Analyze Using Dremio's AI Agent
Dremio's AI agent is built into the Dremio console and lets you work with your data end to end — from exploring and analyzing to creating and troubleshooting — entirely in natural language, without writing a single line of SQL. The AI agent only operates on data and entities that you have the privileges to access or modify.
Open the AI Agent
You can access the AI agent in two ways:
- Type a question into the chat on the Home page in the Dremio console.
- Use the keyboard shortcut Cmd + Shift + G on Mac or Ctrl + Shift + G on Windows.
To use the AI agent, you must be granted CALL MODEL on the default model provider. See Configure Model Providers for setup instructions.
Use the AI Agent
Manage your conversations – The AI agent keeps a history of your past chats. You can start a new chat for a different task, return to a previous conversation to pick up where you left off, and rename or delete chats to stay organized. Use the search bar in the agent chat to quickly find a past conversation.
Reference datasets with @ – Type @ in the chat box to trigger autocomplete and reference a specific table or view in your prompt. This helps the agent focus on the exact data you want to work with instead of searching across your entire catalog.
Audit the SQL – When the agent writes and executes SQL, you can expand the tool calls in the chat window to review exactly what query was run and how the results were generated.
Project Context
A user with the ADMIN role or OWNERSHIP privilege on the project may configure project context, which is a set of instructions that influences how the AI agent reasons and responds for all users in the project. This can affect the agent's focus, terminology, and how it interprets your questions.
To view the project context for your project, go to Admin > Project > AI. For configuration instructions, see Configure AI Agent Project Context.
What the AI Agent Can Do
Dremio's AI agent can help you with core workflows such as discovering and analyzing data, modifying data models, creating visualizations, and troubleshooting query performance. Each section below includes example prompts to help you get the most out of the agent.
Discover and Explore
The AI agent helps you discover available data by breaking down table schemas, identifying relevant tables and views, and surfacing patterns in your data. The agent uses wikis and labels as well as sampling to determine relevance.
| Okay Prompt | Great Prompt |
|---|---|
| What tables can I use? | Which tables or views have customer location data? |
Analyze
The AI agent writes and executes SQL on your behalf based on your natural language input and the information available from the AI Semantic Layer.
| Okay Prompt | Great Prompt |
|---|---|
| I want to see analysis of customer activity | I want to see an analysis of customer purchase activity by region, by customer type for each month of the year. |
| Which customers are the most valuable? | Which customers have spent the most with us over the lifetime of the relationship? |
Visualize
The AI agent generates visualizations based on your natural language input. You can specify chart type, axis requirements, grouping, and trendlines. Each visualization is accompanied by insights that explain the patterns the agent found.
Once a visualization is created, you can:
- Toggle between the chart and a table view of the underlying data
- Download the underlying data as a CSV
- Ask follow-up questions to refine or change the visualization
- Save the visualization to your catalog to share it with others as a live, always-current metric
When you save a visualization, it is added to your catalog and always reflects the latest underlying data when opened. You can control who can view or edit it by managing privileges on the visualization. For more information, see Visualization Privileges.
The agent supports these chart types: bar, line, area, scatter, pie, and heatmap. It can also create diagrams like flowcharts and entity-relationship diagrams when prompted or when it determines a diagram would be helpful.
| Okay Prompt | Great Prompt |
|---|---|
| Visualize the data | Visualize the data as a bar chart with month on the x axis and sum of purchase value as the y axis |
| Create a visual trendline showing me the activity | Create a visualization with a trendline showing customer activity by month |
Create and Manage Data
The AI agent can execute write operations on your behalf, including DML operations such as inserting, updating, or deleting records, as well as DDL operations such as creating, altering, or dropping tables and views.
Before running any write or schema-changing operation, the AI agent pauses and presents the operation for your review. You can execute it or request changes — nothing runs until you approve it.
Your existing Dremio access privileges apply. If you don't have the required permissions to perform an operation, the AI agent will not be able to run it on your behalf.
You don't need to specify every detail — the agent can take on complex, multi-step operations from a high-level description. It will work out the structure and present each operation for your review before anything is executed.
| Okay Prompt | Great Prompt |
|---|---|
| Set up a medallion architecture for my weather data | Create a medallion architecture using @raw.nyc_weather to build daily, monthly, and yearly statistical views for temperature, precipitation, and wind. Start at the Bronze layer. |
| Update the sales table | Update the @quarterly_sales table to set status = 'closed' for all records where close_date is before 2024-01-01 and status is 'pending'. |
| Delete old records. | Delete all records from the @customer_orders table where order_date is before January 1, 2022. |
Troubleshoot and Optimize
Ask the agent to review SQL queries or past jobs to identify bottlenecks and suggest optimizations. For more information, see Explain SQL and Explain Job.
If the AI agent is unable to access a table or view, verify that you have the necessary privileges on that object. The agent can only operate on data you are authorized to access.
Tips for Better Results
- Be specific when updating or deleting – For operations that modify or remove existing records, include clear conditions and filters so the agent affects exactly the right data. For creation tasks, a clear description of your intent is enough. The agent will work out the structure.
- Include business context – Use the terms your team uses. The agent reads wikis and labels to understand your business terminology.
- Ask follow-up questions – The agent maintains context within a conversation. You can refine results with follow-ups like "break that down by product category" or "now show that as a line chart."
- Ask for explanations – If you're unsure about the results, ask "explain how you got this" or "show me the SQL you used."
- Verify the output – Generative AI can make mistakes. Always review SQL queries, data interpretations, and proposed write operations carefully before approving them.
Limitations
- Large tasks: The AI agent may not complete very large or highly complex tasks in a single session. For multi-step analyses, consider breaking your question into smaller parts and iterating across multiple prompts.
Related Topics
- Discover Data – Find and explore data across Dremio in other ways.
- Configure Model Providers – Set up the AI model provider for your organization.
- Data Privacy – Learn more about Dremio's data privacy practices.