CSV to JSON converter (and back).

A parser written for this page follows RFC 4180, the closest thing CSV has to a standard. It reads quoted fields that contain the delimiter, doubled quotes and line breaks, accepts both CRLF and LF line endings and ignores a leading byte-order mark from Excel exports. Everything runs in your browser tab, so exports of customer or order data never leave your machine.

With a header row, each line becomes a JSON object keyed by the column names. Without one, each line becomes an array. Blank header cells are named column_1, column_2 and so on, and repeated names get a suffix (email, email_2) so no column silently overwrites another.

Fields are separated by a delimiter and records by line breaks. A field that contains the delimiter, a double quote or a line break must be wrapped in double quotes, and a double quote inside it is written twice: "She said ""ship it""". Spaces are part of the field, so a, b gives the second field a leading space.

The RFC specifies commas, but real files vary. Spreadsheet apps in regions that use a decimal comma (much of Europe and Latin America) export semicolon-separated files, databases and logs often use tabs (TSV), and some systems use pipes. Auto-detect counts each candidate outside quotes across the first lines and picks the one that is most frequent and most consistent. Some exports add a sep=; first line for Excel; delete it before converting, because it is not part of the data.

CSV has no types; every field is text. With Infer types on, plain integers and decimals become JSON numbers, true and false become booleans, and empty fields become null. Turn it off to keep every value as a string.

Inference is deliberately cautious. Values with leading zeros, such as ZIP codes, phone numbers or product codes (02134), stay strings so the zero is not lost. Integers above 9,007,199,254,740,991 stay strings because JavaScript numbers cannot hold them exactly, which matters for database IDs. Decimal commas (49,90) and dates are left as text; convert them in code where you know the locale and format.

When in doubt, turn inference off and convert types in code, where the rules can be tested. A column of numbers with the odd N/A entry is a typical case that needs a human decision rather than a guess.

The input should be an array of objects, an array of arrays or a single object. Columns are the union of all keys in the order they first appear, so objects with different shapes still line up, and missing values become empty cells. Nested objects and arrays are written as JSON text inside the cell. Pick the output delimiter to suit the destination: semicolons for spreadsheets in decimal-comma locales, tabs when values are full of commas.

Fields are quoted only when needed, quotes are doubled and records end with CRLF as RFC 4180 specifies. To import the result into Excel with accented characters intact, use Data, From Text/CSV, and choose UTF-8, since double-clicking a CSV may assume a legacy encoding.

Ragged rows (a row with more or fewer fields than the header) usually mean an unquoted delimiter inside a value, such as an address with a comma. The status line names the first ragged row, and the table marks missing cells. An unclosed quote is reported at the line and column where the field starts.

Curly quotes pasted from a word processor are not CSV quotes; replace them with straight double quotes. When you export CSV that other people will open in a spreadsheet, be aware of formula injection: a cell starting with =, +, - or @ may be executed as a formula, so prefix such values with an apostrophe or validate them on export. Building repeatable imports like this is part of our data engineering work.

CSV is still the common language between systems: spreadsheet exports, bank statements, CRM and e-commerce reports and bulk uploads to SaaS tools all speak it, while APIs and document databases expect JSON. Converting between the two is everyday ETL work, and most problems come from the same four places: encoding, delimiters, quoting and types.

For a one-off file, this tool is enough. For a recurring import, write the rules down: expected columns, delimiter, encoding (UTF-8), how empty values and dates are written, and what happens to rows that fail validation. Then load files with a real CSV library, such as Python's csv module or a streaming parser in Node.js, rather than splitting lines on commas, which breaks the first time a value contains a comma or a line break.

Very large files belong in a pipeline rather than a browser tab. Columnar formats such as Parquet suit analytics storage better, and a data pipeline can convert CSV on arrival so analysts never handle the raw file.

Nothing is uploaded: parsing, conversion and the table preview all happen in JavaScript in this tab. The preview renders the first 500 rows to stay fast, while the JSON or CSV output always includes every row. Files of several megabytes convert in well under a second on a modern laptop.

If a conversion looks wrong, open the Table view first: it shows exactly how each row was split, which makes delimiter and quoting problems obvious. Fix the source or override the delimiter rather than editing the JSON by hand.