https://parquetkit.com/docs Practical guides on working with Parquet files — no Spark, no installs, everything in the browser. en https://parquetkit.com/docs/append-to-existing-parquet-file https://parquetkit.com/docs/append-to-existing-parquet-file Parquet files are written once, so appending means adding a row group or a new file. fastparquet append=True, a DuckDB rewrite and the dataset-directory pattern. https://parquetkit.com/docs/check-if-parquet-file-is-corrupted https://parquetkit.com/docs/check-if-parquet-file-is-corrupted Validate a Parquet file in your browser: check the footer and schema instantly, then scan every row group with one SQL query. No pyarrow, no Spark. https://parquetkit.com/docs/combine-csv-files-into-one-parquet https://parquetkit.com/docs/combine-csv-files-into-one-parquet Turn a folder of CSV exports into a single compressed Parquet file with one DuckDB command, handling schema drift — or do it entirely in your browser. https://parquetkit.com/docs/compare-two-parquet-files https://parquetkit.com/docs/compare-two-parquet-files Diff two Parquet files by key to find added, removed and changed rows — in your browser, with DuckDB SQL, or with a short pandas script. https://parquetkit.com/docs/convert-excel-xlsx-to-parquet https://parquetkit.com/docs/convert-excel-xlsx-to-parquet Get a spreadsheet into Parquet without pandas: export the sheet as CSV, convert it in your browser, and fix the types Excel mangled on the way out. https://parquetkit.com/docs/convert-folder-of-parquet-files-to-csv https://parquetkit.com/docs/convert-folder-of-parquet-files-to-csv Convert every Parquet file in a directory to its own CSV in one go — with a DuckDB shell loop, PowerShell, a pyarrow script, or a multi-file drop in the browser. https://parquetkit.com/docs/convert-parquet-to-csv-command-line https://parquetkit.com/docs/convert-parquet-to-csv-command-line Three ways to turn a Parquet file into CSV from a terminal — a DuckDB one-liner, Python with pandas or pyarrow — plus a no-install browser fallback. https://parquetkit.com/docs/convert-parquet-to-excel-xlsx https://parquetkit.com/docs/convert-parquet-to-excel-xlsx Produce a real .xlsx workbook from a Parquet file — dates stay dates, types survive — with a DuckDB one-liner or pandas, plus fixes for the two errors you will hit. https://parquetkit.com/docs/duckdb-parquet-viewer https://parquetkit.com/docs/duckdb-parquet-viewer DuckDB can view Parquet files through its CLI, its local web UI, Python, or a browser-based tool. Here is how each works and when to pick one. https://parquetkit.com/docs/fastparquet-install-and-import-errors https://parquetkit.com/docs/fastparquet-install-and-import-errors pip install fastparquet fails to build a wheel, or importing it raises a numpy binary incompatibility. What each error means and how to get to a working reader. https://parquetkit.com/docs/fastparquet-parquetfile-read-columns-filters https://parquetkit.com/docs/fastparquet-parquetfile-read-columns-filters Use fastparquet's ParquetFile to inspect a file, load only the columns you need, prune row groups with filters and iterate in chunks to keep memory flat. https://parquetkit.com/docs/fastparquet-vs-pyarrow https://parquetkit.com/docs/fastparquet-vs-pyarrow fastparquet and pyarrow both read and write Parquet from pandas, but they differ on speed, type support and maintenance. Here is how to choose. https://parquetkit.com/docs/fastparquet-write-options https://parquetkit.com/docs/fastparquet-write-options Write Parquet with fastparquet that other tools read correctly: per-column compression, partition_on with hive layout, row group size, object_encoding and INT96. https://parquetkit.com/docs/find-duplicate-rows-in-parquet https://parquetkit.com/docs/find-duplicate-rows-in-parquet Count exact duplicates, find repeated business keys and keep only the newest row per key — with DuckDB SQL in your browser or a short pandas script. https://parquetkit.com/docs/flatten-nested-parquet-columns https://parquetkit.com/docs/flatten-nested-parquet-columns Parquet struct and list columns become unusable text in CSV. Flatten them properly with DuckDB SQL, or keep the nesting by exporting JSON instead. https://parquetkit.com/docs/large-parquet-to-csv-without-running-out-of-memory https://parquetkit.com/docs/large-parquet-to-csv-without-running-out-of-memory pandas dies converting a multi-GB Parquet file to CSV? Stream it instead — DuckDB COPY, pyarrow batches or Polars sink_csv write the CSV without loading it all. https://parquetkit.com/docs/merge-multiple-parquet-files https://parquetkit.com/docs/merge-multiple-parquet-files Combine several Parquet files into a single dataset with DuckDB SQL in your browser — including files whose schemas don't quite match — or with pyarrow. https://parquetkit.com/docs/open-parquet-file-without-spark https://parquetkit.com/docs/open-parquet-file-without-spark Five ways to open and inspect a Parquet file — from zero-install browser tools to DuckDB one-liners — and when each one makes sense. https://parquetkit.com/docs/open-parquet-in-excel https://parquetkit.com/docs/open-parquet-in-excel Excel cannot read Parquet natively. Convert the file to CSV in your browser — and trim oversized files with SQL first — to get it into a spreadsheet. https://parquetkit.com/docs/pandas-parquet-integer-column-becomes-float https://parquetkit.com/docs/pandas-parquet-integer-column-becomes-float An int column with nulls comes back from pd.read_parquet as float64, and large IDs lose digits. Find whether the file or the reader is at fault, and fix it. https://parquetkit.com/docs/pandas-read-parquet-no-engine-error https://parquetkit.com/docs/pandas-read-parquet-no-engine-error pandas raises ImportError: Unable to find a usable engine when no Parquet reader is installed. What actually causes it, and how to fix it for good. https://parquetkit.com/docs/parquet-column-data-type-mismatch https://parquetkit.com/docs/parquet-column-data-type-mismatch Spark or Databricks fails reading a Parquet dataset when one file's column type changed. Here is how to find the drifted file and column, and fix it. https://parquetkit.com/docs/parquet-compression-snappy-vs-gzip-vs-zstd https://parquetkit.com/docs/parquet-compression-snappy-vs-gzip-vs-zstd Snappy, Gzip, and Zstd trade file size for CPU time differently. Benchmarks, defaults by engine, and how to check or change the codec in an existing Parquet file. https://parquetkit.com/docs/parquet-int96-timestamps https://parquetkit.com/docs/parquet-int96-timestamps Timestamps shifted by hours, wrong by centuries, or rejected outright? Your Parquet file probably uses legacy INT96. What it is and how to deal with it. https://parquetkit.com/docs/parquet-magic-bytes-not-found https://parquetkit.com/docs/parquet-magic-bytes-not-found The PAR1 magic bytes error means your Parquet file is truncated or not Parquet at all. Here is how to diagnose it in seconds and where the file broke. https://parquetkit.com/docs/parquet-row-group-statistics-min-max https://parquetkit.com/docs/parquet-row-group-statistics-min-max Read the per-row-group min/max and null counts stored in a Parquet footer with DuckDB's parquet_metadata(), and use them to explain why a filter is fast or slow. https://parquetkit.com/docs/parquet-to-csv-null-and-timestamp-formatting https://parquetkit.com/docs/parquet-to-csv-null-and-timestamp-formatting Your exported CSV has NaN, shifted timestamps or mangled accents. How to control null strings, date formats, delimiters and encoding when writing CSV. https://parquetkit.com/docs/parquet-to-duckdb-or-sqlite-database https://parquetkit.com/docs/parquet-to-duckdb-or-sqlite-database Turn Parquet files into a queryable .duckdb or .sqlite database file — with the right commands, the type traps, and when to skip the import entirely. https://parquetkit.com/docs/parquet-to-json-not-json-serializable https://parquetkit.com/docs/parquet-to-json-not-json-serializable json.dumps fails on Timestamp, Decimal, bytes and int64 values read from Parquet. Why each type breaks, and three ways to get valid JSON out. https://parquetkit.com/docs/parquet-to-jsonl-for-llm-fine-tuning https://parquetkit.com/docs/parquet-to-jsonl-for-llm-fine-tuning Fine-tuning APIs expect JSONL, but datasets ship as Parquet. Filter, reshape and export training data as JSONL — in the browser or with one DuckDB command. https://parquetkit.com/docs/parquet-to-nested-json https://parquetkit.com/docs/parquet-to-nested-json Turn flat Parquet rows into nested JSON — orders with an items array, customer objects — using DuckDB's list() and struct syntax. Includes JSON array and JSONL output. https://parquetkit.com/docs/parquet-vs-csv https://parquetkit.com/docs/parquet-vs-csv A practical comparison of Parquet and CSV — file size, query speed, type safety and tool support — with concrete guidance on choosing. https://parquetkit.com/docs/print-parquet-schema-without-installing https://parquetkit.com/docs/print-parquet-schema-without-installing See a Parquet file's columns, types and row count in seconds — in your browser, with no pyarrow, Spark or parquet-tools. Works for multi-GB files. https://parquetkit.com/docs/query-parquet-on-s3-without-downloading https://parquetkit.com/docs/query-parquet-on-s3-without-downloading Read Parquet straight from S3, GCS or any HTTPS URL with DuckDB httpfs — credentials, globs, partition pruning, and how to keep the bytes you pay for down. https://parquetkit.com/docs/query-parquet-with-sql-in-browser https://parquetkit.com/docs/query-parquet-with-sql-in-browser Run real SQL against local Parquet and CSV files using DuckDB in the browser: joins, aggregates and window functions with zero setup. https://parquetkit.com/docs/read-parquet-key-value-metadata https://parquetkit.com/docs/read-parquet-key-value-metadata Every Parquet footer stores key-value metadata: the writer, the pandas index, Spark's schema, custom tags. Read it with DuckDB SQL or pyarrow, and write your own. https://parquetkit.com/docs/read-partitioned-parquet-dataset https://parquetkit.com/docs/read-partitioned-parquet-dataset Query a directory of Hive-partitioned Parquet files with DuckDB SQL, filter by partition columns, and understand how partition pruning works. https://parquetkit.com/docs/rename-columns-in-parquet-file https://parquetkit.com/docs/rename-columns-in-parquet-file Parquet files can't be edited in place. Rename, drop, reorder or retype columns by rewriting the file with one DuckDB statement or a few lines of pyarrow. https://parquetkit.com/docs/split-large-parquet-file https://parquetkit.com/docs/split-large-parquet-file Split an oversized Parquet file by target size, by row count, or by a column value — with DuckDB SQL, pyarrow, or in the browser with no install.