csvQL

The problem

A lot of real data lives in CSV exports and Excel workbooks. Answering a quick question about one usually means opening a spreadsheet app and clicking through filters, or writing a throwaway script. I wanted to point SQL at the file and ask.

What I built

csvQL is a terminal app and CLI that loads CSV files, whole directories of them, or every sheet of an .xlsx workbook as tables, then runs SQL against them. The interactive mode is a full-screen terminal UI: a data grid with virtual scrolling, a query bar with autocomplete for tables, columns and values, and a side panel that shows the selected cell, the inferred schema and the loaded tables. One-off queries also run straight from the command line.

Key decisions

• A real SQL parser, not string matching. Queries go through a SQL parser, and WHERE clauses are evaluated over its syntax tree: comparisons, AND/OR/NOT, IS NULL and BETWEEN.
• Types are inferred, then respected. Each column’s type is inferred from a sample of its values, so amount > 100 compares numbers rather than strings.
• Messy headers get cleaned up. Header rows are detected automatically (or chosen with --header-row), and header names are turned into valid SQL identifiers.
• Reshaping built in. Spreadsheets often spread dates across columns. :unpivot turns those columns into rows, so they can be filtered like any other data, and :reset puts them back.

What I’d do next

• Add ORDER BY, GROUP BY with aggregates, and joins across files.
• Write query results back out to CSV.

← All projects