Skip to main content
Last updated on

Genie chat

Preview

This feature is in Public Preview.

Chat in Genie is a unified, full-screen natural-language interface for business users to ask data questions. It first searches available Genie Spaces for a match, then falls back to Genie Agent mode across dashboards, queries, and metric views. You can also connect external sources (such as Google Drive and SharePoint) so chat can answer from your company documents and schedule recurring chat tasks that post results back as a thread.

Chat in Genie.

note

Chat in Genie is not available in workspaces with the compliance security profile enabled. See Compliance security profile.

Enable chat​

Workspace admins can enable chat for all workspace users:

  1. As a workspace admin, click your username in the top bar of the Databricks workspace.
  2. From the menu, select Previews.
  3. Search for and turn on New chat experience in Genie.

How chat works​

When you submit a question to chat:

  1. It searches available Genie Spaces for one relevant to your question.
  2. If a matching Genie Space is found, it uses that space to generate a response.
  3. If no matching Genie Space is found, it uses Genie in Agent mode to search for relevant data assets and answer your question.

You can also connect external data sources such as Google Drive, SharePoint, GitHub, Glean, and Atlassian to give chat access to your company documents. See Connect to external tools and sources.

To open chat, click New chat from in the sidebar or use Ask mode in the search bar. You must have the CAN USE permission on at lease one SQL warehouse to use chat

Genie Ontology​

The Ontology is a map of your data and business, automatically generated and maintained by Genie.

Genie automatically extracts snippets of knowledge from your tables, queries, dashboards, docs, and connected apps. For example, snippets can include:

  • Metric definitions: For example, "An 'active user' is a distinct user, deduplicated across all platforms."
  • Authoritative sources: For example, "Revenue should be pulled from the main.finance.certified table."
  • Business rules: For example, "A 'qualified lead' only counts once a demo is booked."

Each snippet gets an authority score based on where it was generated from, how often it is used, and how fresh it is. Snippets are also gated using Unity Catalog permissions. When you ask a question, Genie ranks the most relevant snippets, resolves conflicts, and answers using only the sources you are permitted to see. This generates more accurate responses and reduces latency by reducing the amount of data Genie needs to crawl and query.

To see which knowledge sources Genie used to answer your question, click the Snippet icon. citation icons in a response.

Add workspace instructions for chat​

Beta

This feature is in Beta.

Workspace admins can add custom instructions that apply to every chat conversation in the workspace. Instructions can include information about your organization's data conventions, preferred terminology, or guidelines for how chat should respond.

To add workspace instructions, create a Markdown file at the following path in your workspace:

/Workspace/.genie_workspace_instructions.md

Instructions must be under 20,000 characters. Chat reads this file automatically with no additional configuration required. Instructions apply to chat only and do not affect Genie Spaces or Genie Code.

For best practices on adding custom instructions, see Best practices for Genie Code instructions.

User skills​

User skills let you extend Genie chat with custom capabilities tailored to your personal needs. Skills are available only to you and are loaded automatically when Genie determines they are relevant to your request. You can also invoke a skill manually by @-mentioning it by name.

note

For agent skills in Genie Code, see Extend Genie Code with agent skills.

Create a user skill​

prompt

Tell Genie to do this for you:

Create a user skill that summarizes weekly sales trends for my region.

Genie creates the skill and saves it to your workspace at /Workspace/Users/{email}/.assistant/skills/.

Edit a user skill​

To edit a user skill, ask Genie to make changes in a conversation, or open the skill in Canvas and edit it directly.

To rename or delete a user skill, access the skill file directly in your workspace at /Workspace/Users/{email}/.assistant/skills/.

Invoke a user skill​

Genie automatically loads skills when they are relevant to your request. To invoke a skill, @-mention it by name in the chat input.

Scheduled tasks​

Scheduled tasks run automatically at a specified time and post the results in a chat thread. When a task runs, chat sends you an email with the results.

To create or run scheduled tasks, you must have the Databricks SQL access entitlement. See Manage entitlements.

You can create a scheduled task in two ways:

  • Type a request in natural language in chat, for example: "Send me a daily briefing of all new customer reviews for my store location." Chat might ask clarifying questions, such as the time or which location to use, then creates the task and adds it to the Scheduled tasks section.

  • In the sidebar, click Scheduled tasks, then click the drop-down menu next to + Create in chat and select Create manually.

    Create scheduled task manually

    Fill in the Title, Instructions, Connections, Schedule, and Timezone, then click Create.

    Create scheduled task dialog

To view past runs, edit, or delete a scheduled task, click the task in the Scheduled tasks section. Each run opens in a chat thread.

Connect to external tools and sources​

You can connect external data sources such as Google Drive, SharePoint, GitHub, Glean, and Atlassian to a Genie chat conversation. See Connect to external tools and sources.

Limitations​

  • You cannot specify which SQL warehouse chat uses. Databricks selects one automatically.
  • Workspaces with many Genie Spaces might experience reduced routing accuracy.

Next steps​

On this page