• fix: csvstat no longer errors when a column is a time delta and --json is set.

2.0.0 - May 1, 2024

This is the first major release since December 27, 2016. Thank you to all contributors, including 44 new contributors since 1.0.0!

Want to use csvkit programmatically? Check out agate, used internally by csvkit.


  • csvclean now writes its output to standard output and its errors to standard error, instead of to basename_out.csv and basename_err.csv files. Consequently:

    • The --dry-run option is removed. The --dry-run option changed error output from the CSV format used in basename_err.csv files to a prosaic format like Line 1: Expected 2 columns, found 3 columns.

    • Summary information like No errors., 42 errors logged to basename_err.csv and 42 rows were joined/reduced to 24 rows after eliminating expected internal line breaks. is not written.

  • csvclean no longer reports or fixes errors by default; it errors if no checks or fixes are enabled. Opt in to the original behavior using the --length-mismatch and --join-short-rows options. See new options below.

  • csvclean no longer omits rows with errors from the output. Opt in to the original behavior using the --omit-error-rows option.

  • csvclean joins short rows using a newline by default, instead of a space. Restore the original behavior using the --separator " " option.

In brief, to restore the original behavior for csvclean:

csvclean --length-mismatch --omit-error-rows --join-short-rows --separator " " myfile.csv

Other changes:

  • feat: csvclean adds the options:

    • --length-mismatch, to error on data rows that are shorter or longer than the header row

    • --empty-columns, to error on empty columns

    • --enable-all-checks, to enable all error reporting

    • --omit-error-rows, to omit data rows that contain errors, from standard output

    • --label LABEL, to add a “label” column to standard error

    • --header-normalize-space, to strip leading and trailing whitespace and replace sequences of whitespace characters by a single space in the header

    • --join-short-rows, to merge short rows into a single row

    • --separator SEPARATOR, to change the string with which to join short rows (default is newline)

    • --fill-short-rows, to fill short rows with the missing cells

    • --fillvalue FILLVALUE, to change the value with which to fill short rows (default is none)

  • feat: The --quoting option accepts 4 (csv.QUOTE_STRINGS) and 5 (csv.QUOTE_NOTNULL) on Python 3.12.

  • feat: csvformat: The --out-quoting option accepts 4 (csv.QUOTE_STRINGS) and 5 (csv.QUOTE_NOTNULL) on Python 3.12.

  • fix: csvformat: The --out-quoting option works with 2 (csv.QUOTE_NONUMERIC). Use the --locale option to set the locale of any formatted numbers.

  • fix: csvclean: The --join-short-rows option no longer reports length mismatch errors that were fixed.

1.5.0 - March 28, 2024

  • feat: Add support for Zstandard files with the .zst extension, if the zstandard package is installed.

  • feat: csvformat adds a --out-asv (--A) option to use the ASCII unit separator and record separator.

  • feat: csvsort adds a --ignore-case (--i) option to perform case-independent sorting.

1.4.0 - February 13, 2024

  • feat: csvpy adds the options:

    • --no-number-ellipsis, to disable the ellipsis () if max precision is exceeded, for example, when using table.print_table()

    • --sniff-limit`

    • --no-inference`

  • feat: csvpy removes the --linenumbers and --zero output options, which had no effect.

  • feat: in2csv adds a --reset-dimensions option to recalculate the dimensions of an XLSX file, instead of trusting the file’s metadata. csvkit’s dependency agate-excel 0.4.0 automatically recalculates the dimensions if the file’s metadata expresses dimensions of “A1:A1” (a single cell).

  • fix: csvlook only reads up to --max-rows rows instead of the entire file.

  • fix: csvpy supports the existing input options:

    • --locale

    • --blanks

    • --null-value

    • --date-format

    • --datetime-format

    • --skip-lines

  • fix: csvpy: --maxfieldsize no longer errors when --dict is set.

  • fix: csvstack: --maxfieldsize no longer errors when --no-header-row isn’t set.

  • fix: in2csv: --write-sheets no longer errors when standard input is an XLS or XLSX file.

  • Update minimum agate version to 1.6.3.

1.3.0 - October 18, 2023

  • csvformat adds a --skip-header (-E) option to not output a header row.

  • csvlook adds a --max-precision option to set the maximum number of decimal places to display.

  • csvlook adds a --no-number-ellipsis option to disable the ellipsis () if --max-precision is exceeded. (Requires agate 1.9.0 or greater.)

  • csvstat supports the --no-inference (-I), --locale (-L), --blanks, --date-format and datetime-format options.

  • csvstat reports a “Non-null values” statistic (or a nonnulls column when --csv is set).

  • csvstat adds a --non-nulls option to only output counts of non-null values.

  • csvstat reports a “Most decimal places” statistic (or a maxprecision column when --csv is set).

  • csvstat adds a --max-precision option to only output the most decimal places.

  • csvstat adds a --json option to output results as JSON text.

  • csvstat adds an --indent option to indent the JSON text when --json is set.

  • in2csv adds a --use-sheet-names option to use the sheet names as file names when --write-sheets is set.

  • feat: Add a --null-value option to commands with the --blanks option, to convert additional values to NULL.

  • fix: Reconfigure the encoding of standard input according to the --encoding option, which defaults to utf-8-sig. Affected users no longer need to set the PYTHONIOENCODING environment variable.

  • fix: Prompt the user if additional input is expected (i.e. if no input file or piped data is provided) in csvjoin, csvsql and csvstack.

  • fix: No longer errors if a NUL byte occurs in an input file.

  • Add Python 3.12 support.

1.2.0 - October 4, 2023

  • fix: csvjoin uses the correct columns when performing a --right join.

  • Add SQLAlchemy 2 support.

  • Drop Python 3.7 support (end-of-life was June 5, 2023).

1.1.1 - February 22, 2023

  • feat: csvstack handles files with columns in different orders or with different names.

1.1.0 - January 3, 2023

  • feat: csvsql accepts multiple --query command-line arguments.

  • feat: csvstat adds --no-grouping-separator and --decimal-format options.

  • Add Python 3.11 support.

  • Drop Python 3.6 support (end-of-life was December 23, 2021).

  • Drop Python 2.7 support (end-of-life was January 1, 2020).

1.0.7 - March 6, 2022

  • fix: csvcut extracts the correct columns when --line-numbers is set.

  • fix: Restore Python 2.7 support in edge cases.

  • feat: Use 1024 byte sniff-limit by default across csvkit. Improve csvstat performance up to 10x.

  • feat: Add support for .xz (LZMA) compressed input files.

  • Add Python 3.10 support.

  • Drop Python 3.5 support (end-of-life was September 30, 2020).

1.0.6 - July 13, 2021



  • csvformat creates default headers when --no-header-row is set, as documented.

  • csvstack no longer errors when --no-header-row is combined with --groups or --filenames.

1.0.5 - March 2, 2020


  • Drop Python 3.4 support (end-of-life was March 18, 2019).


  • Output error message for memory error even if not --verbose.


  • Fix regression in 1.0.4, which caused numbers like 4.5 to be parsed as dates.

  • in2csv Fix error reporting if --names used with non-Excel file.

1.0.4 - March 16, 2019


  • Drop Python 3.3 support (end-of-life was September 29, 2017).


  • csvsql adds a --chunk-size option to set the chunk size when batch inserting into a table.

  • csvkit is tested against Python 3.7.


  • --names works with --skip-lines.

  • Dates and datetimes without punctuation can be parsed with --date-format and datetime-format.

  • Error messages about column indices use 1-based numbering unless --zero is set.

  • csvcut no longer errors on --delete-empty-rows with short rows.

  • csvjoin no longer errors if given a single file.

  • csvsql supports UPDATE commands.

  • csvstat no longer errors on non-finite numbers.

  • csvstat respects all command-line arguments when --count is set.

  • in2csv CSV-to-CSV conversion respects --linenumbers when buffering.

  • in2csv writes XLS sheets without encoding errors in Python 2.

1.0.3 - March 11, 2018


  • csvgrep adds a --any-match (-a) flag to select rows where any column matches instead of all columns.

  • csvjson no longer emits a property if its value is null.

  • csvjson adds --type and --geometry options to emit non-Point GeoJSON features.

  • csvjson adds a --no-bbox option to disable the calculation of a bounding box.

  • csvjson supports --stream for newline-delimited GeoJSON.

  • csvsql adds a --unique-constraint option to list names of columns to include in a UNIQUE constraint.

  • csvsql adds --before-insert and --after-insert options to run commands before and after the INSERT command.

  • csvpy reports an error message if input is provided via STDIN.

  • in2csv adds a --encoding-xls option to specify the encoding of the input XLS file.

  • in2csv supports --no-header-row on XLS and XLSX files.

  • Suppress agate warning about column names not specified when using --no-header-row.

  • Prompt the user if additional input is expected (i.e. if no input file or piped data is provided).

  • Update to agate-excel 0.2.2, agate-sql 0.5.3.


  • csvgrep accepts utf-8 arguments to the --match and --regex options in Python 2.

  • csvjson streams input and output only if --snifflimit is 0.

  • csvsql sets a DECIMAL’s precision and scale and a VARCHAR’s length to avoid dialect-specific errors.

  • csvstack no longer opens all files at once.

  • in2csv respects --no-header-row when --no-inference is set.

  • in2csv CSV-to-CSV conversion streams input and output only if --snifflimit is 0.

  • in2csv supports GeoJSON files with: geometry set to null, missing Point coordinates, altitude coordinate values.

csvkit is no longer tested on PyPy.

1.0.2 - April 28, 2017


  • Add a --version flag.

  • Add a --skip-lines option to skip initial lines (e.g. comments, copyright notices, empty rows).

  • Add a --locale option to set the locale of any formatted numbers.

  • Add a --date-format option to set a strptime date format string.

  • Add a --datetime-format option to set a strptime datetime format string.

  • Make --blanks a common argument across all tools.

  • -I is the short option for --no-inference.

  • csvclean, csvformat, csvjson, csvpy support --no-header-row.

  • csvclean is faster and no longer requires exponential time in the worst case.

  • csvformat supports --linenumbers and –zero (no-op).

  • csvjoin supports --snifflimit and --no-inference.

  • csvpy supports --linenumbers (no-op) and --zero (no-op).

  • csvsql adds a --prefix option to add expressions like OR IGNORE or OR REPLACE following the INSERT keyword.

  • csvsql adds a --overwrite flag to drop any existing table with the same name before creating.

  • csvsql accepts a file name for the --query option.

  • csvsql supports --linenumbers (no-op).

  • csvsql adds a --create-if-not-exists flag to not abort if the table already exists.

  • csvstat adds a --freq-count option to set the maximum number of frequent values to display.

  • csvstat supports --linenumbers (no-op).

  • in2csv adds a --names flag to print Excel sheet names.

  • in2csv adds a --write-sheets option to write the named Excel sheets to files.

  • sql2csv adds an --encoding option to specify the encoding of the input query file.


  • csvgrep no longer ignores common arguments if --linenumbers is set.

  • csvjson supports Decimal.

  • csvpy again supports IPython.

  • csvsql restores support for --no-constraints and --db-schema.

  • csvstat no longer crashes when --freq is set.

  • in2csv restores support for --no-inference for Excel files.

  • in2csv restores support for converting Excel files from standard input.

  • in2csv accepts utf-8 arguments to the --sheet option in Python 2.

1.0.1 - December 29, 2016

This is a minor release which fixes several bugs reported in the 1.0.0 release earlier this week. It also significantly improves the output of csvstat and adds a --csv output option to that command.

  • csvstat no longer crashes when a Number column has None as a frequent value. (#738)

  • csvlook documents that output tables are Markdown-compatible. (#734)

  • csvstat adds a --csv flag for tabular output. (#584)

  • csvstat output is easier to read. (#714)

  • csvpy has a better description when using the --agate flag. (#729)

  • Fix a Python 2.6 bug preventing csvjson from parsing utf-8 files. (#732)

  • Update required version of unittest to latest. (#727)

1.0.0 - December 27, 2016

This is the first major release of csvkit in a very long time. The entire backend has been rewritten to leverage the agate data analysis library, which was itself inspired by csvkit. The new backend provides better type detection accuracy, as well as some new features.

Because of the long and complex cycle behind this release, the list of changes should not be considered exhaustive. In particular, the output format of some tools may have changed in small ways. Any existing data pipelines using csvkit should be tested as part of the upgrade.

Much of the credit for this release goes to James McKinney, who has almost single-handedly kept the csvkit fire burning for a year. Thanks, James!

Backwards-incompatible changes:

  • csvjoin renames duplicate columns with integer suffixes to prevent collisions in output.

  • csvsql generates DateTime columns instead of Time columns.

  • csvsql generates Decimal columns instead of Integer, BigInteger, and Float columns.

  • csvsql no longer generates max-length constraints for text columns.

  • The --doublequote long flag is gone, and the -b short flag is an alias for --no-doublequote.

  • When using the --columns or --not-columns options, you must not have spaces around the comma-separated values, unless the column names contain spaces.

  • When sorting, null values are greater than other values instead of less than.

  • CSVKitReader, CSVKitWriter, CSVKitDictReader, and CSVKitDictWriter have been removed. Use agate.csv.reader, agate.csv.writer, agate.csv.DictReader and agate.csv.DictWriter.

  • Drop Python 2.6 support (end-of-life was October 29, 2013).

  • Drop support for older versions of PyPy.

  • If --no-header-row is set, the output has column names a, b, c, etc. instead of column1, column2, column3, etc.

  • csvlook renders a simpler, markdown-compatible table.


  • csvkit is tested against Python 3.6. (#702)

  • import csvkit as csv defers to agate readers/writers.

  • csvgrep supports --no-header-row.

  • csvjoin supports --no-header-row.

  • csvjson streams input and output if the --stream and --no-inference flags are set.

  • csvjson supports --snifflimit and --no-inference.

  • csvlook adds --max-rows, --max-columns and --max-column-width options.

  • csvlook supports --snifflimit and --no-inference.

  • csvpy supports --agate to read a CSV file into an agate table.

  • csvsql supports custom SQLAlchemy dialects.

  • csvstat supports --names.

  • in2csv CSV-to-CSV conversion streams input and output if the --no-inference flag is set.

  • in2csv CSV-to-CSV conversion uses agate.Table.

  • in2csv GeoJSON conversion adds columns for geometry type, longitude and latitude.

  • Documentation: Update tool usage, remove shell prompts, document connection string, correct typos.


  • Fixed numerous instances of open files not being closed before utilities exit.

  • Change -b, --doublequote to --no-doublequote, as doublequote is True by default.

  • in2csv DBF conversion works with Python 3.

  • in2csv correctly guesses format when file has an uppercase extension.

  • in2csv correctly interprets --no-inference.

  • in2csv again supports nested JSON objects (fixes regression).

  • in2csv with --format geojson prints a JSON object instead of OrderedDict([(...)]).

  • csvclean with standard input works on Windows.

  • csvgrep returns the input file’s line numbers if the --linenumbers flag is set.

  • csvgrep can match multiline values.

  • csvgrep correctly operates on ragged rows.

  • csvsql correctly escapes %` characters in SQL queries.

  • csvsql adds standard input only if explicitly requested.

  • csvstack supports stacking a single file.

  • csvstat always reports frequencies.

  • The any_match argument of FilteringCSVReader works correctly.

  • All tools handle empty files without error.

0.9.1 - March 31, 2015

  • Add Antonio Lima to AUTHORS.

  • Add support for ndjson. (#329)

  • Add missing docs for csvcut -C. (#227)

  • Reorganize docs so TOC works better. (#339)

  • Render docs locally with RTD theme.

  • Fix header in “tricks” docs.

  • Add install instructions to tutorial. (#331)

  • Add killer examples to doc index. (#328)

  • Reorganize doc index

  • Fix broken csvkit module documentation. (#327)

  • Fix version of openpyxl to work around encoding issue. (#391, #288)

0.9.0 - September 8, 2014

  • Write missing sections of the tutorial. (#32)

  • Remove -q arg from sql2csv (conflicts with common flag).

  • Fix csvjoin in case where left dataset rows without all columns.

  • Rewrote tutorial based on LESO data. (#324)

  • Don’t error in csvjson if lat/lon columns are null. (#326)

  • Maintain field order in output of csvjson.

  • Add unit test for json in2csv. (#77)

  • Maintain key order when converting JSON into CSV. (#325.)

  • Upgrade python-dateutil to version 2.2 (#304)

  • Fix sorting of columns with null values. (#302)

  • Added release documentation.

  • Fill out short rows with null values. (#313)

  • Fix unicode output for csvlook and csvstat. (#315)

  • Add documentation for –zero. (#323)

  • Fix Integrity error when inserting zero rows in database with csvsql. (#299)

  • Add Michael Mior to AUTHORS. (#305)

  • Add –count option to CSVStat.

  • Implement csvformat.

  • Fix bug causing CSVKitDictWriter to output ‘utf-8’ for blank fields.

0.8.0 - July 27, 2014

  • Add pnaimoli to AUTHORS.

  • Fix column specification in csvstat. (#236)

  • Added “Tips and Tricks” documentation. (#297, #298)

  • Add Espartaco Palma to AUTHORS.

  • Remove unnecessary enumerate calls. (#292)

  • Deprecated DBF support for Python 3+.

  • Add support for Python 3.3 and 3.4 (#239)

0.7.3 - April 27, 2014

  • Fix date handling with openpyxl > 2.0 (#285)

  • Add Kristina Durivage to AUTHORS. (#243)

  • Added Richard Low to AUTHORS.

  • Support SQL queries “directly” on CSV files. (#276)

  • Add Tasneem Raja to AUTHORS.

  • Fix off-by-one error in open ended column ranges. (#238)

  • Add Matt Pettis to AUTHORS.

  • Add line numbers flag to csvlook (#244)

  • Only install argparse for Python < 2.7. (#224)

  • Add Diego Rabatone Oliveira to AUTHORS.

  • Add Ryan Murphy to AUTHORS.

  • Fix DBF dependency. (#270)

0.7.2 - March 24, 2014

  • Fix CHANGELOG for release.

0.7.1 - March 24, 2014

  • Fix homepage url in

0.7.0 - March 24, 2014

  • Fix XLSX datetime normalization bug. (#223)

  • Add raistlin7447 to AUTHORS.

  • Merged sql2csv utility (#259).

  • Add Jeroen Janssens to AUTHORS.

  • Validate csvsql DB connections before parsing CSVs. (#257)

  • Clarify install process for Ubuntu. (#249)

  • Clarify docs for –escapechar. (#242)

  • Make import csvkit API compatible with import csv.

  • Update Travis CI link. (#258)

  • Add Sébastien Fievet to AUTHORS.

  • Use case-sensitive name for SQLAlchemy (#237)

  • Add Travis Swicegood to AUTHORS.

0.6.1 - August 20, 2013

  • Fix CHANGELOG for release.

0.6.0 - August 20, 2013

  • Add Chris Rosenthal to AUTHORS.

  • Fix multi-file input to csvsql. (#193)

  • Passing –snifflimit=0 to disable dialect sniffing. (#190)

  • Add aarcro to the AUTHORS file.

  • Improve performance of csvgrep. (#204)

  • Add Matt Dudys to AUTHORS.

  • Add support for –skipinitialspace. (#201)

  • Add Joakim Lundborg to AUTHORS.

  • Add –no-inference option to in2csv and csvsql. (#206)

  • Add Federico Scrinzi to AUTHORS file.

  • Add –no-header-row to all tools. (#189)

  • Fix csvstack blowing up on empty files. (#209)

  • Add Chris Rosenthal to AUTHORS file.

  • Add –db-schema option to csvsql. (#216)

  • Add Shane StClair to AUTHORS file.

  • Add –no-inference support to csvsort. (#222)

0.5.0 - August 21, 2012

  • Implement geojson support in csvjson. (#159)

  • Optimize writing of eight bit codecs. (#175)

  • Created csvpy. (#44)

  • Support –not-columns for excluding columns. (#137)

  • Add Jan Schulz to AUTHORS file.

  • Add Windows scripts. (#111, #176)

  • csvjoin, csvsql and csvstack no longer hold open all files. (#178)

  • Added Noah Hoffman to AUTHORS.

  • Make csvlook output compatible with emacs table markup. (#174)

0.4.4 - May 1, 2012

  • Add Derek Wilson to AUTHORS.

  • Add Kevin Schaul to AUTHORS.

  • Add DBF support to in2csv. (#11, #160)

  • Support –zero option for zero-based column indexing. (#144)

  • Support mixing nulls and blanks in string columns.

  • Add –blanks option to csvsql. (#149)

  • Add multi-file (glob) support to csvsql. (#146)

  • Add Gregory Temchenko to AUTHORS.

  • Add –no-create option to csvsql. (#148)

  • Add Anton Ian Sipos to AUTHORS.

  • Fix broken pipe errors. (#150)

0.4.3 - February 20, 2012

  • Begin CHANGELOG (a bit late, I’ll admit).