CSV / Excel to Parquet Converter

About the CSV / Excel to Parquet Converter

Turn a CSV, TSV, Excel or JSON file into Apache Parquet — the columnar format that DuckDB, Spark, pandas, Polars, BigQuery, Snowflake and data lakes read fastest — without installing Python or uploading the data. Each column gets the type that keeps its values exactly: whole numbers as INT32 or INT64 (IDs of up to 19 digits stay exact), decimals as DOUBLE, true/false as BOOLEAN, ISO dates as DATE, date-times as TIMESTAMP, objects and arrays as JSON, and the rest as STRING; codes with leading zeros such as 007 stay text. Change any type, name or column before writing.

Choose Snappy or Gzip compression (or none) and the row-group size. When the file is written it is read straight back and every value is compared with what was meant to be written, so you know the file is good before you download it. The page also shows the schema, the sizes and each column’s minimum and maximum from the file’s own statistics.

Without a pass, the free preview shows the size of the file and the read-back check, with the first rows of its schema and statistics; the .parquet file and the copies unlock with a Premium pass.

Common uses

• Getting spreadsheet exports into a data lake, a warehouse or DuckDB in a compact, typed format.
• Making Parquet test files with exact types for a pipeline, without writing a script.
• Shrinking a large CSV for sharing: Parquet files are usually much smaller and keep the column types.
• Converting API output (JSON or JSON Lines) to Parquet for analysis.

How the types are chosen

A type is chosen only when every value in the column fits it, in this order: BOOLEAN (true / false), INT32 (whole numbers from −2,147,483,648 to 2,147,483,647), INT64 (larger whole numbers, exact up to 19 digits), DOUBLE (other numbers), DATE (YYYY-MM-DD), TIMESTAMP (YYYY-MM-DD hh:mm[:ss[.ffffff]], with or without a zone), JSON (objects and arrays), else STRING. Whole numbers with a leading zero (007, 0042) stay STRING so no digit is lost. Numbers written with a decimal comma (12,5 or 1.234,50, as in many European files) count as numbers when the file uses them: the separator is worked out from the values and shown under Numbers written as text use, where you can change it. Missing values do not count: every column is OPTIONAL, so they are written as nulls. Overriding a type also reads looser text: 1,234.50, $12, 12.5 %, a decimal comma, yes/no for BOOLEAN, or dates such as 15/03/2026 for DATE (day or month first, as you choose; it starts as your country writes dates). Whole numbers beyond INT64 (above 9,223,372,036,854,775,807) stay STRING, as a DOUBLE would lose their last digits.

Dates and timestamps

DATE columns hold days since 1970-01-01. TIMESTAMP columns hold microseconds since 1970-01-01, as the Parquet specification defines: when the values name their zone (Z, +05:30) they are converted to UTC and marked as instants (isAdjustedToUTC = true); when they do not, they are written as local times (isAdjustedToUTC = false, read the same everywhere — Spark calls this TIMESTAMP_NTZ). You can switch a column between the two; local times turned into instants are taken as UTC. Excel dates and times are read from the workbook’s own values, not from how they are displayed.

Compression and row groups

Snappy is fast to write and read and the most common choice; Gzip (the GZIP codec of the Parquet specification, RFC 1952) usually gives smaller files; None writes the pages as they are. Rows are stored in row groups — 100,000 rows each by default; smaller groups let readers skip more data with the min/max statistics, larger ones compress better. Columns with many repeated values are dictionary-encoded automatically.

The read-back check

After writing, the file is opened again with a Parquet reading library and every value of every column is compared with what was meant to be written — integers exactly, dates as days, timestamps to the microsecond, JSON as text. The result is shown above the schema; a file that did not pass would say which values differ.

Limitations

• Tables that open in the page: CSV and TSV files up to 100 MB, workbooks up to 50 MB, JSON up to 100 MB (up to 2,000,000 records). Everything is held in memory while the file is written.
• Flat columns only: nested JSON is written as JSON text values, not as Parquet structs, lists or maps. There is no DECIMAL type: use DOUBLE, or STRING to keep the digits exactly as written.
• Zstandard, LZ4 and Brotli compression are not offered.

Privacy

On this device: nothing is sent

Everything happens in your browser. What you enter or open here is not uploaded or stored by MySmartCoPilot.