FAQ

The ten things people write to me about, answered from the code rather than from hope.

How big a file can I open?

There is no fixed limit; the ceiling is the memory your browser tab is allowed. Files up to a few hundred megabytes are routine on a laptop. Past 100 MB the app asks first: This file is large (250MB). Processing may be slow. Continue? Parquet is the format to reach for at that size, because DuckDB reads only the columns a query touches. A load that runs longer than 60 seconds is stopped with Loading is taking too long. Try a smaller file or filter it externally first. PDFs, images and database files get 180 seconds, since OCR is slow.

Where does my data go?

Nowhere. The file is read into DuckDB running inside your tab, and the badge in the header says exactly that: Connected. Your file never leaves this tab. Files, pipelines and settings are saved to your browser's IndexedDB so a refresh does not lose them; the Browser Storage section of Settings clears it. The site records anonymous usage events, which operation was applied or which size bucket a file fell into, never a cell value. If you add an AI key, the model sees your column names, types and row, null and distinct counts, not your rows. Up to five example values per column are added only if you turn that option on in Settings.

My Excel file has several sheets. Which one loads?

You choose. A workbook with more than one non-empty sheet opens a Select sheets to load dialog listing each sheet with its columns. Every sheet you tick becomes its own table with its own tab. Sheets with only a header, or nothing at all, are skipped, and a workbook with no data anywhere is refused with Excel file has no non-empty sheets. Both .xlsx and .xls work.

Accented letters came out as garbage.

The file is not UTF-8. The app checks the bytes first: a UTF-16 file is recognised by its byte-order mark, and anything else that fails a strict UTF-8 scan is read as Windows-1252, with the warning This file is not valid UTF-8, so it was read as Windows-1252 (Latin). Check that accented characters look right. That guess is right for most Excel exports. When it is wrong, click the gear on the file's tab, Configure CSV parsing, and pick the Encoding yourself: UTF-16, Latin-1, Shift-JIS, GB2312 and EUC-KR are all there. Try semicolon-eu.csv from the samples for a German file with umlauts and a semicolon delimiter.

My dates were read wrong, or stayed as text.

DuckDB decides the format once per column from a sample of rows. A column holding both 2024-01-05 and 05/06/2024 has no single format, so it stays text. So does a column with an impossible date like 2024-13-45 in it. Fix it with Convert Type to date: auto-detect handles the common layouts, and for day-first files you clear the checkbox and type %d/%m/%Y under Date format in the data. A value that still will not parse becomes empty rather than failing the step.

Did I lose my leading zeros?

Not here. When a column carries values like 00001 or 03301, DuckDB keeps the whole column as text, even when a few rows lack the zero. The leading-zeros.csv sample loads with every zip code and account number intact. If a column did become a number, open Configure CSV parsing from the gear on the file tab and tick Load all columns as text under Type Detection. The tool that eats leading zeros is Excel, on the way back in.

Can I share a dashboard?

Yes, without uploading anything. Switch to Canvas and click Share link. The whole state travels in the URL: filters, the breakdown column, the panel layout, the pivot. When the data itself compresses to under 200 KB it rides along too and the dialog says Link includes the data, so the recipient sees the dashboard with no file at all. For one of the site's own sample files the link carries the sample path instead. Anything larger sends the layout only, and the recipient drops the same file in to fill it.

What can I export?

The Export / download button writes the current grid, with every pipeline step applied, as CSV, TSV, JSON, JSON Lines, XML, Parquet, Excel, PDF, HTML or a DuckDB database file. The Canvas has its own export with SQL (CREATE plus INSERT statements), Markdown and PNG added. Copy to clipboard needs an https page; on plain http you get Clipboard API is not available (requires HTTPS).

Does it work offline?

Once the app has loaded, yes. Pull the network and the badge flips to Offline. Tools keep working., because nothing you do afterwards needs a server. The app shell is cached for return visits, but the DuckDB engine is fetched from a CDN on first load, so open the app once while online before relying on it in a train tunnel.

Can I use it on my phone?

It runs in a mobile browser. Below 1024 pixels wide the Pipeline panel turns into a drawer that slides over the grid. Two honest caveats: a phone gives a tab far less memory than a laptop, so the file-size ceiling is lower, and building a condition with three clauses is nicer with a mouse. Reading a shared dashboard on a phone is fine.