Querying Data
Ask precise warehouse questions, inspect the generated query and data scope, and validate Agent results before sharing or saving them.
Last reviewed July 13, 2026
The Agent can inspect your catalog and run read-only warehouse queries for the active team account. It can return a table, explain the result, and create a visualization when the result shape supports one.
How querying works
- The Agent identifies likely tables from the catalog and workspace context.
- It inspects schemas and, when useful, a small data sample.
- It builds a read-only query with an explicit time range and grain.
- It executes the query within the account’s available datasets.
- It presents the result and the query used to produce it.
Ask a precise question
| Include | Example |
|---|---|
| Metric | Net revenue after refunds |
| Time range | Last 12 complete weeks |
| Grain | One row per week and channel |
| Population | Completed orders only |
| Timezone | Workspace timezone |
| Output | Table and line chart |
Avoid relying on an ambiguous business term such as “active customer” or “revenue.” Ask the Agent to use a metric from the Metrics Library, or define the term in the prompt.
Inspect the result
Before you share or save an answer, check:
- The selected dataset and tables.
- Join keys and whether the join can multiply rows.
- Date field, timezone, and inclusive/exclusive boundaries.
- Filters for status, test data, currency, refunds, or deleted records.
- Aggregation grain and denominator.
- Row limits or sampling disclosed in the result.
Use SQL Editor when you want to iterate directly on the query or compare it with a known warehouse result.
Create charts
Ask for the analytical intent instead of only a chart type:
Show the trend and highlight the largest week-over-week change.Compare channels without hiding small values.Use a table if the categories are too dense for a readable chart.
The Agent selects from the visualizations available in the current canvas. Keep the supporting table or query with the chart so the result remains auditable.
Query safety
- The Agent’s warehouse query surface is read-only.
- Account and dataset access are resolved from the active team workspace.
- Large results may be limited, sampled, summarized, or moved to an artifact.
- Operational changes use explicit platform tools; they are not hidden inside SQL.
Troubleshooting
| Symptom | Resolution |
|---|---|
| Table not found | Ask the Agent to refresh the catalog and show the available datasets |
| Query is slow | Reduce the time range, columns, or table scan before expanding it |
| Count is too high | Inspect joins and compare distinct entity IDs before aggregation |
| Dates differ from another tool | Align timezone, attribution window, and source-processing delay |
| Result changed since yesterday | Check source freshness, mutable definitions, and model run history |