The last mile of
code review is still human
A native desktop app for reviewing diffs you didn't write —
whether they came from a teammate or an AI.
Two kinds of review.
Same problem.
Maybe you used Claude Code or Copilot and now you're staring at a 400-line diff of your own code, trying to verify what it actually did. Or maybe a teammate's PR landed in your queue and you need to understand it well enough to approve it.
Either way, the job is the same: read the diff, understand the changes, and make a judgment call. For some teams this is a compliance requirement. For others it's just good practice. Either way, it's not something you can automate away.
The problem is that the tools for this haven't kept up. GitHub's diff viewer shows every changed line with equal weight — the meaningful refactor, the auto-formatted import, the whitespace fix. When the diff is hundreds of lines, that's a wall of green and red.
Review is built for this job. It triages the noise, surfaces the signal, and gives you a native, keyboard-driven interface to move through large diffs with actual understanding.
Trust patterns
Define rules that auto-approve trivial hunks — import reordering, formatting, whitespace. In a 400-line diff, half the hunks might be noise. Trust patterns let you skip them instantly.
The taxonomy is extensible with custom patterns per project.
Batch similar changes
Review automatically groups similar hunks across files — like the same import removed from a dozen modules. Approve or reject them all at once instead of clicking through each one.
When you're staring at 47 hunks and 18 are the same mechanical change, batch actions turn minutes into seconds.
Symbol-level diffs
Powered by tree-sitter. Instead of scrolling through files line-by-line, navigate the diff as a tree of changed functions, classes, and methods. Supports 10+ languages.
AI classification
Claude reads each hunk and classifies it — refactor, bug fix, new feature, test update. Plus narrative summaries that explain why the change was made.
Local-first
Reads directly from your local git repos. No data uploaded, no accounts, no tokens. Works offline, works on private repos, works on repos that aren't on GitHub at all.
Read-only by design — you can't edit code from inside Review, and that's the point.
Standalone by design — Review doesn't try to bundle an editor, terminal, or AI agent into the same window. Those tools are evolving fast and you probably already have ones you like. Review does one thing well and stays out of the way.
Some diffs need a real window.
Terminal diffs are fine for small changes. But when you're reviewing 30 files with hundreds of hunks, you need spatial navigation — a sidebar, tabs, a minimap. That's not a TUI problem. That's a GUI problem.
Compare different branches in separate windows, each with independent review state. Auto-reloads when files change on disk. Still keyboard-driven:
Starts in your terminal.
Review has a full CLI alongside the desktop app. Check status, classify hunks, manage trust patterns, or open the GUI — all from the command line.
Open source · macOS · Experimental