Run SQL on a CSV file in your browser

Open a file and run SELECT country, max(co2) AS peak FROM data GROUP BY 1 ORDER BY 2 DESC LIMIT 10. Column names come from the header; quote them with double quotes if they contain spaces, like "Total revenue". Anything typed into the bar that starts with SELECT, WITH or FROM runs as SQL; anything else is treated as a filter.

It is full DuckDB SQL: CTEs, window functions, GROUP BY ALL, QUALIFY, regex functions, date arithmetic, UNNEST for nested JSON, PIVOT. The same dialect you'd use in the DuckDB CLI, so a query that works here works in a script later.

The same bar takes a filter language: country:NL, amount:>500, year:2010..2020, -status:ok, sku:AB*, free words that search every text column, OR and parentheses. Click a header to sort, shift-click for a second key, hide columns, and the column panel shows non-null counts, distinct values, min, max and the top values.

Press the SQL toggle and the current filter, sort and visible columns are rewritten as a SELECT you can keep editing. It is the fastest way to learn what a filter compiles to.

One file per page, up to 1.5 GB, read into the memory of your tab. Above that a CSV loads partially and a note says how many rows made it. Results are capped at 1,000,000 rows and a badge says first 1M when the cap hits; aggregate or LIMIT to stay under it.

There is one table, data, so joins across two files aren't possible here. Self-joins, subqueries and CTEs over that table are fine. Queries run in a Web Worker, so a slow GROUP BY doesn't freeze the page.

Export writes the query result: CSV with a header, Parquet with zstd compression, a JSON array, JSONL, or a Markdown table of the first 1,000 rows. The CO2 sample is 13.7 MB as CSV and 4.3 MB as Parquet.

Press Share to upload the file once. On the hosted page the query, sort and visible columns live in the URL after #, so the link you copy opens on the same result for whoever gets it. Anonymous links take files up to 100 MB and are deleted 7 days after creation; a free account keeps 10.

Files over 1.5 GB, joins across several files, or writing results back to disk: install DuckDB and run duckdb -c "SELECT count(*) FROM read_csv('data.csv')". It is the same engine and the same SQL, minus the browser's memory ceiling.

For a dataset that updates every night, push it instead: curl -T data.csv https://rows.page, then push again to the same link and the page shows how many rows are new, gone or changed, with the same SQL bar on top.