Text Diff

A fast, private, browser-only text comparison tool. Paste any two versions of a document — two drafts of a paragraph, two versions of a config file, two log outputs, two prompts — and Text Diff shows you exactly what changed, line by line, with word-level highlighting inside each changed line so you never miss a single-word substitution.

Why a dedicated text diff tool matters

Comparing two blocks of text is one of the most common operations in a developer, writer, or editor's day, and yet most online "difference checkers" are either ad-infested pages that POST your text to a server, or bare-bones textarea-pair widgets that just dump the two inputs side by side without actually diffing them. This tool ships a structural diff that runs the classic Longest Common Subsequence algorithm in your browser, identifies which lines are equal, inserted, or deleted, and then — for lines that were replaced — runs a second pass at the word level so you see exactly which tokens changed.

The result is the kind of diff you used to need a desktop tool for: line-accurate, word-precise, with proper side-by-side panes, stable line numbers, and a clean unified-patch export. No accounts, no uploads, no analytics watching your payloads. The entire algorithm is under 200 lines of pure TypeScript and runs in your browser — nothing about the two texts you compare leaves the page.

What it does

• Line-level LCS diff — classifies every line as equal, inserted, or deleted using the dynamic-programming Longest Common Subsequence algorithm. Guaranteed minimal edit distance, deterministic output.
• Word-level highlighting within changed lines — when a delete and an insert pair up, a second LCS pass over word tokens tints the specific words that differ. Catches single-character typos that line-only diffs bury in noise.
• Side-by-side view with line numbers — left and right panes render as a two-column table with 1-based line numbers, colored backgrounds for inserts (green) and deletes (red), and per-row alignment so your eye moves horizontally across the change.
• Ignore case option — folds Hello and HELLO to the same line key without mutating the surfaced text.
• Ignore whitespace option — collapses runs of whitespace inside a line so "a b" and "a b" compare equal. Useful for diffing re-wrapped Markdown or reformatted YAML.
• CRLF, LF, and lone CR handling — the splitter normalises every line terminator, so a Windows-saved file and a Unix-saved file with identical content show zero differences.
• Unicode-aware — word tokenisation uses \p{L} and \p{N} property escapes, so Cyrillic, CJK, accented Latin, and emoji all tokenise sensibly.
• Unified patch export — the Copy patch button emits ---/+++/@@ hunk headers followed by space/plus/minus prefixed lines, the format understood by patch(1), every modern code review tool, and most LLM prompts.
• URL-hash share — pressing ⌘/Ctrl + Enter or clicking Share writes both texts and the option toggles into the URL fragment, so you can send a colleague a link that opens to the exact comparison you were looking at. The fragment never reaches a server.

How the algorithm works

The Longest Common Subsequence (LCS) algorithm builds a two-dimensional table where each cell (i, j) holds the length of the longest subsequence common to the first i lines of the left text and the first j lines of the right text. Building the table takes O(n·m) time. A backtrack from the bottom-right corner of the table then emits the edit script: whenever both current lines match the current cell value came from a diagonal move (equal); otherwise the move came from the side with the higher score (insert or delete).

For the word-level pass, the exact same algorithm runs on the tokens of the two lines that were paired as a replace. Because most replaced lines are short, this second pass is effectively free — but it's what makes the diff view actually readable when you are comparing two drafts of a sentence, two slightly different log lines, or two versions of a config key.

Line keys are normalised — lower-cased if Ignore case is on, whitespace-collapsed if Ignore whitespace is on — before the LCS table is built, so the options affect equality without mutating what you see in the output.

Common use cases

• Comparing two drafts of documentation, a contract, or an email.
• Auditing a config file for unexpected edits (migrations, CI reruns, accidental reformat).
• Hunting the exact line that changed in a failing log snapshot.
• Checking an LLM rewrite against the original prompt or response.
• Confirming a find-and-replace didn't over-fire.
• Reviewing a translation against its source paragraph by paragraph.

What's not supported (yet)

• Three-way merge view — today Text Diff is strictly two-way. A three-pane merge with a computed base is on the roadmap.
• Block-level move detection — if you cut a paragraph from one place and paste it elsewhere, the tool reports that as a delete and an insert, not a move. A move-aware mode would require a more expensive diff variant (like Patience or Histogram diff); worth it for code, but noisier for prose.
• Syntax-aware diff — for structured data, use the dedicated JSON Diff tool, which compares parsed values, ignores key order inside objects, and emits path-tagged changes.
• Character-level diff for very short lines — Text Diff's word-level pass bottoms out at word tokens. For single-character comparison (DNA sequences, obscure codes), a character-level fork is on the backlog.

Related tools

• JSON Diff — structural diff for parsed JSON documents; ignores key order, compares values by type, reports changes as paths.
• Case Converter — normalise identifier casing before or after a diff to avoid noise when comparing renamed variables.