Query CSV files with SQL — without a database.

The idea

A CSV file is a table; SQL is the language for asking tables questions. Usually you would have to create a database, define a table and import the file first. In Intact CSV Editor the query console treats every open document as a table already — named after its file, customers.csv as customers — and copies in only the tables a statement names.

Your first query

1. Open the file — a CSV, a workbook, a Parquet file, anything the app reads.
2. Press ⇧⌘R to open the query window. The sidebar lists the tables and their columns.
3. Type a statement and press ⌘R:

   SELECT city, COUNT(*) AS customers, SUM(revenue_chf) AS revenue
   FROM customers
   GROUP BY city
   ORDER BY revenue DESC;

4. Use the result: replace the table with it, export it to any format, or open it as a new document.

Columns that look numeric become real numbers for the query, so 1'234.50 and 987,25 compare and add up correctly — while the file itself keeps them exactly as written.

Joining two files

SELECT c.name, o.order_id, o.amount
FROM customers c
JOIN orders o USING (customer_no)
WHERE o.date >= '2026-01-01';

A file that is not open can be named by its path: SELECT * FROM "@file:/Exports/archive.csv".

MySQL, on SQLite

The engine underneath is SQLite; on top of it the dialect is MySQL's. Backtick identifiers, # comments, SHOW TABLES, DESCRIBE, GROUP_CONCAT(… SEPARATOR …), DATE_FORMAT, STR_TO_DATE, REGEXP_REPLACE, JSON_EXTRACT and most of MySQL's function library work as they do there, and a MySQL dump can be pasted in as it stands. Switch MySQL mode off in Settings for plain SQLite.

Regular expressions — RLIKE, REGEXP_LIKE, REGEXP_REPLACE — are compiled once per query rather than once per row: a grouped query with RLIKE over a million rows takes about 0.9 seconds. A pattern that can take exponential time, a repeat inside a repeat such as (a+)+b, is refused with a message saying why, instead of freezing the window on the first long value.

Views and long results

Save a query under a name and it becomes a table of its own — SELECT * FROM open_items — which remembers the files it is built from and can be opened as a document that is always current. Results longer than the console shows are read a page at a time, 101–200 of 4,318, without rewriting your statement.

Other ways to do it

On the command line, sqlite3 can import a CSV with .import, csvkit's csvsql runs SQL on CSV files, and DuckDB reads CSV and Parquet directly. They are excellent in a terminal; Intact CSV Editor puts the same questions next to the grid, where you can see and fix the data the answer came from.