Remove duplicate CSV rows

Collapse repeated records down to one while leaving the rest of the file in its original order. The usual source is a list assembled from more than one place: three event exports with the same attendee in each, a mailing list built from two systems, or a stock file where a product was entered twice on different days.

Complete-row mode removes a row only when every value matches an earlier one, which makes it the safe default. Key-column mode is the one that does real work: name the field that identifies a record, such as email, customer_id, or sku, and later rows sharing that value are dropped even when everything else about them differs. Several comma-separated names combine into one key, so first_name, last_name, postcode treats a match on all three as the same person. A key column that does not exist is reported by name rather than being quietly ignored.

Matching is exact, and this is the limit to plan around. ann@example.com and Ann@Example.com are two different values here, as are Acme Ltd and Acme Ltd., and a trailing space is enough to keep a row alive. There is no fuzzy matching, and none is pretended: a tool that decided for itself that two slightly different names were the same person would be guessing about your data. Normalise the key column first if the source systems disagree about case or spacing, then deduplicate.

The first occurrence is the one kept, so ordering the file before running this decides which version survives. If the newest record should win, sort the file so that it comes first; otherwise you keep the oldest address on file rather than the current one.

Quoted commas, multiline cells, semicolons, tabs, and Unicode text are handled by the same parser used across the CSV tools, and the summary afterwards states how many rows were removed, which is worth a glance. A number far larger than you expected usually means the chosen key is not as identifying as it looked. Everything runs in the browser, so a list of real customers never leaves your machine to be cleaned.