How to Get Started with Tableau Agent

Tableau Agent is a generative AI feature built into Tableau that turns natural-language prompts into visualizations, calculations, and data insights. Available in Tableau Desktop (2025.1+), Tableau Cloud, and Tableau Server (2025.3+), it lets analysts describe what they want instead of building views from scratch. Enabling it takes five minutes once you have the right license in place. Since October 2025, AI in Tableau no longer consumes Einstein Request credits for standard agent usage.

What Tableau Agent Can Do

Tableau Agent covers three stages of a typical analytics workflow.

In the data preparation phase, use natural language to reshape and clean your data before analysis. Prompts like "remove rows where revenue is blank" or "pivot this table so each month is a column" translate into preparation steps. The agent shows you a plan before executing, so you can confirm each transformation.

In the exploration phase, describe the analysis you want and the agent builds a draft visualization. "Show me monthly revenue by region as a bar chart" or "what month had the largest growth in donors compared to the previous month" will produce a starting view you can then refine. The agent can write calculated fields in Tableau's expression language and explain what each formula does in plain English, which removes the most common barrier for analysts who understand the business question but have never written a level-of-detail expression.

In the consumption phase, Dashboard Narratives (available as a beta in version 2026.1) generates a written summary of what is happening in a dashboard. It identifies significant trends and notable outliers across the charts on screen. This is useful for preparing briefings or sharing findings with stakeholders who do not open Tableau directly.

What You Need Before You Start

The requirements differ depending on where you use Tableau Agent.

In Tableau Desktop (2025.1 or later), the Agent icon is active only when you are signed into a Tableau Cloud site that has Tableau+ and AI in Tableau enabled for web authoring. Without a Tableau+ subscription, the icon is visible but inactive. Desktop connects to the agent's backend through your Cloud site, so a live internet connection is required.

In Tableau Cloud, a site administrator must turn on AI in Tableau under site settings. Once enabled, users with the Creator or Explorer role can access the agent. Viewers do not have access. All conversation data sent to the language model is processed through Salesforce's Einstein Trust Layer, which means it is not stored by the model and is not used for training.

In Tableau Server (2025.3 or later), setup mirrors Cloud: an administrator enables AI in Tableau in the server settings, but the backend connects directly to your own language model provider rather than the Einstein Trust Layer. Role requirements are the same as Cloud.

If your organization has not yet upgraded to Tableau+ or enabled AI features, you will need to work with your site administrator or Salesforce account manager before proceeding. A free Tableau Cloud trial that includes Tableau Agent features is available at tableau.com/products/trial.

How to Enable Tableau Agent on Tableau Cloud

These steps assume you are a site administrator:

1. Log into your Tableau Cloud site and navigate to Settings, then General.
2. Scroll to the AI in Tableau section and enable Tableau Agent for web authoring.
3. Optionally, opt into Dashboard Narratives (Beta) if you want dashboard authors to configure written summaries for their consumers.
4. Save settings. Creators and Explorers will see the Tableau Agent icon appear in the toolbar on their next page load.

For Tableau Desktop users, no additional configuration is needed beyond signing into a Cloud site where the feature is already active.

Running Your First Analysis

Start with an extract or uploaded file. Tableau Agent performs best with data extracts and file-based connections (.csv, .txt, .xlsx, .hyper). Live connections work but respond more slowly on large datasets, and cubes are not currently supported.

Open a workbook, connect to your data, and navigate to a worksheet. The Tableau Agent icon appears in the top toolbar, immediately to the right of the Show Me button. Click it to open the Agent pane on the right side of the screen.

Type a prompt describing what you want to see. For a first test, something concrete works better than something open-ended. "Show me total sales by product category for the last 12 months as a horizontal bar chart" will produce a more reliable result than "give me insights on my data."

The agent will propose a visualization plan, show you which fields it intends to use, and ask for confirmation before building. Review the field mapping carefully at this step. If the agent picks the wrong date field or the wrong aggregation, correct it before proceeding. Once confirmed, it builds the view.

From there, you can iterate conversationally. "Sort the bars by value descending," "add a reference line at the average," or "break this down by region" each modify the existing view rather than starting from scratch.

For calculated fields, try: "Create a calculated field that shows year-over-year revenue growth as a percentage." The agent writes the calculation, adds it to your data pane, and explains what each part of the formula does in plain English. In teams where fewer than 30 percent of analysts write custom calculations regularly, this single capability tends to drive the highest reported time savings.

Using Dashboard Narratives

Dashboard Narratives is available as an opt-in beta for Tableau Cloud sites running version 2026.1 or later. It is not available for Tableau Server.

As a site administrator, opt into the beta in site settings. As a dashboard author, open a published dashboard that uses a published data source (locally embedded sources are not supported), then open the Tableau Agent pane and configure which worksheets should be included in the narrative summary. You can toggle Dashboard Overview and Dashboard Insights separately.

Dashboard consumers who open the published dashboard will see a Narratives button at the top. Clicking it generates a written summary of the charts on screen. The narrative refreshes when the underlying extract refreshes, so it stays current with your scheduled data updates. You can copy the text directly into a briefing document or an email.

If you work primarily with uploaded spreadsheets and want to run natural-language analysis without configuring a Tableau Cloud site, VSLZ handles the same kind of query from a file upload with no server setup required.

Prompt Strategies That Work

Several patterns consistently produce better results with Tableau Agent:

Be specific about the output format. "Bar chart" performs better than "a chart." "Sorted descending by value" avoids a follow-up correction.

Name the fields explicitly. If your data has a field called gross_revenue rather than revenue, use the exact name. The agent can infer common synonyms but explicit naming eliminates ambiguity.

Use multi-step prompts for complex calculations. Rather than asking for a complex calculation in one prompt, ask the agent to compute an intermediate result first, confirm it looks correct, then proceed to the next step. This reduces compounding errors in long formula chains.

Ask for explanations. "Explain this calculated field" and "Why did you choose this aggregation?" both work. Understanding the agent's reasoning helps you catch incorrect assumptions before they carry through to the final view.

Known Limitations

Tableau Agent does not yet support all visualization types. Map-based views, custom chart extensions, and some advanced formatting options require manual authoring. The agent will tell you when a request falls outside what it can automate.

Dashboard Narratives does not support dashboards that use locally embedded data sources. The data source must be published to the Tableau Cloud site before Narratives can be configured. This also means Dashboard Narratives is unavailable for Tableau Server deployments.

On very large live data sources, response time can be slow. Switching to an extract before an agent session is the most reliable workaround.

Getting the Most Out of Your Setup

Once Tableau Agent is running, the fastest way to build fluency is to use it for tasks you already know how to do manually. Building a calculated field you could write yourself lets you compare the agent's output against your own understanding and learn its patterns. After a few sessions on familiar tasks, the agent becomes more useful for genuinely novel analysis where you would otherwise spend time experimenting.

For teams deploying this across multiple analysts, enabling the feature at the site level with a shared extract library gives everyone consistent, performant access without each user configuring their own connection.