Analyze without SQL

Open a workbook and choose Analyze. Every operation runs in DuckDB through a validated, typed plan; your browser receives bounded results rather than downloading source tables. The tools work even on a server where SQL is disabled.

Summaries and pivots

Choose a source, up to four group columns, and a measure: count, sum, average, minimum or maximum. Count counts rows; the other aggregates ignore NULL inputs. Formula columns can be grouped and measured like any other column.

A report extends a summary with up to eight measures (revenue = sum(amount), orders = count()), up to four calculated measures written as formulas over the measures (average = [@revenue]/[@orders]), optional subtotals per group level plus a grand total, and up to eight filters.

Optionally pick a pivot column and list up to 24 explicit categories, one per line — (NULL) selects missing values. Unlisted categories are excluded, and the result headers name each category. Calculated measures and a pivot cannot be combined.

Previews contain at most 500 rows; a truncation notice tells you more groups exist. The aggregates themselves always use the full source — the limit applies only to the output.

Drill-down. Every summary or report row has Show rows: it opens the source table in the grid, filtered to that row’s group values, paged like any view.

Charts

A saved analysis keeps a chart: bars (the default), a line, or points of the first measure over the groups, at most the first 100 groups, with a label when the chart is partial. A list beside the chart gives every exact value, so the bars are never your only copy of a number.

Saved analyses can be laid out as tiles on a dashboard, with slicers that filter every report on it over its table. A workbook holds up to 32 dashboards of up to 12 tiles and 6 slicers each.

Joins

A join takes one equality key from each source and is inner or left. Key SQL types must match exactly — convert incompatible types first — and NULL keys do not match. Output columns are prefixed left_ and right_, up to 100 of them.

Before you materialize anything, a diagnostic table reports repeated non-NULL keys on each side and the complete output row count, computed from key frequencies — so many-to-many blow-ups are visible without running the full join. The preview stays within 500 rows.

Column profiles

Profile up to 32 columns at once: declared SQL type, total rows, NULL count, exact distinct non-NULL values, and duplicate occurrences. Profiling and grouping may scan the full source; Cancel run stops the active request at any time.

Clean-up tools, previewed first

Each tool shows a preview — the first 50 rows, the row count, and how many rows it would drop, change or fail to convert — and then applies as a new table. The source is never modified, so undoing means deleting the new table.

Tool	What it does
Deduplicate	By key columns; the first row is kept, in row order
Trim	Removes surrounding spaces, optionally inner runs too
Split	One column by a delimiter into up to 16 new columns
Convert	To BIGINT, DOUBLE, DATE, BOOLEAN or VARCHAR; values that do not convert become blank

Tables can also carry validation rules (one of a list, a number between bounds, not blank) that every write path enforces — typing, pasting, and even SQL changes that would add violations are refused.

Saving and refresh

Save analysis persists the plan and chart choice; opening it reruns it on current data, and workbook changes mark a displayed result stale rather than silently rerunning expensive work. Create result table snapshots the result (up to 10,000 rows) as a table that stays unchanged until you explicitly refresh it.

Live selection statistics

Select any cells and the bar under the grid shows the live sum, average and count of the selection — no menu, no formula.

The full specification, including every limit, is in the repository’s analysis guide.

The SQL workspace — when you do want to write the query yourself
Big tables
Importing and exporting data