Databricks Genie apps for Slack and Microsoft Teams are now generally available

5 hours ago

8 min read

Databricks Genie apps for Slack and Microsoft Teams generally available cover over a teal city skyline. Data Studios original cover

Databricks announced on November 9, 2026 that the Databricks Genie app for Slack and the Databricks Genie app in Microsoft Teams are generally available. Both apps let people ask questions about their data from a direct message, or by mentioning Genie in a channel or group conversation, and receive answers drawn from Genie One or from a specific Genie Agent that has been pinned to the conversation. The general availability entries appear in the Databricks release notes for AWS and Google Cloud dated November 9.

The change moves conversational analytics out of the Databricks workspace and into the chat tools where business teams already discuss numbers. Each question runs with the Databricks credentials of the person who asked it, so Unity Catalog permissions keep deciding which tables and rows Genie can read. Responses posted in a channel are public by default, though, and the documentation states that public responses are visible to everyone in the channel or group conversation, including people who do not have access to the underlying data.

For BI administrators the practical work therefore sits in the workspace settings, in the channel configuration and, for workspaces that use Private Link or IP access lists, in network rules that have to admit the relay service Databricks runs for each app. The Azure Databricks release notes for November did not list the general availability entries when they were checked on November 10, 2026, and the AWS documentation page for the Teams app still carried a Public Preview label on the same day, so Azure customers in particular should confirm the status in their own account console before planning a rollout.

··········

WHAT GENIE IN SLACK AND MICROSOFT TEAMS DOES.

In both chat platforms a user opens the app from the Apps area and asks a question in a direct message, or types the mention (@Genie in Slack, @Databricks Genie in Teams) in a channel or group chat. Answers include sources and suggested follow-up prompts, and charts arrive as image files in the reply. Slack replies also offer a View conversation button that opens the same conversation in Databricks Genie, which gives analysts a path from a quick answer in chat to the full context in the workspace.

The scope of each conversation is either Genie One, the general entry point, or a specific Genie Agent. When an agent is pinned, new conversations go straight to that agent through the Agent mode API; otherwise the app uses the Genie One API. In a direct message the app routes the question to the most appropriate Genie Agent. In channels and group chats only a channel-wide pin, saved with the Save as channel-wide setting option, sends conversations to an agent, while a personal configuration keeps using Genie One. Conversations that started on Genie One stay there after a pin is set, and feedback always goes to Genie One.

Both apps can also create scheduled tasks from a direct message. A user can ask for a schedule in natural language or use the button offered under an answer, and the results are delivered back by direct message. The Slack documentation adds that Genie skills can be created from the app, and that creating new Genie Agents from Slack is not supported.

........

Capability	Slack app	Teams app
Mention in channels	@Genie	@Databricks Genie
Configuration command	/databricks-genie config	config
Scheduled tasks	Direct messages	Direct messages
File uploads	Not documented	DMs without a pinned agent
Save documents as PDF	Not documented	DMs, app 1.0.17 or later
Relay service region	us-west-2	East US 2

........

··········

INSTALLING THE GENIE APPS AND WHO SIGNS IN.

In Slack, a Slack workspace administrator installs the app from the install link in the Databricks documentation, reviews the requested permissions and approves them; if the Slack workspace requires admin approval for third-party apps, a Slack admin must approve the request first. Users can then start from a channel with an @Genie message or from the App Home. On first use each person signs in to a Databricks workspace, choosing a recent workspace, using the account selector or entering a workspace URL.

In Microsoft Teams, a Microsoft 365 administrator installs Databricks Genie from the Microsoft marketplace and approves its permissions. The app requests the Entra ID permissions offline_access and User.Read, plus resource-specific permissions for chat messages, channel messages, channel members and channel settings that apply only to the chat, channel or team where the app is installed. If users see a "Need admin approval" message when signing in, a Microsoft Entra administrator has to grant tenant-wide consent to the Databricks enterprise application; the Slack documentation describes the same step for organizations that sign in to Databricks through Entra ID. The Teams page also lists an account console preview toggle as a requirement, which is the label that had not been updated when the page was checked.

Compute is chosen for the user. Neither app lets anyone pick a SQL warehouse; Databricks selects one that the asking user is authorized to use. Teams administrators who want document features should know that file uploads and PDF saving rely on each user having OneDrive for Business.

··········

PERMISSIONS AND PUBLIC ANSWERS IN CHANNELS.

Genie answers every question with the credentials of the person who asked it, and that person sees only data already granted to them in Unity Catalog. The governance gap sits in the visibility of the answer. Responses are public by default, so a sales manager with access to a revenue table who asks a question in a busy channel posts the result where every member can read it, whether or not those members could query the table themselves.

Visibility is controlled at three levels. A workspace administrator sets the outer limit under Settings, Security, External access; a channel owner can apply a setting to the whole channel; and an individual user can set a personal preference that applies when no channel-wide setting exists. The workspace level takes precedence over the channel, and the channel over the person.

........

Setting	Where	Options
Allowed collaboration platforms	Workspace, External access	Allow all (default), Allow Slack only, Allow Teams only, Deny all
Allow public messages in collaboration platforms	Workspace, External access	Off makes every response private to the asker
Message visibility	Channel or personal configuration	Visible to all members or only to the asker
Genie scope	Channel or personal configuration	Genie One or a pinned Genie Agent

........

Workspace changes can take up to an hour to apply. When a platform is not allowed, users cannot connect the app and existing users receive a message saying the connection is not allowed. The Deny all option is also the only documented way to switch the apps off from the Databricks side, since neither page describes uninstall steps.

Approval prompts add a second control for agents that use connected external tools. When Genie needs permission to act in another system, the reply shows Allow and Deny buttons, sometimes with a scope of Once, Always in this chat or Always, and only the person who asked the question can respond. That approval feature was released on November 3 as a Public Preview and did not become generally available with the apps.

··········

[[ADPLUS_300x250_1]]

PRIVATE LINK, IP ACCESS LISTS AND DATA RESIDENCY FOR GENIE CHAT APPS.

Requests from both apps pass through a central Databricks control plane service, located in us-west-2 for Slack and in East US 2 for Teams, regardless of where the workspace runs. Databricks states that this service stores only the mapping between the chat workspace or tenant and the Databricks workspace and does not persist request or answer content. Genie processes the question and stores conversation history in the workspace's own region, and the language model runs according to the workspace's Geo and its cross-Geo processing setting, so a workspace in the Europe Geo without cross-Geo processing keeps model processing in Europe.

Workspaces with IP access lists must add the outbound IP addresses of the relevant control plane region. Workspaces with front-end Private Link need an account administrator to add a context-based ingress rule: on the Ingress tab, an Allow rule with API access and the Genie scope, and a source of Partner Platforms with the matching Genie app selected (Databricks Teams App in the Teams documentation), after which the private access settings object is set to have public access governed by the ingress policies. Databricks added partner platform sources for Power BI, Tableau Cloud, dbt Cloud and the two Genie apps on November 5, so administrators no longer track those IP ranges by hand. The Teams page warns that the public access change applies to every workspace using the same private access settings object.

Usage can be audited in the system.access.audit table, where chat requests appear under the genieChat service with the createGenieChatResponse action and a source value of slack, teams or webapp.

SELECT event_time,
       user_identity.email AS user_email,
       request_params.source AS source
FROM system.access.audit
WHERE service_name = 'genieChat'
  AND action_name = 'createGenieChatResponse'
  AND request_params.source IN ('slack', 'teams')
ORDER BY event_time DESC;

··········

LIMITS OF THE GENIE SLACK AND TEAMS APPS.

The documented limits are specific. In Teams, file uploads work only in direct messages without a pinned agent, with up to five PDF, Word or CSV files per message and 20 MiB per file; files stay available for follow-ups until the user sends newchat. Saving a document as a PDF is limited to direct messages, the offer expires 15 minutes after it is posted, and files larger than 20 MiB cannot be saved. In Slack, group conversations can add the app only when the group DM is created, ephemeral messages in channels disappear after events such as a client reload or session expiry, and public chart images require the files:write permission, which may mean reinstalling the app.

When an agent is pinned, Genie One may rewrite a question before passing it to the agent, which matters for teams that tune an agent around precise business definitions. Databricks also states that its agents may generate inaccurate responses and that output should be verified before acting on it. Organizations that already map Power BI measures into Genie, a capability covered in Databricks Genie can now reuse Power BI DAX measures in Beta, should test whether answers in chat match the definitions they expect, and agents that read documents, described in Databricks Genie Agents now answer from PDFs and slides in Unity Catalog volumes, deserve the same check before they are pinned to a channel.

Neither documentation page mentions pricing or billing for the apps. Questions asked in chat are answered by Genie and run on a SQL warehouse the user can access, but the pages do not state how that usage is metered.

··········

DATABRICKS GENIE IN CHAT: WHAT ADMINS SHOULD SET BEFORE ROLLOUT.

The release notes for AWS and Google Cloud confirm general availability of both apps as of November 9, 2026, along with the per-user permission model, Genie One or pinned agent routing, scheduled tasks from direct messages and the relay services in us-west-2 and East US 2. Azure Databricks customers should confirm availability in their own account, since the Azure release notes had not listed the change when checked.

The settings that decide the risk profile belong to each organization: whether the workspace allows Slack, Teams or neither, whether public answers are permitted at all, which channels get a pinned agent, and whether Private Link workspaces receive the ingress rule. Turning off public messages before the first install is the most conservative choice for workspaces that hold sensitive financial or customer data. Pricing for chat usage, uninstall steps and the general availability of tool approvals remain undisclosed or in preview, and the same conversational pattern is now appearing across vendors, as with the OpenAI Data agent in ChatGPT Work.

··········

·····

DATA STUDIOS

·····

datastudios.org