JSON Formatter

Three modes, one workspace. Format, Validate and Diff share the same parser and the same editor, so a document behaves identically whichever mode reads it, and switching between them never reformats what you pasted.

What each mode does

Faults are located, not just reported

Nineteen distinct causes are named, each with its line, its column and its byte offset, and the surrounding lines are printed with a caret under the character. A trailing comma, a single-quoted key, a bare word, a comment, an invalid escape and a raw control character inside a string all read as themselves instead of as one generic parse error.

Two faults a valid document can still carry

Both are reported on documents that parse cleanly, because both change your data silently. A duplicate key cannot be recovered by anything: JSON.parse keeps only the last value, and so does every other parser you are likely to use, so the warning names the key and its line in order to fix the producer. An integer beyond 2^53 changes value the moment JavaScript reads it, which is how a 64-bit identifier quietly becomes a different identifier.

What the diff refuses to call a difference

Indentation, line endings and the order object keys were written in carry no meaning in JSON, so they are not differences. Array order does carry meaning and is compared as a sequence, which is why one inserted element reads as one insertion rather than as every element after it having changed.

Reading a document too large to scroll

• The tree opens one level at a time, so a large file does not become a large page. Every row carries the path that reaches it, users[4].email, with a button to copy the path or the value.
• Search by key, by value or both, with a match count and a step through the hits.
• Statistics appear once the document parses: keys, objects, arrays, strings, numbers, booleans, nulls, maximum depth, size and lines.
• Conversions to YAML, to a TypeScript interface or to CSV. When a document cannot be a table, the CSV path says which shape it needed rather than emitting something misleading.

The ceiling is 5 MiB. Above roughly 300 KiB the syntax colours switch off so typing stays smooth, and on very large documents the detailed diff stops early and says so; validation and statistics keep working across the whole range.

There is no API endpoint behind this page. The document is parsed in browser memory and never reaches a server, nothing is written to storage or to the address bar, and the only thing kept between visits is which indentation you picked.