Aptrics AI
Aptrics AI answers questions about your data in plain language. It is not a text-to-SQL tool pointed at raw tables — it plans over your semantic layer, so every answer is composed from governed metrics and dimensions you have already defined and approved.
That constraint is what makes the answers trustworthy: if a metric is not in the model, Aptrics AI will not invent it, and every response can be traced back to the definition and the SQL that produced it.
Asking Questions
Start a Conversation
Open Aptrics AIfrom the project sidebar and ask a question the way you would ask a colleague — “How did revenue trend by region last quarter?” Aptrics AI resolves the metrics and dimensions it needs from the active project and environment, then returns a chart or table with a short written summary.
Refine with Follow-ups
Conversations keep context. After a first answer you can say “now split that by product category” or “compare it to the same period last year” and the query is rebuilt on top of what came before, rather than starting over.
Write Questions That Land
Use the vocabulary from your semantic model — metric labels and dimension names — and state the time window explicitly. Clear descriptions and synonyms on your metric definitions in Studio directly improve how well Aptrics AI resolves the question, so investing in the model pays off here.
Trust and Verification
See How an Answer Was Built
Every answer can be expanded to show which metrics, dimensions, and filters were selected and the SQL that ran against your warehouse. If the interpretation was not what you meant, adjust the selection directly and the answer re-renders.
Governance Boundaries
Aptrics AI runs with the permissions of the person asking. It can only reach metrics in projects and environments that user already has access to, and it cannot query tables that the semantic layer does not expose.
Hand Off to Explore or a Dashboard
Any answer can be opened in the Metrics tab for deeper slicing, or pinned straight to a dashboard as a tile. Questions that get asked repeatedly are usually better served as a saved dashboard tile than as a fresh conversation each time.
Related
Answer quality follows model quality — see Create Your First Model (Studio) for defining metrics, and Explore and Analyze for manual analysis.