Data Agent

Ask the Data Agent

This page walks through a conversation with the Data Agent: choosing what to query, asking, answering the agent’s clarifying questions, reading the result, refining it, and charting it. If you’re new to the feature, start with the Data Agent overview.

In the app: Agents › Data Agent

Before you start

  • The Data Agent feature must be on for your organization, and at least one ontology must be bound to a database or lakehouse. See Data Agent overview.
  • If your organization is in read-only mode, you can read past results but can’t ask new questions.

Pick an ontology and connection

  1. Choose the ontology from the Ontology dropdown at the top. It lists every ontology in your organization with a database or lakehouse binding.
  2. If the ontology has more than one binding environment, choose one from the Connection dropdown. Each option reads environment — connection (system). With only one environment, it’s shown as a caption and chosen for you.

Each ontology keeps its own conversation. Switch ontologies and you get that ontology’s chat and results; switch back and the first one is where you left it.

Ask a question

  1. Type your question in the box that says Ask for data in plain English…. Press Enter to send; Shift+Enter adds a new line.
  2. While the agent works, a status line shows what it’s doing, such as Understanding your question…, Writing the query…, or Executing query…. Click Stop to cancel the turn.
  3. When it’s done, the agent replies in the chat and the result opens on the right. The reply has two chips: View · N rows shows the result, and SQL shows the query.

Not sure what to ask? In a new conversation, click Suggest questions. The agent proposes a few questions based on the ontology’s bound classes. Click one to ask it, or click Suggest others for a new set.

Tip: Specific questions get better SQL. “Total revenue by region for 2025” gives the agent more to work with than “revenue stuff”.

If something isn’t set up, such as a missing connection or a class with no binding, the agent says so in plain language instead of running a query. If something goes wrong, the message has a Retry button.

Answer clarifying questions

When a question is ambiguous, the agent asks before it runs anything. For example, it might ask whether “this year” means the calendar year or your fiscal year.

  • Each question offers options to pick from, and a box that says or type your own.
  • Click Send answers when every question has an answer.
  • Or click Answer with your best assumptions to let the agent decide and tell you what it assumed.

Your answers carry forward through the rest of the conversation, so you won’t be asked the same thing twice.

[screenshot]
A clarifying question in the chat with its option buttons, the free-text box, and Send answers.

Read the result

The workspace has three tabs: Results, Chart, and SQL.

Assumptions

Above the tabs, Assumptions lists anything the agent decided that you didn’t say. Each one has a type, such as Time range, Time zone, Fiscal calendar, Status filter, Deduplication, Fan-out, Metric, or Entity. Where there’s an obvious alternative, click Change to put “Redo that, but …” in the question box, then edit and send it.

Results

The Results tab shows the rows.

  • Click a column header to sort, and use the column menu to filter.
  • The grid shows the first 1,000 rows. When there are more, a banner says so and Show all N loads the rest.
  • Download CSV exports every fetched row, regardless of the grid’s sort and filters.
  • Empty values show as NULL.

Each query fetches at most your organization’s row limit: 10,000 rows by default. An admin can change it with Data Agent max records on the Organization Settings › Agents tab. When a query matches more rows than the limit, the banner says the fetch was capped.

SQL

The SQL tab shows the query the agent ran, with Copy and Download .sql. Read it to check the agent answered the question you asked.

[screenshot]
A completed answer: the chat with View and SQL chips, and the Assumptions list above the Results grid.

Older results

To keep the page responsive, Kenseme eventually releases the rows of older results in a long conversation. Their View chip, Download CSV, and Chart this are then disabled with the note Result data released — ask again to re-run. The SQL is always kept.

Refine with a follow-up

Ask a follow-up in the same conversation, such as “only 2025”, “add the region”, or “group by month instead”. The agent adjusts the query it ran last. The grid, chart, and SQL update, and a new reply appears with its own chips. A follow-up that only asks the agent to explain something doesn’t change the grid.

To start over, click Clear above the chat and confirm. It removes the conversation and its results for the current ontology. Conversations aren’t stored anywhere else — they last only as long as your current session — so clearing can’t be undone. To keep a result, save it as a report first.

Turn a result into a chart

Let the agent propose one. When a result suits a chart (a category column plus one or more number columns), or when you ask for one, the agent builds it and the Chart tab opens. You can also ask directly: “make that a pie chart”.

Build one yourself. On the Results tab, click Chart this. The Chart tab opens with a builder. If the agent already proposed a chart, the builder starts from its settings.

Control What it does
Chart type Column, bar, line, area, pie, donut, scatter, bubble, heatmap, funnel, treemap, stacked column, or stacked bar. Bubble adds Size column; heatmap adds Group column.
X column / Y columns The category axis and the series to plot. Number columns are listed first.
Title, axis titles Labels for the chart and its axes.
Data labels None, Ends, or All.
Value format Plain, Currency, Percent, or Compact.
Color scheme The palette.
Legend, Normalize to 100% Toggles. Normalize applies to the stacked types only.

The chart updates as you change settings. To keep a result, chart and all, see Data Agent saved reports.

Messages you might see

Message What to do
The Data Agent is busy right now. Please try again in a moment. Many questions are running at once. Wait a few seconds and retry.
The Data Agent took too long to answer (over N seconds) and was stopped. Ask a narrower question, such as a shorter date range.
The database for this connection could not be reached. Check the connection with Test on the Database Connections page, or ask an admin.
That query references a table that is not part of this ontology's bindings. The agent tried to use an unbound table. Bind the table’s class, or rephrase.

If your organization has used up its AI allowance, the agent can’t answer until the allowance resets or more is added. See AI token allowance and usage.