsoft Teams (which also connect Genie to Microsoft 365 Copilot), and Genie plugins for ChatGPT, Claude and Grok Bot
Limitations
- Requires a Databricks account with data registered in Unity Catalog; Genie Agents need a pro or serverless SQL warehouse, and chatting in Genie One needs CAN USE on at least one SQL warehouse
- Genie One and Genie Agents usage by users is free only until January 31, 2027; service principals are billed, and SQL warehouse or serverless compute is billed separately at standard Databricks rates
- Answer quality depends on curation: Databricks says well-annotated tables, instructions and example queries are key, and a Genie Agent only queries the sources attached to it
- Chat mode handles structured data only; files in Unity Catalog volumes (Public Preview) and MCP connectors in Genie Agents (Beta) work only in Agent mode
- The always-on autonomous Genie, Genie Search and Ontology Manager announced on November 11, 2026 are listed as coming soon to preview and beta programs, not generally available
- Genie Agents become unavailable if an account admin disables partner-powered AI features, which route requests to models from Azure OpenAI, OpenAI or Anthropic hosted on Databricks
- Prompts are wrapped in English, so answers to questions asked in other languages can come back in English
- In Slack, a response in a channel with public message visibility is shown to everyone there, including members without access to the underlying data
- Account-level Genie One may process Databricks-generated metadata in the US; teams with strict residency rules can disable it
Use cases
- Letting a sales manager ask which products sold best by region last quarter and get the SQL, a results table and a chart without filing a request with the data team
- Asking Agent mode why gross margin dropped in one region and getting a cited report built from several queries
- Scheduling a Monday-morning briefing of new customer reviews or pipeline changes that arrives by email with a PDF
- Giving a finance team a curated Genie Agent with certified revenue definitions so everyone's numbers match
- Asking governed data questions from ChatGPT or Claude through the Databricks Genie plugin, with answers linking back to the Databricks sources
- Embedding a Genie Agent in an internal portal or app, as an iframe or through the Genie Agents API, with each user's Unity Catalog permissions still applied
Our take
Genie is the obvious first thing to try if your company already runs on Databricks with Unity Catalog, because it adds no new data copy, no new permission model and, until January 2027, no extra LLM charge for business users. Where it is strongest is governance: read-only queries executed as the asking user, with row filters and column masks applied automatically, are exactly what security teams worry about with generic chat-with-your-data tools. Where it is weakest is the same place as every text-to-SQL product: accuracy depends on how well analysts annotate tables, write instructions and maintain trusted queries, and Databricks' own docs say as much. Agent mode and scheduled tasks make it more than a query box, but read the November 2026 announcement carefully: the always-on Genie that takes on long-running work is still on its way to preview and beta programs. Plan for compute costs, and for an unknown LLM price once the promotion ends, before rolling it out to thousands of users.
Who should use it
Databricks customers with data in Unity Catalog who want finance, sales, marketing and operations staff to self-serve answers, deeper analysis and scheduled briefings, and data teams willing to curate Genie Agents with trusted definitions and test them with benchmarks.
Who should skip it
Organizations whose data lives mainly in Snowflake, BigQuery or SaaS tools rather than Databricks, teams without analysts to curate metadata and definitions, and buyers who need the always-on autonomous agent or fixed per-seat pricing today.
Strengths
- Runs where the data already is, with every query read-only and checked against the asking user's Unity Catalog permissions, row filters and column masks
- Analysts control quality with instructions, example SQL, trusted queries and benchmark tests for each Genie Agent
- Agent mode goes beyond one-shot text-to-SQL, running several queries to explain why a metric moved and returning a cited report
- Reaches people where they work: web, mobile and desktop apps, Slack, Microsoft Teams, email, and plugins for ChatGPT, Claude and Grok Bot
- Genie One and Genie Agents are free for users until January 31, 2027, apart from the compute they use
- Documented APIs and an MCP server for embedding Genie in your own apps and agents
Weaknesses
- Only useful if your data is in Databricks with Unity Catalog
- Accuracy depends on curated metadata, instructions and example queries
- SQL warehouse compute is always billed, and LLM usage pricing after the January 2027 promotion is not yet published
- The always-on autonomous Genie announced in November 2026 is not yet available outside preview and beta programs
- Several useful pieces, including file analysis, MCP connectors in Genie Agents, Flash, artifacts and the desktop app, are still in Public Preview or Beta
Databricks Genie pricing
Genie One and Genie Agents
Free for users until January 31, 2027
- Promotional pricing for identified users
- Service principal usage is billed
- SQL warehouse or serverless compute billed separately
Genie Code
150 DBUs free per user, then pay-as-you-go
Billed monthly
- 150 DBUs is about $10.50 in US East; varies by region
- Usage beyond the allowance billed in DBUs, with a 25% promotional discount through January 31, 2027
- Account admins can set budgets per workspace, group or user
Compute
Standard Databricks SQL rates
- Genie Agents need a pro or serverless SQL warehouse
- Not covered by free Genie usage or Genie budgets
Note: Genie is part of the Databricks platform and is billed through Databricks usage rather than a separate seat license. Free Genie usage covers LLM usage only. Billed Genie usage appears under the Serverless Real-Time Inference SKU in billing system tables, and through January 31, 2027 a 25% promotional discount is applied to all billed Genie usage. Databricks has not published Genie One and Genie Agents pricing for after January 31, 2027, and says it may introduce charges for curating Genie Ontology snippets, which are currently free.
Technical specs
API pricing
Genie One and Genie Agents usage by users free until January 31, 2027; service principal usage billed in DBUs; SQL warehouse compute billed separately
Available models
Where Databricks Genie excels
Self-serve answers for a sales or marketing team
An analyst curates a Genie Agent with the relevant tables, metric definitions and example queries, and business users ask follow-up questions in Genie One, Slack or Teams instead of queuing requests, with each person seeing only the rows they are allowed to see.
Investigating why a KPI moved
Agent mode breaks the question into steps, runs several SQL queries, adjusts as results come in and returns a report with citations and charts that an analyst can review before it is shared.
Governed data inside ChatGPT or Claude
With the Databricks Genie plugin, employees who already work in ChatGPT or Claude can ask business questions there, and answers come from Genie Ontology and Unity Catalog under their own sign-in rather than from a model reading raw tables.
Databricks Genie vs. competitors
Databricks Genie vs. AgentSpot
AgentSpot is ThoughtSpot's hosted, no-code workspace where business teams build and share agents, scheduled workflows and data apps grounded in ThoughtSpot data models through Spotter, priced in per-outcome credits with a free plan. Genie is built into Databricks itself, works directly on Unity Catalog tables and metric views under each user's own permissions, and is included with the Databricks platform, with user access to Genie One and Genie Agents free until January 31, 2027 apart from compute.
Databricks Genie vs. Jedify
Jedify is a separate analytics platform that builds its own versioned context graph from warehouses (including Databricks), BI tools and dbt, and runs Data Agents on demand, on a schedule or when a threshold is crossed, sold on contracts starting at $1,500 a month on AWS Marketplace. Genie lives inside Databricks, uses Genie Ontology and analyst-curated Genie Agents for context, and only works on data in Databricks, while Jedify can span several warehouses.
Frequently asked questions
What is Databricks Genie?
Genie is Databricks' natural-language data agent. Business users ask questions in plain English in Genie One, Slack, Teams, the mobile apps or an AI assistant such as ChatGPT or Claude, and Genie writes and runs read-only SQL against Unity Catalog data they are allowed to see, returning answers, charts and cited reports. Analysts curate Genie Agents, domain workspaces with chosen tables, instructions and example queries, to make answers reliable. Genie is also the umbrella name for Genie One, Genie Agents and Genie Code, the coding assistant for data teams.
What is the difference between Genie One, Genie Agents and Genie Code?
Genie One (formerly Databricks One) is the simplified interface for business users, with chat, dashboards, apps, scheduled tasks and documents. Genie Agents (formerly Genie Spaces) are domain-specific workspaces that analysts configure with datasets, SQL examples and instructions; Genie One routes questions to them. Genie Code is the AI assistant for developers in notebooks, the SQL editor, pipelines and dashboards. All three share Genie Ontology, Databricks' business context layer.
How much does Databricks Genie cost?
Genie is part of the Databricks platform rather than a separate subscription. Until January 31, 2027, Genie One and Genie Agents usage by users is free; service principals are billed. Genie Code uses pay-as-you-go pricing with 150 free DBUs per user per month (about $10.50 in US East) and DBU billing beyond that, discounted 25% during the same promotional period. In every case, the SQL warehouse or serverless compute Genie uses to run queries is billed separately at normal Databricks rates. Databricks has not published what Genie One and Genie Agents will cost after the promotion.
Is the always-on autonomous Genie available now?
Not generally. Databricks' November 11, 2026 announcement says Genie and Genie Ontology are generally available to all customers, while Genie Search, Ontology Manager and the new always-on Genie, which handles long-running work, delegates to subagents and asks for decisions through an inbox and email, are coming soon to preview and beta programs. What is available today includes Genie One chat, Genie Agents, Agent mode and scheduled tasks.
Can I use Databricks Genie in ChatGPT or Claude?
Yes. Databricks offers Genie plugins for ChatGPT, Claude and Grok Bot that connect to the Genie One MCP server. An admin of the ChatGPT, Claude or Cursor workspace sets up the plugin first, then each user connects and signs in with their Databricks account, and answers respect that user's Unity Catalog permissions. Databricks' launch announcement also lists a plugin for Microsoft Copilot Cowork, Genie reaches Microsoft 365 Copilot through the Databricks Genie app in Microsoft Teams, and any other MCP client can connect to the Genie One MCP server directly. If your workspace restricts inbound network access, an admin must allow ChatGPT or Claude traffic.
Does Genie respect data permissions?
Yes. Genie runs read-only SQL, and Unity Catalog evaluates data access as the user asking the question, so row filters and column masks apply per person and queries are attributed to that user. For Genie Agents, the warehouse runs on compute credentials embedded by the agent's author, but data access is still checked against the end user's identity. External MCP connectors use each user's own OAuth credentials. One caveat: in Slack, a response posted with public visibility is visible to everyone in the channel.
Which clouds does Databricks Genie run on?
Genie is documented for Databricks on AWS, Azure Databricks and Databricks on Google Cloud. Some features have regional conditions: Agent mode availability follows Databricks Designated Services rules, and Flash runs in-Geo only in US and EU workspaces unless cross-Geo processing is enabled.
Databricks Genie vs ThoughtSpot AgentSpot: what is the difference?
Both let business users ask questions of governed data in plain language. Genie is built into Databricks and works directly on Unity Catalog tables and metric views, with access to it included in the Databricks platform. AgentSpot is ThoughtSpot's hosted workspace for building and sharing agents, scheduled workflows and data apps on ThoughtSpot data models, with its own free plan and per-outcome credit pricing. If your data and semantic models live in Databricks, Genie avoids adding another layer; if your company standardizes on ThoughtSpot, AgentSpot fits that stack.
Integrations & fit
Alternatives to consider
ThoughtSpot
Business teams in operations, finance, sales, HR and legal that want non-technical staff to build agents and scheduled workflows once and share them company-wide, especially organizations already modelling their data in ThoughtSpot who want answers grounded in governed metrics and row-level security, with usage billed per conversation or run rather than per seat
Freemium
Jedify
Data teams at mid-size and large companies with a cloud warehouse who are tired of rebuilding the same reports, and who want finance, RevOps, operations or supply chain staff to set up their own monitoring and root-cause agents on governed metric definitions with per-user row-level permissions
PaidAbout Databricks Genie
Genie is the name Databricks uses for a family of AI experiences. This page covers the data agent business users talk to: Genie One and the Genie Agents behind it. Genie Code, the coding assistant for data engineers, shares the name and context layer but is a separate developer tool. Genie One (formerly Databricks One) is a simplified interface where people browse dashboards and apps and chat with their data. A question goes to a relevant Genie Agent (formerly a Genie Space), a domain workspace an analyst builds from up to 50 Unity Catalog tables, views or metric views plus instructions, example SQL, trusted queries and benchmark questions; if none fits, Genie One searches the data assets you can access. Agent mode, formerly Research Agent, handles open-ended questions such as why a metric changed: it plans the research, runs several SQL queries, adjusts as results come in and returns a report with citations, charts and tables. Genie One also runs scheduled tasks that email results with a PDF, drafts documents, and can search or act in tools such as Gmail, Jira and Slack through MCP connectors under each user's own OAuth sign-in. Governance is the strongest argument for it: queries are read-only and run under the asking user's Unity Catalog permissions, so row filters and column masks apply per person, and answers draw on Genie Ontology, a context layer built from metric views, certified assets, dashboards, query history and curated agents. Genie is available on the web, in iOS and Android apps, in a desktop app (Beta), in Slack and Microsoft Teams, through REST APIs, and through Databricks Genie plugins for ChatGPT, Claude and Grok Bot built on the Genie One MCP server. The tradeoffs: it only makes sense if your data is in Databricks with Unity Catalog; answer quality depends on analysts curating agents and metadata; the always-on autonomous Genie announced on November 11, 2026 is still listed as coming to preview and beta programs; and Genie One and Genie Agents usage is free for users only until January 31, 2027, while SQL warehouse compute is billed at standard rates throughout.
Updates from Databricks Genie
Databricks announced a second-generation Genie designed to take on long-running work, delegate to subagents, track progress and reach out through a priority inbox and email when it needs a decision. Databricks said Genie and Genie Ontology are generally available to all customers, while Genie Search, Ontology Manager and the new always-on Genie are coming soon to preview and beta programs.
Databricks released Genie plugins for ChatGPT, Claude, Microsoft Copilot Cowork and Grok Bot. The ChatGPT, Claude and Grok Bot plugins connect to the generally available Genie One MCP server through default OAuth apps, so users sign in with their Databricks account. The older Beta MCP endpoint reaches end of life on November 30, 2026.
Genie Agents gained the ability, in Public Preview, to analyze PDFs, Word documents, slide decks and images stored in Unity Catalog volumes alongside structured tables, with citations that point to the source page. File analysis runs in Agent mode.
Databricks began charging for Genie Code usage beyond a per-user free monthly allowance of 150 DBUs, with account admins able to set budgets per workspace, group or user. Genie One and Genie Agents usage by users stays free under a promotion that runs until January 31, 2027.
Are you the founder? Claim this listing →