Databricks Genie 插件接入 ChatGPT、Claude、Microsoft Copilot Cowork 与 Grok Bot

Databricks：Blog·2026-11-11 23:00· 53 分钟前AI 评分49

Bring your business context to ChatGPT, Claude, Microsoft Copilot Cowork and Grok Bot

AI 导读

Databricks 推出 Genie 插件，让 ChatGPT、Claude、Microsoft Copilot Cowork 和 Grok Bot 中的智能体通过受治理接口访问企业业务上下文。

正文 · 原文

Every business is adding more data to their AI-powered workflows. But more data doesn't automatically mean better insights. Inevitably, it means more places for an agent to look and more chances to find the wrong answer. Ask a frontier agent a question about your business and it starts a live search. It hops between MCP servers, opens a document, reads a schema, tries a table, backs out and tries another. It has little context on where the right answer lives, so it may guess its way through a random walk across the organization's data.

What agents need is a map of the business before they start looking. That's why we announced Genie Ontology, the enterprise context layer, that combines governed definitions with continuously learned organizational knowledge giving AI agents the business understanding they need to deliver trusted insights.

Take your business context anywhere you work with the Databricks Genie plugins

Ontology hero

Today we are introducing new Databricks Genie plugins that give your team’s favorite agents a governed interface to access your enterprise context. These plugins work natively today in ChatGPT, Claude, Microsoft Copilot Cowork and Gok Bot so teams can maintain business context in Databricks and make it available across their preferred agents.

Frontier agents can now draw on Genie Ontology to supply the definitions, relationships, and source context needed to interpret the business. Genie works out what the business terms mean, finds the right data across tables and documents, writes and runs SQL and returns an answer with citations to its sources.

For questions requiring specialized analysis, Genie can route work to domain-specific Genie Agents. A question about pipeline attribution routes to the marketing Genie Agent. A question about operational costs routes to the finance Genie Agent. Genie also supports insights from unstructured data so teams can access insights from large volumes of unstructured data that reside inside their enterprise

For clients that support the MCP Apps extension, Genie returns interactive visualizations and Genie Ontology citations directly in the conversation. You see the charts, sources, and definitions without leaving your workflow.

How your agents work with the Genie plugin

Some requests are complex business related questions: why is activation tracking low? Which accounts are at risk and why? The agent infers that the question belongs to Genie and hands it over. Genie then runs its own loop:

It works out what the business terms mean using Genie Ontology.
It finds data the user is allowed to see and runs It writes and runs the SQL,
it routes to a domain Genie Agent if needed for further investigation.
It returns an answer with citations.

Other times the agent owns a bigger job, like changing codebase, or monitoring a new product release. Here the agent checks for the business context it's missing and then keeps going with its own tools: editing files, drafting a plan, posting to Slack.
Genie Ontology

Lets see a few examples of this in action:

ChatGPT: Dots keep product teams on track with real-time insights

Dots are always-on agents in ChatGPT with its own computer that keeps working between conversations. A product manager can give a dot a standing job: watch the launch of a new feature and flag anything that needs attention. Each morning the dot asks Genie how adoption is tracking, using the certified definition of an active user, and checks it against launch targets. When a segment slips, it pulls related support themes and posts a short summary to the team's Slack channel.

A product manager might ask their dot:

Track weekly adoption of the new feature against target, by plan tier
Engage accounts tried the feature once and never came back
Investigate drop in enterprise usage after last Tuesday's release

Claude: Deep reasoning that understands your business

Claude is built for long, agentic work across many sources. With the Genie plugin, Claude can check important facts before it draws any conclusions. Say a business analyst has to explain why quarterly revenue in a given region came in 10% below forecast. Claude can ask Genie for revenue, order volume and average deal size by product and channel, using the certified metric definitions finance relies on. It then has Genie search pricing memos, sales call notes and competitor reports for what changed during the quarter. From there, Claude goes through a deep analysis with citations behind each claim and totals that match the official forecast.

A business analyst might ask:

Why did Northeast revenue miss forecasts last quarter? Back it up with our sales and competitor data.
Which products and channels drove most of the gap, and was it fewer orders or smaller deals?
Write the variance summary for the QBR using our certified revenue metric

Microsoft Copilot Cowork: Automate account planning with governed data

Microsoft Copilot Cowork automates agentic work across Microsoft 365. It plans multi-step work, runs it in the background, and hands back finished spreadsheets, decks and email drafts. With the Azure Databricks Genie plugin, Cowork can now help an account team prepare for a client meeting. Through Genie, Cowork can pull usage and account trends from Azure Databricks, using governed and certified metric definitions. From there it fills in the account team's briefing template in Word, builds an Excel overview with charts, and creates a client-ready presentation. Cowork also books the client call with the account update, then waits for approval before anything is sent.

An account director might ask:

Prep me for this week's customer meetings with a brief on usage, spend against commit, and open escalations.
Build the slides for Thursday's business review using our certified consumption metrics
After each meeting, use the Teams recap to draft an Outlook follow-up with the numbers we discussed

Grok Bot: Hire an AI teammate to run your growth campaigns end to end

Grok Bot gives your team always-on AI teammates. Each one has its own role, memory and tools, and its own cloud computer that stays on. You can hire a growth marketing teammate and give it a job, and it proactively runs the workflow on a schedule without being prompted. With Genie connected, it works from the campaign spend, qualified opportunity and pipeline definitions your company already governs in Databricks, so its numbers match the ones your team reports.

Every week, the teammate runs the whole campaign review loop. It asks Genie which campaigns are missing targets, digs into why, and drafts budget shifts with the evidence attached. It also pulls product-usage and CRM data to build re-engagement audiences. Your growth lead reviews and approves, and the teammate then sets up the retargeting or lifecycle campaigns in your connected ad tools. One person's weekly analysis can become a workflow that runs itself.

Jobs a bot teammate might take on:

Every Monday, find the three campaigns with the highest cost per qualified opportunity, recommend budget changes
Build a re-engagement shortlist of accounts whose usage dropped more than 30% over the last 28 days.
Find the campaigns that are behind their quarterly targets, work out the gaps, and draft an action plan

How the Genie plugin works

Genie plugins extend the same governed business context through our Genie MCP server into AI assistants, coding agents, and client interfaces. Genie Ontology is the governed context layer in Databricks that helps Genie understand your business. Together, they allow organizations to build business meaning once in Genie Ontology and access it across an evolving AI ecosystem. Organizations can also leverage the MCP server directly in your tools to access insights from Genie

Get Started with the Databricks Genie plugin

Enable the Databricks Genie plugin in your agent workspace. Once connected, start asking business questions in your normal workflow or assigning tasks to your agents. Genie handles the data retrieval and governance. Agents without native plugins can also access Genie via the MCP server.

Join us for the Data + AI Live: Context is just the beginning event with Ali Ghodsi and Greg Brockman to learn more about how Databricks is powering the enterprise context for your agents.

For more information, review these resources:

FAQs

What are Databricks Genie plugins?

Databricks Genie plugins give the AI agents your teams already use a governed interface to your enterprise context in Databricks. They work natively in ChatGPT, Claude, Microsoft Copilot Cowork and Grok Bot. Agents without a native plugin can access Genie through the MCP server.

How do AI agents use the Genie plugin?

Agents use it in two ways. They can hand off a business question, and Genie works out what the terms mean using Genie Ontology, finds data the user is allowed to see, writes and runs the SQL and returns an answer with citations. Or the agent keeps the job and asks Genie only for the business context it's missing, then continues with its own tools.

What does Genie Ontology add to an AI agent?

Genie Ontology is the governed context layer in Databricks. It supplies the definitions, relationships and source context an agent needs to interpret the business, such as the company's agreed activation metric. Organizations build that business meaning once and apply it across every agent connected to Genie.

Can Genie route questions to specialized Genie Agents?

Yes. For questions that need specialized analysis, Genie can route work to domain-specific Genie Agents. For example, a pipeline attribution question goes to the marketing Genie Agent and an operational cost question goes to the finance Genie Agent.

How do I get started with the Databricks Genie plugin?

Enable the Databricks Genie plugin in your agent workspace, then ask business questions or assign tasks to your agents in your normal workflow. Genie handles data retrieval and governance. In clients that support the MCP Apps extension, Genie also returns interactive visualizations and Genie Ontology citations in the conversation.

来源：Databricks：Blog · databricks.com