Paste a git diff and read it as a proper diff, with files, hunks, and line numbers. Instant local checks run in your browser the moment it parses, with no AI involved at all. When you want a deeper pass, only the changed lines are sent for review, and every finding is pinned to the line it belongs to instead of arriving as a wall of text.
Generate one with git diff, git diff --staged, or git diff main...HEAD. Never paste secrets or credentials.
Parsing, rendering, the file and line statistics, and a set of instant local checks: leftover debugger statements and console logs, merge conflict markers, likely hardcoded secrets, empty catch blocks, eval, innerHTML assignment, and insecure http:// URLs. All of that happens in your browser the moment you paste, costs nothing, and never leaves your machine.
Because a full-file review burns compute re-reading code you didn't touch and buries the findings that matter in observations about code that was already there. Sending only the hunks makes the review faster, cheaper, and sharper, and it's what a human reviewer does anyway.
No. It's a fast first pass that catches obvious bugs and omissions before a teammate looks. Human review still matters for architecture, product intent, and team conventions.
It's treated as if every line were newly added, so you get the same review with line numbers that match the file. A real diff gives better results, because knowing what changed is most of what makes a review useful.
The local checks never send anything anywhere. When you run the deeper review, only the changed hunks are sent to the AI engine to produce the findings, and they are not published or shared. Don't paste secrets or credentials into any web tool, including this one.
The diff view works with any language. The instant local checks are tuned for JavaScript, TypeScript, Python, Go, Java, Ruby, and PHP; the AI pass handles any mainstream language.