Parsing options

Every option is optional.

Option	Type	Default	Description
header	boolean	true	Use the first row as keys; false yields string[].
columns	(keyof T)[] or label-to-key map	none	Name, select, or rename columns.
separator	string or 'auto'	,	Field separator, or detect it.
quote	string	"	Quote character.
skipEmptyLines	boolean | 'greedy'	true	Drop blank (or whitespace-only) lines.
comment	string	none	Skip lines starting with this.
trim	boolean	false	Trim whitespace around every field.
bom	boolean	auto	Strip a leading UTF-8 byte-order mark.
dynamicTyping	boolean	false	Coerce numbers and booleans losslessly.
strict	boolean	false	Throw on a row whose field count differs.
maxRows	number	none	Stop after this many records.
row	(record, context) => T	none	Map or validate each record.

header and columns are covered in Choosing columns; dynamicTyping and row in Typing and validation. The rest are below.

Separator and quote

separator is a single character, or 'auto' to detect it. Detection reads the first line that is neither blank nor a comment, so metadata or comment lines above the header never skew it. It considers comma, semicolon, tab, and pipe, and ignores any of them inside quotes. It also works while streaming, buffering only up to that first line.

parse('a;b\n1;2', { separator: 'auto' });
// [{ a: '1', b: '2' }]  (; detected among the candidates , ; tab |)

quote sets the quote character (default "). Doubled quotes inside a quoted field are unescaped per RFC 4180.

Empty lines and comments

skipEmptyLines is true by default, dropping fully blank lines. Use 'greedy' to also drop lines that contain only whitespace, or false to keep every line.

parse('a\n1\n   \n2', { skipEmptyLines: 'greedy' });
// [{ a: '1' }, { a: '2' }]

comment skips any line whose first field begins with the given string. The first field is checked after quotes are parsed, so a quoted first field that starts with the comment string also counts as a comment.

parse('a\n1\n# note\n2', { comment: '#' });
// [{ a: '1' }, { a: '2' }]

Whitespace

trim removes surrounding whitespace from every field. It applies to quoted fields too, so leave it off when whitespace inside quotes is significant.

parse(' a , b \n 1 , 2 ', { trim: true });
// [{ a: '1', b: '2' }]

Strict mode

By default parsing is lenient: a short row pads missing fields with '' and extra fields are dropped. strict instead throws a CsvPipeError naming the row whose field count differs from the header. The number is the 1-based row in the source, where the header is row 1 (a quoted field spanning newlines counts as one row).

parse('a,b\n1', { strict: true });
// throws CsvPipeError: Row 2 has 1 fields, expected 2.

Limiting and BOM

maxRows stops after that many records, which pairs well with streaming a large file to sample the start. bom strips a leading UTF-8 byte-order mark when present; set it to false to keep one.

parse('a\n1\n2\n3', { maxRows: 2 });
// [{ a: '1' }, { a: '2' }]

See the API reference for the typed definitions.