iReview
AI codes. I review.
Review AI-generated code locally — and give Claude Code, Cursor, Copilot, or any coding agent precise instructions to fix it.
Comment on the diff, paste the review back telling the agent to fix it — done.
Why iReview?
vs git diff
You can actually leave comments
The terminal lets you read; iReview lets you mark up. Typed must-fix / suggestion / note comments per line, structured for copy-back.
vs the agent self-reviewing
Don't grade your own homework
The agent that wrote the code is not the right reviewer. iReview keeps you in the loop where you actually catch the deadlocks, races, and silly errors.
vs a GitHub PR
Works pre-commit, no push needed
Review staged + unstaged + recent commits as one diff, locally, before anything leaves your machine. Faster loop with the agent.
Get started in 30 seconds
Install once, Run from any repo
macOS & Linux
Install via Homebrew, then run from any repo.
brew tap suhothayan/tap brew install ireview
cd /path/to/your/repo ireview
Tap source: Formula/ireview.rb ↗
Windows
Paste into PowerShell — downloads the latest release and adds it to your user PATH. Open a new terminal afterwards.
irm https://suhothayan.github.io/iReview/install.ps1 | iex
cd C:\path\to\your\repo ireview
MIT licensed · runs 100% locally · no telemetry.
Find all downloads on the GitHub Releases page.
How it works
Pick changes
View commits, staged, and unstaged work as a single diff.
Review like a PR
GitHub-style diff viewer with file tree, two view modes, and mark-as-reviewed.
Add comments
Add must-fix, suggestion, or a note comment on a line or block of code.
Tell the agent what to fix
One-click copy of all your comments — paste into Claude Code, Cursor, or Copilot.
Review output for AI agents
The exported review is structured Markdown optimised for pasting into AI agent chats — numbered for easy reference (you can ask the agent to "address comment #2").
I reviewed your code and have the following comments. Please address them. Comment types: MUST FIX (must be addressed), SUGGESTION (improvements), NOTE (observations) 1. [MUST FIX] - `src/db.ts:42-50` - this transaction can deadlock under contention 2. [SUGGESTION] - `src/auth.ts:12` - add a more specific error message 3. [SUGGESTION] - `src/auth.ts:50` - prefer Result<> over throwing here the calling layer would rather branch on a typed result than wrap a try/catch 4. [NOTE] - `src/auth.ts` - consider extracting this into a dedicated module
FAQ
Does iReview send my code anywhere?
No. The diff stays on your machine. The server runs on
127.0.0.1, talks only to your local
git, and there is no telemetry or analytics. See
server/index.js.
Is it free? Will it stay free?
Yes. MIT licensed, open source, no paid tier, no signup. Run it forever.
Why not just use my agent's built-in review?
Letting the agent that wrote the code grade its own homework tends to miss the issues you'd catch. iReview keeps a human in the loop — fast — without forcing you to leave the terminal loop.
Does it work on Windows?
Yes — install via the PowerShell one-liner above. The same comment / copy-review flow works identically.
Can I use it for non-AI code reviews?
Yes. Nothing about iReview is AI-specific — it's a local diff reviewer. The Markdown export just happens to be very useful for AI agents. Pre-commit reviews, self-reviews before pushing, and quick sanity checks all work fine.