Research Agent in Genie spaces
Research Agent is in Beta. Workspace admins control access using the previews page. See Manage workspace-level previews.
Research Agent extends Genie's capabilities to help you uncover deeper insights and tackle complex business questions using multi-step reasoning and hypothesis testing. When you ask a question, Research Agent creates and refines a research plan, running multiple SQL queries, learning from each result, and iterating until it has enough evidence to provide a comprehensive answer.
Unlike standard Genie queries, Research Agent:
- Creates research plans: Develops a structured approach and hypotheses for answering complex questions.
- Runs multiple queries: Executes several SQL queries to gather evidence from different angles.
- Learns and iterates: Continuously adjusts its approach based on what it discovers, refining its reasoning until it's confident in the answer.
- Delivers comprehensive reports: Provides detailed summaries with citations, visualizations, and supporting tables.
Research Agent is designed for complex, exploratory questions like:
- Why did revenue spike in June 2025?
- How can we reduce order cancellations this holiday season?
- What factors contributed to customer churn last quarter?
- Which marketing campaigns had the best ROI and why?
Requirements and availability
To use Research Agent, your workspace must meet the following criteria.
Have a well-prepared Genie space
Research Agent relies heavily on the context within your Genie space to reason about your data. You must meet all requirements for setting up a Genie space. See Set up a Genie space.
To optimize your Genie space for Research Agent, follow the comprehensive guidance in:
- Set up and manage an AI/BI Genie space: Initial setup and configuration
- Curate an effective Genie space: Best practices for setting up an effective Genie space
- Build a knowledge store for more reliable Genie spaces: Adding space-specific context
Regional availability
Region | Availability |
|---|---|
Americas & Europe workspaces | Available (cross-Geo not required) |
All other regions (Australia, India, Asia, etc.) | Available only if cross-Geo is enabled. See Enable cross-Geo processing. |
Manage Research Agent
Workspace admins can control access to Genie Research Agent using the previews page. See Manage workspace-level previews.
After Research Agent is enabled, it is available for all Genie spaces within that workspace. All users with access to a Genie space can use Research Agent. All requirements for accessing a Genie space and its data apply. See Set up a Genie space.
Use Research Agent
To ask questions using Research Agent:
- Open a Genie space.
- Click
the Research Agent icon in the chat box. - Enter a question and send your prompt. Research Agent begins its research and may ask follow-up questions to clarify your intent. Research Agent responses may take longer to generate than standard Genie space responses. When complete, it answers with a final report containing specific findings, visualizations, and citations to research steps.
- After the report is generated, you can send additional clarification prompts to refine your report or generate new reports on different research questions. You can provide feedback to help improve Research Agent.
What questions can Research Agent answer?
Research Agent performs best when you ask specific, data-driven questions about trends, patterns, or relationships in your business data. It excels at questions that need multi-step reasoning, such as:
- Why did revenue spike in June 2025?
- How can we reduce the number of order cancellations this holiday season?
- Tell me how our new product is doing in EMEA
- What factors contributed to customer churn last quarter?
- Which marketing campaigns had the best ROI and why?
What Research Agent cannot do
Research Agent has the following limitations:
- Data scope: Research Agent only uses data available in the current Genie space. It cannot query your full catalog or data outside the space.
- Unstructured data: Research Agent can only reason on structured data sources included in the Genie space. It cannot leverage unstructured data.
- Web search: Research Agent does not search the web for additional context or information.
Monitor and improve
Research Agent improves as you refine your Genie space based on user feedback and usage patterns.
- Review feedback: Use the Monitoring tab in your Genie space to see how users interact with Research Agent and what questions they ask.
- Identify gaps: Look for questions where Research Agent struggles or provides incomplete answers. Add targeted instructions or examples to address these gaps.
- Iterate continuously: Think of your Genie space as a living resource that improves over time with user feedback and refinement.
For Genie space users, you can leave response feedback for space authors to review. See Provide feedback.
Frequently asked questions
Are there additional costs for Research Agent while in Beta?
No. During the Beta period, Research Agent does not have additional costs beyond the standard warehouse compute costs associated with running SQL queries.
Does Research Agent use different data than regular Genie?
No. Research Agent uses the same context and data as the standard Genie experience. To learn about the context that Genie uses to generate responses, see How does Genie generate a response? and Privacy and security.
Is Research Agent available through the API?
No. During the Beta period, users can only submit Research Agent prompts through the Databricks UI.
What models power Research Agent?
Today, Research Agent uses a combination of GPT and Sonnet models. Genie is a managed service where we constantly evaluate new models from every provider and leverage the most performant and accurate options. For more information, see Databricks AI assistive features trust and safety.
How do I provide feedback on Research Agent?
Share feedback with your Databricks account team. In particular, Databricks is seeking feedback in the following areas:
- Research steps: Whether Research Agent investigates the right dimensions and generates accurate SQL.
- Answer summaries: Preferences for format, level of detail, and clarity.
- Usability: Any challenges in finding, enabling, or using the feature.
- Context handling: Whether Research Agent uses relevant context effectively.
- Success stories: Examples where Research Agent delivered valuable insights.