Databricks Launches Genie on Slack with Access Control

Databricks makes its Genie application available to everyone in Slack. The tool promises contextualized answers based on Unity Catalog permissions, actions via MCP with approval, and centralized administration, all without leaving the discussion threads.

Governance of Rights and Centralized Management in Slack

Every question posed to Genie in Slack is processed with the identity of the person asking it, applying Unity Catalog permissions and those of connected sources. In channels, responses utilize the inquirer’s rights, and each channel can decide whether they are shared with the channel or sent privately. Workspace administrators can set a default visibility for the entire workspace. The use of AI is governed through Unity Gateway, with MCP policies, cost controls, and monitoring functions. These policies determine the accessible MCP servers and tools, authorized users, and actions that require human approval, while logs keep track of what has been done and by whom. Environments using Private Link are supported through context-based entry policies that an account administrator must create.

Contextualized Responses and Source Citation

Genie responds directly in the channel, thread, or private message where the question was asked, with answers anchored in the business context accessible to the user. The tool only reads the content of the channels it has been invited to and responds within the context of the relevant channel. Responses are published in the thread that called the assistant, preserving the history of exchanges. By default, they are visible to all channel members, but administrators can establish private responses by default for the entire workspace. Each response cites its sources and allows for further exploration, including in Genie One via links.

From Questioning to Action with MCP and Scheduling

Genie is not limited to providing answers and can act in other tools through MCP connectors, creating or updating documents, commenting on tickets, or modifying records with the user’s credentials. Each change is prepared for approval before execution. Users can schedule recurring analyses from Slack, via a dedicated button or a natural language instruction, with results delivered in their direct messages. It is possible to attach files for analysis and receive documents such as PDFs back in the thread. Genie also sends visualizations directly in Slack. These same capabilities are available in Genie One.

Setup for Administrators and User Experience

Administrators install the Databricks Genie application from the Slack Marketplace and configure the channels as well as the visibility of responses. Users start by sending a DM to Genie or mentioning it in a channel, then connect to Databricks upon first use and choose a workspace. In channels where Genie has been added, anyone who has linked their Databricks account can invoke it, and each channel can be associated with a specific Databricks workspace. It is also possible to pin a channel to a specific Genie agent to frame the scope of responses.

Functional Scope: Genie One by Default, Agents, and Integrations

In Slack, channels use Genie One by default, which routes questions through Genie Search and Genie Agents to cover the entire data estate. The application supports Genie One and agents, and a conversation or channel can be pinned to an agent via the Genie Agentic API. Users can discover and visualize insights from Databricks data, integrate context from sources like Google Drive, Confluence, or Jira, and make updates or comments via MCP after approval. Notifications from the Databricks platform can be received in Slack, and responses can be rated with a thumbs up or down. Recurring analyses can be configured either in Slack or in Databricks via scheduled tasks, with Slack as the destination, provided a prior DM connection has been made. The application is now available in general release and can execute scheduled analyses. Databricks presents it as a governed AI operating in Slack, where teams make decisions. Processing occurs within the Databricks workspace, and the routing service does not store questions or answers. In DM, the user can query their data, search connected knowledge, and obtain results in tables or graphs, as well as upload files for analysis.

Access and Additional Resources

A webinar titled "Just Ask Your Data: Bring Databricks Genie into Slack" is offered to delve deeper into the use of the application. The results of scheduled analyses are delivered via direct message in Slack. It is also possible to create scheduled tasks in Genie One by choosing Slack as the destination.