Parquet Viewer

What it's for

Look inside an Apache Parquet file without installing Python, Spark or a database. Parquet is the column-based file format that data tools such as pandas, Polars, DuckDB, Spark and many data lakes write, and it can’t be opened in a text editor or a spreadsheet.

For example, when an export or a pipeline hands you a .parquet file, open it here to check its columns and types, look at the rows, and save them as CSV for a spreadsheet.

How to use it

Choose a Parquet file, or drop one on the page. You’ll see how many rows, columns and row groups it has, its compression and the program that wrote it, then a table of its first 1,000 rows. Type a row number in Show 1,000 rows from row to see another part. Choose a column’s name to sort the rows shown by it, or type in Filter these rows to keep only rows containing that text.

Schema lists every column and its type, with nested columns (lists and structs) indented under their parent. Column sizes gives each stored column’s compression, its size in the file and unpacked, and its number of nulls. Save every row as CSV or as JSON reads the whole file and downloads it: dates are written as 2026-10-23, times in ISO 8601, and lists and structs as JSON text in their CSV cell.

Example

Make a small file with pandas (with pyarrow installed):

python -c "import pandas as pd; pd.DataFrame({'name': ['alice', 'bob'], 'age': [34, 28]}).to_parquet('people.parquet')"

Open people.parquet here. You’ll see 2 rows, 2 columns and 1 row group, compressed with SNAPPY, written by parquet-cpp-arrow, with pandas under other metadata, and a table with alice, 34 and bob, 28. Schema shows name STRING and age INT64.

Good to know

It reads Parquet files compressed with Snappy, gzip, Zstandard, Brotli or LZ4, or not at all. Encrypted Parquet files and the old LZO compression aren’t supported. Decimals are shown as ordinary numbers, so one with more than about 15 significant digits loses its last digits; integers too big for JavaScript are kept exactly.

Sorting and filtering work on the 1,000 rows shown, not the whole file. Saving reads every row into memory, so a file of many millions of rows may be slow or too big for the browser. Only the parts of the file that are needed are read, and nothing is uploaded.