CSV to Parquet Converter
Free CSV to Parquet converter. Paste CSV and download a real .parquet file with inferred column types and uncompressed plain encoding.
Loading tool...
What is CSV to Parquet Converter?
The CSV to Parquet Converter writes a genuine Apache Parquet file from pasted CSV, entirely in your browser. Column types are inferred per column — all-numeric columns become DOUBLE, everything else UTF8 text — and the output uses uncompressed plain encoding with a single row group, the most widely readable Parquet subset. The exact byte size is reported before download, and the writer is tested by reading its own output back with a Parquet reader. Nothing is uploaded; the bytes are built locally and handed straight to your downloads folder.
Key Benefits
- Standard-conformant Parquet that opens in every reader
- Automatic per-column type inference, no schema drafting
- Quoted-field CSV parsing for messy exports
- Exact byte size shown before you download
- Completely local — confidential data stays in the browser
Common Use Cases
- •Feeding spreadsheet exports into analytics or notebook tools
- •Producing sample .parquet fixtures for tests and demos
- •Teaching how Parquet schemas and row groups are laid out
- •Converting small reference tables for a columnar workflow
- •Creating files for the site Parquet viewer without any install
How to Convert CSV to Parquet
- Paste the CSV: Paste your data with the header row included. Quoted fields containing commas or newlines are handled.
- Convert: Click Convert to Parquet. Each column is inferred as DOUBLE when every value is numeric, otherwise UTF8 text, and the file is written with uncompressed plain encoding.
- Check the summary: The result cards show row count, column count and the exact byte size of the generated file before you download anything.
- Download the file: Click Download .parquet to save the file. It opens in the site Parquet viewer, pandas, duckdb, spark or any standard reader.
Key Features
- Real Parquet output: thrift metadata, row group and data pages
- Per-column type inference between DOUBLE and UTF8
- Quoted-field CSV parsing with comma and newline support
- Round-trip verified output readable by standard Parquet tools
- Fully local conversion with one-click download
Frequently Asked Questions
Is the output a real Parquet file?
Yes. The writer emits a full thrift compact-protocol footer with the schema tree, one row group and one plain-encoded data page per column, wrapped in PAR1 magic. The test suite reads the output back with a standard Parquet reader and compares every value, so compatibility is verified rather than assumed.
How are column types decided?
Each column is checked after the header row: if every value parses as a finite number the column is written as DOUBLE, otherwise as UTF8 text. A single non-numeric value makes the whole column text, which keeps files like postal codes or phone numbers readable instead of lossy.
Why is the file uncompressed?
Uncompressed plain encoding is the most widely supported Parquet subset — every reader from 2013 onward opens it. CSV data converts to roughly its own text size, and any downstream tool can re-compress with snappy or zstd if smaller files matter.
How large a CSV can it handle?
Comfortably into the tens of megabytes, since parsing and writing are linear passes over in-memory strings. For very large exports, split the CSV and convert the parts, or use a streaming local tool — the schema this tool infers tells you exactly what to expect.