Free Text, JSON & CSV Diff Checker

Plain text, diffed twice over: Lines are matched first, then only the lines that changed are re-diffed at word level. A pure line diff would mark a whole paragraph changed because you swapped one adjective; a pure word diff over the whole document would lose the line structure that keeps the two panes aligned. Doing both gives you aligned panes and a highlight tight enough to point at the single edited word.

JSON compared as data, not as text: Both sides are parsed, then compared structurally. Key order and indentation are ignored, so a reserialised file reads as identical; array order is preserved, because in JSON the order of an array is data. Every difference is reported at its full path — $.limits.maxWords, not "line 6" — and a value that changed type from "5" to 5 is flagged as exactly that.

CSV reconciled by row and by column: Both files are parsed to RFC 4180, so quoted commas and embedded newlines survive, and the separator (comma, semicolon, tab or pipe) is detected from the data. Columns are matched by header name, so a reordered or newly inserted column is reported once at the column level instead of as a change on every single row. Rows can be matched by position or by a key column, which is how you compare two exports whose rows were sorted differently.

Word or character granularity, and it is grapheme-aware: Character mode compares grapheme clusters, not UTF-16 code units, so a family emoji or a flag stays a single unit instead of being split into replacement boxes and counted as four changes. Word mode uses Intl.Segmenter rather than splitting on spaces, which is what lets an edit inside a Chinese or Japanese sentence be isolated to one word instead of lighting up the whole line.

Differences you can choose to ignore: Line endings are always normalised — a Windows file and its Unix copy are identical text, not 400 changed lines. Whitespace, case and substituted punctuation are switchable, which is the answer to the most common complaint about diff tools: two documents that look the same on screen reported as completely different because Word replaced every apostrophe with a curly one.

Private and off the main thread: Nothing is uploaded, and there is no account. Everything happens in a Web Worker in your own tab, so a 100,000-word manuscript does not freeze the textarea while it computes — and a confidential contract, an unreleased manuscript or a customer export never leaves your machine, which is not something a server-side diff tool can offer you.