apply

Apply series of string, date, math & currency transformations to given CSV column/s. It also has some basic NLP functions (similarity, sentiment analysis, profanity, eudex, language & name gender) detection. Its summarize subcommand condenses a column or group of columns using an OpenAI API-compatible LLM (local or commercial) with customizable, Mini Jinja-templated per-record prompts.

Table of Contents | Source: src/cmd/apply.rs | 📇🧠🤖🚀🔣👆⛩️

Description | Examples | Usage | Arguments | Apply Options | Operations Options | Summarize Options | Common Options

Description ↩

Apply a series of transformation functions to given CSV column/s. This can be used to perform typical data-wrangling tasks and/or to harmonize some values, etc.

It has five subcommands:

1. operations* - 40 string, format, currency, regex & NLP operators.
2. emptyreplace* - replace empty cells with <--replacement> string.
3. dynfmt - Dynamically constructs a new column from other columns using the <--formatstr> template.
4. calcconv - parse and evaluate math expressions, with support for units and conversions.
5. summarize* - summarize a column or group of columns using an OpenAI API-compatible LLM (local or commercial), with customizable, Mini Jinja-templated per-record prompts.

• subcommand is multi-column capable.

OPERATIONS (multi-column capable) Multiple operations can be applied, with the comma-delimited operation series applied in order:

trim => Trim the cell trim,upper => Trim the cell, then transform to uppercase lower,simdln => Lowercase the cell, then compute the normalized Damerau-Levenshtein similarity to --comparand

Operations support multi-column transformations. Just make sure the number of transformed columns with the --rename option is the same. For example, to trim and fold to uppercase the col1,col2 and col3 columns & rename them to newcol1,newcol2 and newcol3:

qsv apply operations trim,upper col1,col2,col3 -r newcol1,newcol2,newcol3 file.csv

It has 40 supported operations:

• len: Return string length
• lower: Transform to lowercase
• upper: Transform to uppercase
• squeeze: Compress consecutive whitespaces
• squeeze0: Remove whitespace
• trim: Trim (drop whitespace left & right of the string)
• ltrim: Left trim whitespace
• rtrim: Right trim whitespace
• mtrim: Trims --comparand matches left & right of the string (Rust trim_matches)
• mltrim: Left trim --comparand matches (Rust trim_start_matches)
• mrtrim: Right trim --comparand matches (Rust trim_end_matches)
• strip_prefix: Removes specified prefix in --comparand
• strip_suffix: Remove specified suffix in --comparand
• escape - escape (Rust escape_default)
• encode62: base62 encode
• decode62: base62 decode
• encode64: base64 encode
• decode64: base64 decode
• crc32: crc32 checksum
• replace: Replace all matches of a pattern (using --comparand) with a string (using --replacement) (Rust replace)
• regex_replace: Replace all regex matches in --comparand w/ --replacement. Specify as --replacement to remove matches.
• titlecase - capitalizes English text using Daring Fireball titlecase style https://daringfireball.net/2008/05/title_case
• censor: profanity filter. Add additional comma-delimited profanities with --comparand.
• censor_check: check if profanity is detected (boolean). Add additional comma-delimited profanities with -comparand.
• censor_count: count of profanities detected. Add additional comma-delimited profanities with -comparand.
• round: Round numeric values to the specified number of decimal places using Midpoint Nearest Even Rounding Strategy AKA "Bankers Rounding." Specify the number of decimal places with --formatstr (default: 3).
• thousands: Add thousands separators to numeric values. Specify the separator policy with --formatstr (default: comma). The valid policies are:
  comma, dot, space, underscore, hexfour (place a space every four hex digits) and indiancomma (place a comma every two digits, except the last three digits). The decimal separator can be specified with --replacement (default: '.')
• currencytonum: Gets the numeric value of a currency. Supports currency symbols (e.g. $,¥,£,€,֏,₱,₽,₪,₩,ƒ,฿,₫) and strings (e.g. USD, EUR, RMB, JPY, etc.). Recognizes point, comma and space separators. Is "permissive" by default, meaning it will allow no or non-ISO currency symbols. To enforce strict parsing, which will require a valid ISO currency symbol, set the --formatstr to "strict".
• numtocurrency: Convert a numeric value to a currency. Specify the currency symbol with --comparand. Automatically rounds values to two decimal places. Specify "euro" formatting (e.g. 1.000,00 instead of 1,000.00 ) by setting --formatstr to "euro". Specify conversion rate by setting --replacement to a number.
• gender_guess: Guess the gender of a name.
• copy: Mark a column for copying
• simdl: Damerau-Levenshtein similarity to --comparand
• simdln: Normalized Damerau-Levenshtein similarity to --comparand (between 0.0 & 1.0)
• simjw: Jaro-Winkler similarity to --comparand (between 0.0 & 1.0)
• simsd: Sørensen-Dice similarity to --comparand (between 0.0 & 1.0)
• simhm: Hamming distance to --comparand. Num of positions characters differ.
• simod: Optimal String Alignment (OSA) Distance to --comparand.
• eudex: Multi-lingual sounds like --comparand (boolean) Tested on English, Catalan, German, Spanish, Swedish and Italian dictionaries. It supports all C1 letters (e.g. ü, ö, æ, ß, é, etc.) and takes their sound into account. It should work on other European languages that use the Latin alphabet.
• sentiment: Normalized VADER sentiment score (English only - between -1.0 to 1.0).
• whatlang: Language Detection for 87 supported languages, with default confidence threshold of 0.9, which can be overridden by assigning 0.0 to 1.0 to --comparand. If language detection confidence is below the threshold, it will still show the best language guess, followed by the confidence score, ending with a question mark. If you want to always displays the confidence score, end the --comparand value with a question mark (e.g. 0.9?) https://github.com/greyblake/whatlang-rs/blob/master/SUPPORTED_LANGUAGES.md

EMPTYREPLACE (multi-column capable) Replace empty cells with <--replacement> string. Non-empty cells are not modified. See the fill command for more complex empty field operations.

Dynfmt

Dynamically constructs a new column from other columns using the <--formatstr> template. The template can contain arbitrary characters. To insert a column value, enclose the column name in curly braces, replacing all non-alphanumeric characters with underscores.

If you need to dynamically construct a column with more complex formatting requirements and computed values, check out the py command to take advantage of Python's f-string formatting.

Calcconv

Parse and evaluate math expressions into a new column, with support for units and conversions. The math expression is built dynamically using the <--formatstr> template, similar to the DYNFMT subcommand, with the addition that if the literal '' is found at the end of the template, the inferred unit will be appended to the result.

For a complete list of supported units, constants, operators and functions, see https://docs.rs/cpc

Examples ↩

OPERATIONS

Trim, then transform to uppercase the surname field.

qsv apply operations trim,upper surname file.csv

Trim, then transform to uppercase the surname field and rename the column uppercase_clean_surname.

qsv apply operations trim,upper surname -r uppercase_clean_surname file.csv

Trim, then transform to uppercase the surname field and save it to a new column named uppercase_clean_surname.

qsv apply operations trim,upper surname -c uppercase_clean_surname file.csv

Trim, then transform to uppercase the firstname and surname fields and rename the columns ufirstname and usurname.

qsv apply operations trim,upper firstname,surname -r ufirstname,usurname file.csv

Trim parentheses & brackets from the description field.

qsv apply operations mtrim description --comparand '()<>' file.csv

Replace ' and ' with ' & ' in the description field.

qsv apply operations replace description --comparand ' and ' --replacement ' & ' file.csv

Extract the numeric value of the Salary column in a new column named Salary_num.

qsv apply operations currencytonum Salary -c Salary_num file.csv

Convert the USD_Price to PHP_Price using the currency symbol "PHP" with a conversion rate of 60.

qsv apply operations numtocurrency USD_Price -C PHP -R 60 -c PHP_Price file.csv

Base64 encode the text_col column & save the encoded value into new column named encoded & decode it.

qsv apply operations encode64 text_col -c encoded file.csv | qsv apply operations decode64 encoded

Compute the Normalized Damerau-Levenshtein similarity of the neighborhood column to the string 'Roxbury' and save it to a new column named dln_roxbury_score.

qsv apply operations lower,simdln neighborhood --comparand roxbury -c dln_roxbury_score boston311.csv

You can also use this subcommand command to make a copy of a column:

qsv apply operations copy col_to_copy -c col_copy file.csv

EMPTYREPLACE

Replace empty cells in file.csv Measurement column with 'None'.

qsv apply emptyreplace Measurement --replacement None file.csv

Replace empty cells in file.csv Measurement column with 'Unknown Measurement'.

qsv apply emptyreplace Measurement --replacement 'Unknown Measurement' file.csv

Replace empty cells in file.csv M1,M2 and M3 columns with 'None'.

qsv apply emptyreplace M1,M2,M3 --replacement None file.csv

Replace all empty cells in file.csv for columns that start with 'Measurement' with 'None'.

qsv apply emptyreplace '/^Measurement/' --replacement None file.csv

Replace all empty cells in file.csv for columns that start with 'observation' case insensitive with 'None'.

qsv apply emptyreplace --replacement None '/(?i)^observation/' file.csv

DYNFMT

Create a new column 'mailing address' from 'house number', 'street', 'city' and 'zip-code' columns:

qsv apply dynfmt --formatstr '{house_number} {street}, {city} {zip_code} USA' -c 'mailing address' file.csv

Create a new column 'FullName' from 'FirstName', 'MI', and 'LastName' columns:

qsv apply dynfmt --formatstr 'Sir/Madam {FirstName} {MI}. {LastName}' -c FullName file.csv

CALCCONV

Do simple arithmetic:

qsv apply calcconv --formatstr '{col1} + {col2} * {col3}' --new-column result file.csv

Arithmetic with support for operators like % and ^:

qsv apply calcconv --formatstr '{col1} % 3' --new-column remainder file.csv

Convert from one unit to another:

qsv apply calcconv --formatstr '{col1} Fahrenheit in Celsius' -c metric_temperature file.csv

Mix units and conversions are automatically done for you:

qsv apply calcconv --formatstr '{col1}km + {col2}mi in meters' -c meters file.csv

You can append the inferred unit at the end of the result by ending the expression with '':

qsv apply calcconv --formatstr '({col1} + {col2})km to light years <UNIT>' -c light_years file.csv

You can even do complex temporal unit conversions:

qsv apply calcconv --formatstr '{col1}m/s + {col2}mi/h in kilometers per h' -c kms_per_h file.csv

Use math functions - see https://docs.rs/cpc/latest/cpc/enum.FunctionIdentifier.html for list of functions:

qsv apply calcconv --formatstr 'round(sqrt{col1}^4)! liters' -c liters file.csv

Use percentages:

qsv apply calcconv --formatstr '10% of abs(sin(pi)) horsepower to watts' -c watts file.csv

Use very large numbers:

qsv apply calcconv --formatstr '{col1} Billion Trillion * {col2} quadrillion vigintillion' -c num_atoms file.csv

SUMMARIZE

The summarize subcommand sends each record to an OpenAI API-compatible LLM and stores the returned summary in a new column. The prompt is a Mini Jinja template (like qsv template) rendered per record - reference any column by its "safe" name (non-alphanumeric chars become '_'); the 1-based row number is available as {{QSV_ROWNO}}. If no --prompt/--prompt-file is given, a default prompt that summarizes the selected column/s is used. The LLM endpoint, model & API key are resolved with this precedence:
CLI flag > env var (QSV_LLM_BASE_URL / QSV_LLM_MODEL / QSV_LLM_APIKEY) > built-in default (default base-url: http://localhost:1234/v1 ; default model: openai/gpt-oss-20b). An API key is required when the base URL is not a localhost endpoint (set --api-key to "NONE" to suppress sending a key). One HTTP call is made per uncached row, so summarizing large files can be slow & costly. Results are cached on disk (keyed by base-url, model, max-tokens, addl-props & the rendered prompt) so repeated rows & re-runs don't re-pay for inference; use --no-cache to disable & --fresh to force fresh calls that refresh the cache.

Summarize the support_ticket column into a new "summary" column using a local LLM:

qsv apply summarize support_ticket -c summary file.csv

Use a custom prompt referencing multiple columns against a cloud provider:

qsv apply summarize subject,body -c summary --prompt 'Summarize: {{subject}} - {{body}}' -u https://api.openai.com/v1 -m gpt-4o-mini -k $OPENAI_API_KEY file.csv

Also capture per-row inference time & token usage in extra columns:

qsv apply summarize notes -c summary --stats file.csv

For more examples, see tests.

See also https://github.com/dathere/qsv/wiki/Transform-and-Reshape#apply

Usage ↩

qsv apply operations <operations> [options] <column> [<input>]
qsv apply emptyreplace --replacement=<string> [options] <column> [<input>]
qsv apply dynfmt --formatstr=<string> [options] --new-column=<name> [<input>]
qsv apply calcconv --formatstr=<string> [options] --new-column=<name> [<input>]
qsv apply summarize [options] --new-column=<name> <column> [<input>]
qsv apply --help

Arguments ↩

Argument	Description
<column>	The column/s to apply the transformation to. Note that the argument supports multiple columns for the operations & emptyreplace subcommands. See 'qsv select --help' for the format details.
<operations>	The operation/s to apply.
<column>	The column/s to apply the operations to.
<column>	The column/s to check for emptiness.
<column>	The column/s whose values are summarized by the LLM. Used to build the default prompt. With a custom prompt (via the --prompt/--prompt-file options), any column can be referenced.
<input>	The input file to read from. If not specified, reads from stdin.

Apply Options ↩

Option	Type	Description	Default
‑c, ‑‑new‑column	string	Put the transformed values in a new column instead.
‑r, ‑‑rename	string	New name for the transformed column.
‑C, ‑‑comparand=<string>	string	The string to compare against for replace & similarity operations. Also used with numtocurrency operation to specify currency symbol.
‑R, ‑‑replacement=<string>	string	The string to use for the replace & emptyreplace operations. Also used with numtocurrency operation to conversion rate.
‑f, ‑‑formatstr=<string>	string	This option is used by several subcommands:

Operations Options ↩

Option	Type	Description	Default
‑j, ‑‑jobs	integer	The number of jobs to run in parallel. When not set, the number of jobs is set to the number of CPUs detected.
‑b, ‑‑batch	integer	The number of rows per batch to load into memory, before running in parallel. Automatically determined for CSV files with more than 50000 rows. Set to 0 to load all rows in one batch. Set to 1 to force batch optimization even for files with less than 50000 rows.	50000

Summarize Options ↩

Option	Type	Description	Default
‑u, ‑‑base‑url	string	Base URL of the OpenAI API-compatible endpoint. Precedence: this flag > QSV_LLM_BASE_URL env var > http://localhost:1234/v1
‑m, ‑‑model	string	Model name compatible with the OpenAI API spec. Precedence: this flag > QSV_LLM_MODEL env var > openai/gpt-oss-20b
‑k, ‑‑api‑key	string	API key for Bearer token authentication. Precedence: this flag > QSV_LLM_APIKEY env var. Set to "NONE" to suppress sending a key. Required for non-localhost URLs.
‑t, ‑‑max‑tokens	integer	Maximum number of tokens in the LLM output. Set to 0 to not send a max_tokens limit (automatically used for localhost endpoints).	10000
‑‑timeout	integer	Timeout for each LLM request in seconds (0 = no timeout).	300
‑‑addl‑props	string	Additional model properties as a JSON object, e.g. '{"reasoning_effort": "high", "temperature": 0.2}'
‑‑prompt	string	Mini Jinja prompt template rendered per record. Overrides the default prompt. Reference columns by their safe name.
‑‑prompt‑file	string	Read the prompt template from a file. Ignored if --prompt is set.
‑‑rate‑limit	float	Seconds to sleep between LLM requests to avoid provider rate limits. Accepts fractional seconds (e.g. 0.5).	0
‑‑on‑error	string	What to do when an LLM request fails: "fail" aborts; "skip" writes an "<ERROR: ...>" cell and continues.	fail
‑‑user‑agent	string	Custom user agent for LLM requests. Supports variables like $QSV_VERSION.
‑‑cache‑dir	string	Directory for the disk cache.	~/.qsv-cache/apply-summarize
‑‑no‑cache	flag	Disable the disk cache (one LLM call per row, always).
‑‑fresh	flag	Force fresh LLM calls, refreshing any cached values.
‑‑stats	flag	Append two extra columns per row alongside --new-column: "_elapsed_ms" (inference time in milliseconds) and "_tokens" (total tokens used). For cached rows, these report the values captured when the summary was first computed.

Common Options ↩

Option	Type	Description	Default
‑h, ‑‑help	flag	Display this message
‑o, ‑‑output	string	Write output to instead of stdout.
‑n, ‑‑no‑headers	flag	When set, the first row will not be interpreted as headers.
‑d, ‑‑delimiter	string	The field delimiter for reading CSV data. Must be a single character. (default: ,)
‑p, ‑‑progressbar	flag	Show progress bars. Not valid for stdin.

---

Source: src/cmd/apply.rs | Table of Contents | README