Ask your data
Ask a question in plain language and get a chart, plus the query that produced it.
Ask turns a question into a chart. It picks datasets you already have access to, writes the query, and shows you that query alongside the result.
Ask a question
Open a project and click Ask, or press ⌘J. Type a question:
revenue by month for the last year, split by dealer region
You get back three things:
- A chart — type chosen from the shape of the answer.
- The query it ran, in full.
- The datasets it used, named.
Always read the query
The chart can look right and be wrong — the wrong date column, an unintended join, a filter you did not ask for. The query is shown by default for exactly this reason.
What makes a good question
| Better | Weaker | Why |
|---|---|---|
| “orders per week since January, by status” | “how are orders doing” | Names a measure, a grain, a dimension |
| “top 10 dealers by revenue in 2026” | “best dealers” | “Best” is undefined |
| “customers with no order in 90 days” | “inactive customers” | States the rule |
Name the dataset when it is ambiguous
“revenue from Orders, by month” removes the guesswork when several datasets have a total
column.
Refining
Follow-ups keep context, so you can iterate:
› revenue by month for the last year
› now split it by dealer region
› only EU regions
› make it a stacked bar
Each turn re-runs and shows the updated query. Save to project turns the result into a normal chart — from then on it is bound to its datasets and refreshes with them, with no AI involved.
What it can reach
| Scope | Behaviour |
|---|---|
| Datasets in the current project | Always available |
| Datasets you can access elsewhere | Offered, with the project named |
| Datasets you cannot access | Invisible — not named, not hinted at |
| Restricted datasets | Require an access request first |
Ask cannot widen your permissions
It runs as you. If you cannot open a dataset in the catalogue, no phrasing will surface it here.
Accuracy
Ask is reliable for filtering, aggregating, grouping, and time bucketing within a dataset. It is weaker at:
- Business definitions — “active customer” means whatever your company decided; say the rule.
- Cross-dataset questions — it works within one dataset; joining two needs field mapping, which is coming soon.
- Nested arrays — array fields are not exploded into rows; do that upstream.
Do not publish an Ask result unread
Treat generated queries the way you treat a colleague’s first draft — check the joins, the date column, and the filters before a dashboard depends on it.
Cost and limits
| Plan | Questions / month | History |
|---|---|---|
| Free | 50 | 7 days |
| Pro | 2,000 | 90 days |
| Enterprise | Negotiated | Configurable |
Saved charts do not consume the quota — only asking does.
Question text and the generated query are retained for your history; result rows are not. See Data handling.