Skip to main content

Alteryx Agents

Create an Alteryx Agent to help you ask natural-language questions about trusted datasets in Alteryx One. Alteryx Agents are AI-powered analytic experiences that return trusted insights within Alteryx One permission boundaries.

Before You Start

To create and use Alteryx Agents, you must have the Agent Studio capability assigned to you with the AI Agent privilege enabled. Alteryx Agents are available to users on Professional and Enterprise Editions.

You can build an agent only with datasets that you can access. To learn more about how Alteryx Agents use datasets, go to Set Up Assets.

Create an Alteryx Agent

Suggerimento

Not sure what to put in your custom instructions? Use the AI Agent Builder to work through your goals and draft a configuration. To learn more, go to Build and Improve an Agent with AI.

  1. Open Agent Studio in Alteryx One.

  2. Select New Alteryx Agent.

  3. Enter a Name and Description.

  4. Select Choose Datasets to give your agent access to 1 or more datasets. The Description column shows what each dataset contains. Hover over a description to read all of it. AI-generated descriptions are labeled with an icon.

  5. Add custom instructions to guide the agent's behavior and analysis approach.

  6. Add starter prompts to guide users when they first interact with your agent.

  7. Save the agent.

  8. Use the preview window to review your agent configuration.

Datasets

Selecting a dataset allows the agent to answer questions and generate insights from that dataset. If you select multiple datasets, the agent decides which datasets are relevant to the user's question and performs calculations against any or all of them individually.

Datasets aren't merged. The agent combines the resulting calculations and insights to produce a richer analysis for the user.

Dataset descriptions help you find the right dataset for your agent. Descriptions appear in the Choose Datasets dialog and in a tooltip when you hover over a selected dataset. When a dataset doesn't have a description written by a person, Alteryx One can use AI to generate one, so you can quickly see what a dataset contains. AI-generated descriptions are labeled with an icon.

On the Alteryx Agents tab, the Agent Assets column shows how many datasets each agent can use.

If you can't find a dataset, confirm that you have permission to use it. If you have view access but can't set up the dataset, contact the dataset owner or an editor.

The AI Agent Builder suggests only datasets that you can access, and it can't prepare a dataset for insights yet. If the AI Agent Builder tells you a dataset isn't prepared for insights, open the link it provides. The dataset's Data Insights tab starts preparing the dataset when it opens, and preparation continues in the background. When you're notified that insights are ready, return to the AI Agent Builder and follow the prompts.

Dataset Size and Refresh

Agents analyze prepared datasets in Alteryx One. When a dataset is loaded or refreshed, Agent Studio retrieves the latest data from the underlying source and prepares it for insight generation. After the data is prepared, questions from an Alteryx Agent or supported MCP client run against the Alteryx One-hosted dataset, not directly against the original source.

Agents use the latest successfully refreshed version of the dataset. If the source data changes, refresh the dataset so the agent can use the updated dataset. You can refresh a dataset manually, schedule a refresh, or refresh it through a plan in Orchestration.

If dataset preparation fails, the dataset might be too large or too detailed for the current processing environment. Use a filtered or aggregated version of the dataset that includes only the rows, time period, and columns required for the agent's intended questions.

Custom Instructions

Custom instructions let you guide agent behavior. Use custom instructions to describe the agent's:

  • Objectives

  • Role

  • Tone

  • Dataset Context

  • Domain-Specific Terms

  • Key Measures

  • Analysis Methods

Custom instructions strongly influence how the agent interprets user questions and frames responses. Spend time refining them so the agent has enough context to answer consistently and in the right style for its intended audience.

Not sure where to start? Use the AI Agent Builder to draft your instructions, or chat with your agent first to see what guidance it needs. To learn more, go to Build and Improve an Agent with AI.

Starter Prompts

Starter prompts are suggested questions shown at the beginning of the agent chat experience. They help users understand what the agent is intended to answer.

Good starter prompts reflect the core questions the agent is designed to support. If you don't put any starter prompts in yourself, 3 basic starter prompts will be added for you.

The AI Agent Builder can propose starter prompts when you create an agent, and refine them when you edit one.

Preview the Agent

Preview the agent before you share it. Ask questions that match the agent's intended use and review whether the answers use the expected datasets and analysis approach.

If you used the AI Agent Builder, review its proposed instructions before you preview the agent. The proposed instructions are a starting point and might need your edits.

Chat With an Agent

When you chat with an agent, ask questions that match the agent's intended purpose and available datasets. Review responses before you use or share the results.

The Agent Assets card lists all datasets available to the agent, not only the datasets used in a response. Expand or collapse the card to show or hide the list, and hover over a dataset to see its description. AI-generated descriptions are labeled with an icon.

Alteryx Agents can provide traceability through links back to the underlying data queries or datasets when that information is available in the response.

Some analytics responses do not yet have a traceability link.

To view links back to source data in agent responses, your role must have the Auto Insights capability assigned. Without it, source links in responses don't work. Contact your Workspace Admin to request this capability.

Share and Manage Access

Creators and editors can share agents. You can interact with an agent only when you have access to all underlying dependencies.

  • Creators can create agents.

  • Creators and editors can edit and share agents.

  • Viewers can interact with agents if they have access to all underlying dependencies.

  • If a user is missing any dependency, they are blocked from accessing the agent.

  • Dependencies can be shared only by owners or editors of those dependencies.

What Agents Can Analyze

Agent analysis can include:

  • Trend analysis, such as how metrics evolve over time.

  • Root cause analysis, such as why changes occurred.

  • Common cause analysis, such as relationships between metrics.

  • Outlier detection, such as unexpected or anomalous behavior.

Charts and Images in Agent Responses

When an agent answers with a chart, Alteryx generates the chart as an image and includes a link to it in the response.

  • Generated insight images are stored temporarily and deleted after 12 hours. When an image is deleted, its link stops working and the chart no longer appears in earlier responses. Ask the question again to generate a new chart image.

  • Image links work without signing in. Anyone with the link can view the image until it expires. The links are long, randomly generated, and rate-limited, so they can't be easily guessed.

  • The image contains only the chart. It doesn't include the agent's written answer or explanation.

Nota

Share chart links only with people who are allowed to see the underlying data. If your organization restricts links that work without signing in, consider this when you decide who gets access to agents.

Model Details

Alteryx Agents use Alteryx-managed models. Customer-hosted models aren't supported.