Databricks Genie One

What it is

This page covers Databricks’ business-user agent layer, not the lakehouse: Genie One and the Genie Agents under it. Genie One is the chat front door — on the web, mobile, Slack and Microsoft Teams — where a sales, finance or CS user asks a question in plain language and gets an answer with a chart, the SQL behind it and links to the cited sources. The product was called Genie until it was renamed on June 9, 2026; chat in Genie One went generally available on June 15, 2026. Genie Agents, renamed from Genie Spaces on July 9, are scoped agents an analyst curates for one job — say, pipeline and bookings — with up to 50 Unity Catalog tables, views or metric views, example SQL and written instructions. Underneath sits Genie Ontology, a business-context map Databricks infers from your dashboards, notebooks, queries and pipelines; it is on by default but still in public preview. On September 24, 2026 Databricks bought Row Zero, a Seattle cloud-spreadsheet startup, to give Genie a governed spreadsheet grid. No ship date has been announced.

Why it shows up in RevOps and CS stacks

• One identity model for every answer. A Genie Agent runs SQL on a warehouse using the author’s embedded compute credentials, but data access is always checked as the end user. Row filters and column masks in Unity Catalog apply per person, and every query is attributed to the asker in query history. A CSM sees only their book; a regional VP sees only their region.
• CRM data lands without a third-party pipeline. Lakeflow Connect ships managed Salesforce and HubSpot connectors that write straight into Unity Catalog, so a RevOps team can join opportunities, billing and product usage without buying Fivetran just for the CRM feed.
• It reaches the agents your team already uses. The Genie One MCP server went GA on September 22, 2026. Claude, ChatGPT and Cursor can ask Genie questions under OAuth, and clients that support MCP Apps render Genie’s charts inline. Going the other way, Databricks-managed connectors for Google Drive, Gmail, Microsoft 365, Atlassian, Slack and GitHub went GA on September 10, each authenticated per user.

Pricing reality

There is no seat fee, and today the Genie layer itself is close to free. Per Databricks’ billing docs, Genie One and Genie Agents usage by human users costs nothing through January 31, 2027. Service principals — any automated or API-driven caller — have been billed at standard pricing since September 24, 2026, with no free allowance. Genie Code, the data-team agent, gives each user 150 free DBUs a month and bills overage under the Serverless Real-Time Inference SKU, at a 25% promotional discount during the same window. Databricks has not published what Genie One will cost after January 2027.

What you pay now is the warehouse. Genie Agents need a pro or serverless SQL warehouse, and third-party price guides put serverless SQL at about $0.70 per DBU on AWS Premium list. Our estimate: a 2X-Small serverless warehouse at 4 DBUs an hour, running 10 hours a day for 22 working days, costs about $620 a month at list; a Small at 12 DBUs an hour is about $1,850. If you already run Databricks, budget $7,500–$25,000 a year in added warehouse compute for a 20–100 person ops rollout, then set a reserve line for whatever Genie One pricing becomes. If you do not run Databricks, this is a platform decision, not a tool purchase.

Best for

RevOps and CS ops leaders at companies whose data team already runs a Databricks lakehouse with Unity Catalog, who want 20–200 business users answering their own pipeline, renewal and usage questions — in Slack, Teams or a browser — without buying a BI seat for each of them.

Skip it if your warehouse is Snowflake or BigQuery: moving the data to get the agent is rarely worth it. Skip it if users need to edit numbers — forecast submissions and quota overrides need a writeback tool. And skip it if nobody on the data team will own the curated agents; uncurated Genie produces plausible wrong answers.

Versus the alternatives

Microsoft Power BI Copilot is the incumbent by installed base, available on any paid Fabric capacity from F2 upward. Pick it when the company lives in Microsoft 365 and dashboards are the main output. Pick Genie One when the data already sits in Databricks and you want answers governed by Unity Catalog rather than a separate semantic model.

Snowflake CoWork is the direct warehouse-native rival, billed in AI Credits per token. Pick it if your warehouse is Snowflake; the two are not interchangeable.

Hex Threads is the fastest-moving entrant, with seat pricing — Team is $75 per editor per month — and AI answers that escalate into a notebook an analyst can audit. Pick Hex when the data team wants to review what the AI says before business users rely on it.

Sigma wins when the job is writeback apps — forecast submission, territory plans — on the same Databricks tables. To push Genie-identified segments into the CRM on a schedule, pair Genie with Hightouch.

If none fit because sales and finance disagree on what “active account” means, fix the model first — see data warehouse vs CDP.

Watch-outs

• The free period ends January 31, 2027, and the next price is unpublished. Guard: query system.billing.usage for sku_name = 'GENIE_FREE_USAGE' monthly to see what usage would cost at the Genie Code rate, and get post-promotion pricing written into your renewal before the rollout becomes daily habit.
• The beta MCP endpoint sunsets on October 31, 2026. Integrations still calling /api/2.0/mcp/genie stop working. Guard: move them to /ai-gateway/mcp-services/system.ai.genie_one_mcp and add the ai-gateway OAuth scope; an authorization granted only the old genie scope will not reach the new service.
• Automations bill from the first call. Slack bots, scheduled jobs or Claude agents running under a service principal get no free allowance. Guard: run interactive use on-behalf-of-user, set a Genie budget, and tag service-principal workloads so they show up separately in billing.
• Ontology is inferred, not agreed. It learns business meaning from how people have queried the data, including their mistakes. Guard: define your 20–30 core metrics as metric views, attach only those to each agent, and track accuracy with a fixed set of benchmark questions every time instructions change.
• Scheduled tasks run at most daily. Since June 25, 2026, hourly schedules are not supported. Guard: keep intraday alerts — a deal over $100,000 slipping a stage, say — in a warehouse job or Hightouch sync, not a Genie schedule.