Compare two texts or two files, line by line and word by word
Two versions in, every difference out, compared in the browser
Two boxes, and every difference between them: added lines, removed lines, rewritten lines, and inside a rewritten line the words that actually changed. Side by side or in one column, with a switch for showing only what changed.
The comparison runs in this tab. There is no upload and no request to make: the page has no API call, nothing is written to storage, and closing it leaves nothing behind. That is the difference that matters in a diff tool, because a diff tool is where people paste contracts, credentials out of configuration files and code that is not theirs.
Underneath it is not one algorithm but three. What matches at the start and the end is trimmed first, which turns most real comparisons into a few dozen lines of work. What is left is anchored on the lines that appear exactly once on each side - a line unique in both texts can only match itself - which is what makes a block that was moved read as a move rather than as a deletion plus an addition. The small stretches between those anchors go through Myers' algorithm, which is exact. Ten thousand lines with a change in the middle comes back instantly.
The result exports as a unified diff: the format patch, git apply and every code review tool reads.
The most important capabilities, explained in practical terms.
Added, removed and rewritten lines each have their own ground, and a rewritten line carries the words that moved marked inside it. The state is carried by the background, the line number and the plus or minus sign together, never by colour alone.
Side by side for reading two versions as documents, one column for reading a change the way a code review reads it. A "changes only" switch drops the unchanged lines and keeps one line of context on each side of a run.
Case, whitespace and blank lines. Whitespace collapses runs of spaces and tabs rather than deleting them, so an indentation change stops being a difference while "ab" and "a b" stay different lines. Line numbers always count every line in your file.
Text files up to 2 MB per side: .txt, .md, .csv, .json, .xml, .yml, .html, source, logs, subtitles. A PDF or a .docx is refused by name rather than shown as a screen of binary.
Copy or download the result as a unified diff with three lines of context: the format patch, git apply and every review tool reads. Every hunk header counts the lines the hunk really carries, which is the part broken exporters get wrong.
Common prefix and suffix trimmed, then patience anchors, then Myers on what is left. Ten thousand lines with one change is instant. The view paints in slices of 600 rows so the browser never stops responding, and if two large texts have nothing in common the tool says so instead of grinding.
A clear path from the official source to your first successful workflow.
Use one of the official links available for your device or browser.
Follow the store instructions or open the web application. No third-party installers.
Review the options, choose your preferences and start with the core workflow.
A transparent look at the core technologies used to build and maintain this product.
Your data stays on your device. Always.
