CSV to Parquet Converter
Paste or upload a CSV file and convert it straight to Apache Parquet, the columnar format used by data warehouses, Spark, DuckDB, and pandas. The converter parses your CSV, samples each column to infer whether it's a string, number, or boolean, and writes a real .parquet file you can download right away, with no upload, no Python, and no server round trip.
Reviewed by Vincenzo Manto, Founder & Lead Developer · Last updated September 27, 2026
How it works
- 1
Paste your CSV text or upload a .csv file.
- 2
Pick the delimiter that matches your data (comma, semicolon, tab, or pipe).
- 3
Click Convert; each column is sampled to infer string, double, or boolean types.
- 4
Download the resulting .parquet file, written locally with hyparquet-writer.
Why convert CSV to Parquet?
CSV is universal but inefficient: it stores everything as text, has no schema, and every reader has to guess types on load. Parquet stores data by column, compresses it well, and carries an explicit schema, which is what analytics engines like DuckDB, Spark, and cloud warehouses expect for fast, cheap queries.
No install required
Normally producing a parquet file means writing a Python script with pandas and pyarrow, or spinning up a Spark job. This tool does the same conversion with pure JavaScript in your browser tab, using the hyparquet-writer library, so you get a real, standards-compliant .parquet file in seconds.
Frequently asked questions
How are column types decided?
Each column's non-empty values are sampled: if every value looks like true/false it's written as BOOLEAN, if every value is numeric it's written as DOUBLE, and otherwise it falls back to STRING so no data is lost.
What happens to empty cells?
Blank CSV cells are written as parquet nulls rather than empty strings, and are ignored when inferring a column's type.
Is my CSV data uploaded anywhere?
No. Both the CSV parsing and the Parquet file writing happen in your browser using the hyparquet-writer library, and nothing is sent to a server.
What compression does the output use?
hyparquet-writer compresses column chunks with Snappy by default, which is widely supported by Spark, DuckDB, pandas, and most other parquet readers.
Can I preview the resulting Parquet file?
Parquet is a binary columnar format, so there's no meaningful text preview to show. You get the row and column count plus an instant download, and you can drop the result into the Parquet Viewer tool to inspect it.