Text Diff Checker

// How to read the differences

An unchanged line carries both line numbers. A removed line has its old number and a minus sign; an inserted line has its new number and a plus sign. When one line was rewritten rather than simply added or removed, the smaller marks inside the pair show the words that actually changed. That second pass is what separates a changed sentence from a wall of highlighted text.

Unified view keeps the whole sequence in one list, which is compact and reads like a patch. Split view puts the old and new lines beside each other, which is easier when the shape stayed similar but several details moved. On a narrow screen the split view stacks the two sides so neither column becomes too small to read.

The similarity figure is defined as twice the number of unchanged characters divided by the two text lengths together. That makes it symmetrical: swapping the sides never changes the percentage. Exact scores are rounded to a whole percent with 100% reserved for identical text. If that auxiliary score would exceed its deterministic work limit, the exact patch and counts stay available and the score says it was not computed.

// Lines, words, or characters

Line mode is the useful default for source code, configuration, lists, and any document where a newline has meaning. It keeps old and new line numbers, pairs nearby replacements, and produces the plain plus, minus, and space-prefixed patch copied by the button above the result.

Word mode is better for prose. It keeps spaces and punctuation in place while marking whole inserted and removed words, so a contract clause or edited paragraph reads as language rather than as individual letters. Character mode is the microscope: it exposes a changed digit, a missing space, or the single letter separating colour from color.

The comparison updates as the text changes. The inputs open on a short real example, and the first click selects that example so typing or pasting replaces it. Each side can also read a plain-text file up to 5 MB. Long unchanged sections are folded around the exact edits. Input that is too large, still needs more than 1,000 displayed rows, or takes unusually complex work switches to a clearly marked bounded summary so the tab stays responsive.

// A comparison that does not need your document

A text diff is arithmetic on two strings. It does not need an account, a saved project, or a server copy of the material. This page performs that work in the browser tab, so a contract draft, unpublished article, configuration file, or private note is not sent away merely to find what changed.

The no-upload promise is deliberately narrower than a security guarantee. Browser extensions, a managed device, or someone with access to the screen can still read what you open. What this tool controls is its own network behavior: the comparison, file reading, similarity calculation, and patch generation all happen on your device.

// fair questions

How do I compare two texts for differences?

Paste or type the earlier version into Original and the later version into Changed. The result updates immediately. Removed content is marked with a minus, inserted content with a plus, and rewritten lines highlight the smaller word changes inside them.

Is it safe to compare confidential documents?

The tool does not upload either text, create an account, or save a comparison. Everything it controls runs in this browser tab. That avoids sending the document to this site, though extensions, device management software, and anyone with access to your screen remain outside that promise.

Can I compare two files?

Yes. Use open file above either pane to load a plain-text file up to 5 MB, including TXT, Markdown, CSV, JSON, logs, and source code. Binary office formats such as DOCX and PDF are not plain text and need to be exported or copied first.

Which comparison mode should I use?

Use Lines for code, configuration, and lists; Words for paragraphs, contracts, and essays; Characters for typo hunting or exact small edits. Switching modes never changes either input.

What does the similarity percentage mean?

When computed, it is twice the unchanged character count divided by the combined length of both texts, rounded to a whole percent with 100% reserved for identical text. The score has its own deterministic work limit: an exact patch can remain available while the score says not computed. Bounded summaries report a conservative lower bound only when one can be established honestly.