Turn a CSV export into a JSON array of objects, using the header row as the keys. The parser follows RFC 4180 properly rather than splitting on commas, so a quoted field containing a comma, a doubled quote, or a line break survives intact — which is exactly where naive converters mangle your data without telling you. The delimiter is detected automatically by looking outside quoted regions, so semicolon-separated European exports work without you changing a setting. Turn on type inference and plain numbers, booleans and nulls become real JSON types, while anything ambiguous — a leading-zero product code, a long account number — is left as a string so it cannot lose digits. Drop a whole folder of CSVs and get them all back as a ZIP. Nothing is uploaded: the files are read on your device, which matters because CSV exports are usually customer lists and sales figures.
| CSV input | Correct result | What splitting gives you |
|---|---|---|
| a,"b,c",d | 3 fields | 4 fields — the quoted comma splits |
| "say ""hi""" | say "hi" | Literal doubled quotes left in |
| a,"line1⏎line2",c | One row, 3 fields | Two broken rows |
| a;b;c (European) | 3 fields | 1 field — wrong delimiter |
| BOM + a,b | Header "a" | Header "a" that never matches |
Yes — that is the main reason to use a real parser rather than splitting on the delimiter. A field like "Smith, John" stays one value, doubled quotes inside a field unescape to a single quote, and a line break inside quotes stays part of the value rather than starting a new row.
The delimiter is detected automatically, counting candidates outside quoted regions so a comma inside a quoted field does not mislead it. You can also set it manually if the guess is wrong.
It converts values that are unambiguously numbers, booleans or null into real JSON types. It deliberately leaves alone anything that only looks numeric — leading zeros like 007, values with thousands separators, and integers too large to represent exactly — because turning an identifier into a float silently corrupts it.
Blank headers become column_2, column_3 and so on, and a duplicate name gets a numeric suffix. Both are reported as warnings. The alternative — one key overwriting another — would drop a whole column without any sign that it happened.
Yes. Drop multiple files or a whole folder and each is converted separately, with its own status, then downloaded together as a ZIP that keeps the original filenames.
No. Files are read locally with the browser's File API and parsed in the page. Nothing is transmitted, which is the point — CSV exports are usually customer records, sales figures, or user lists.
CSV is parsed with a hand-written RFC 4180 state machine rather than by splitting on commas, because splitting breaks the moment a field contains the delimiter. The parser tracks whether it is inside a quoted field, treats a doubled quote inside one as a literal quote, and keeps delimiters and line breaks that appear within quotes as part of the value. CRLF and lone LF both end a row, a UTF-8 BOM is stripped so it cannot become part of the first column name, and a trailing newline does not produce a phantom empty row. The delimiter is detected by counting comma, semicolon, tab and pipe candidates outside quoted regions across the first few lines — counting blind picks the wrong one as soon as a comma appears inside a quoted semicolon-separated field, which is how most European exports are written. Header names that are blank or duplicated are renamed rather than silently collapsing columns together, and rows with too few cells are padded with empty values, both reported as warnings. Type inference is conservative: it converts plain numeric, boolean and null forms but leaves leading-zero values and integers beyond the safe range as strings, so an identifier is never turned into a lossy float. Files are read with the File API and never uploaded.