Open Parquet files and export exact values to CSV, JSON or Excel
Bring a local Parquet file into a precise, private data workbench. Inspect its schema, choose a manageable slice, and keep long identifiers, decimals, nulls and nested values intact when you export.
Parquet Desk
The precision example is a real Parquet file generated independently with Apache Arrow. It uses the same local reader and export path as your file.
Supported scope: One unencrypted local .parquet file up to 256 MiB; footer up to 4 MiB, 128 top-level columns and 1,000 row groups. Each scan: up to 64 columns, 20,000 rows, 500,000 output cells, 1,000,000 physical values and 64 MiB declared uncompressed data (32 MiB per selected row group, 16 MiB per page). Supported codecs: uncompressed, Snappy, Gzip, Brotli, Zstd and LZ4_RAW. No INT96, maps, variants, legacy list layouts, dotted/reserved field names, encrypted/external columns, remote URLs, joins or CSV-to-Parquet writing. XLSX is omitted when any selected cell exceeds 32,767 characters; CSV/JSON remain available. Exports are limited to 24 MiB.
Your local columnar file
Required fields are marked with an asterisk.
Schema and selection workbench
Review your output, then save it to your device.
Make the next step clear.
Your findings, assumptions and next steps will appear here after you run the check.
Parquet Desk · https://columnar.kitforma.com/ · KitForma
Exact integers and decimals
Column and row-group selection
No account or data upload
Understand the evidence.
Focused guides explain the rules, the edge cases and the checks worth doing before you act.
Inspect a Parquet schema before exporting rows
Use column types, row groups and a declared scan window to choose a useful slice.
Read the guide ↗Field guide / 02Export Parquet numbers without losing identifiers or decimal precision
Keep INT64, UINT64, scaled decimals and temporal units explicit across formats.
Read the guide ↗Field guide / 03Check nested values, nulls and export boundaries
Validate a Parquet selection without flattening away structure or claiming unsupported recovery.
Read the guide ↗Before you start
Will an identifier such as 9007199254740993 be rounded?
No. INT64 and UINT64 are decoded as BigInt and exported as decimal strings. DECIMAL uses its signed unscaled integer and declared scale, so precision does not pass through JavaScript Number. XLSX uses text cells. CSV importers can still infer numeric types; import those columns as text.
Does the filter search the entire file?
Only when your selected row window covers the entire file within the supported limits. The result reports physical rows scanned and rows matched. Numeric comparisons use exact decimal-string arithmetic; text comparisons are case-sensitive. The tool does not imply that a 1,000-row preview is a whole-dataset query.
Why do timestamps look like integers?
This release preserves the original day, millisecond, microsecond or nanosecond values. The schema records units and UTC-adjustment semantics. It does not silently round timestamps to JavaScript millisecond dates or invent a timezone for local timestamps.
What happens to lists, structs, binary data and nulls?
Supported standard lists and structs remain nested JSON values; they are not multiplied into extra rows. Binary data is hex text beginning with 0x. JSON keeps null distinct from an empty string or empty list. CSV blanks can look identical, so keep JSON/schema when those distinctions matter.
Why can a small file exceed the scan limit?
Compressed size is not decoded size. A selected row group can contain many values or a large dictionary even when you ask for only a few rows. The workbench checks declared column/page sizes and value counts before decoding. Reduce columns or choose a smaller row group when possible.
Are all Parquet files and Excel cells supported?
No. The supported scope and codecs are listed before upload. Encrypted/external columns, maps, variants, INT96 and some legacy layouts are refused rather than guessed. If an output cell exceeds Excel’s 32,767-character limit, the workbook is omitted with an explanation; complete bounded CSV and JSON are still offered.
Sources & assumptions
Formats, algorithms and supported features are documented below. Check the stated scope and inspect your output before using it.
- Apache Parquet: File format ↗
Parquet stores column chunks within row groups, with file metadata in the footer.
- Apache Parquet: Logical types ↗
Integer, DECIMAL and temporal annotations describe the meaning of physical values.
- Hyparquet reader ↗
MIT-licensed reader. This product bypasses lossy scalar conversions and applies its own tested exact conversion.
- Hyparquet compressors ↗
MIT-licensed codec package; this product enables only its independently fixture-tested codec set.
- Apache Arrow: Parquet reading and writing ↗
Independent PyArrow fixtures verify real Parquet decoding and row-group behavior.
- SheetJS: Cell objects ↗
Excel exports write literal text cells, never executable formulas.
Reviewed 2026-10-18. Read our methodology or report an issue.