iReview

AI codes. I review. — browser-based local diff review for AI-generated code.

Point it at a git repo, pick any combination of recent commits + uncommitted work, and review the resulting diff like a GitHub pull request. Leave typed comments (must-fix / suggestion / note) and click Copy review to paste a structured Markdown summary back to your coding agent — Claude Code, Cursor, Copilot, anything.

Website: https://suhothayan.github.io/iReview/

Why iReview

• vs git diff — you can actually leave typed comments instead of just reading.
• vs the agent self-reviewing — don't grade your own homework. A human catches the deadlocks, races, and dumb mistakes the agent confidently shipped.
• vs a GitHub PR — works pre-commit on staged + unstaged + recent commits, all locally, no push needed.

Demo

Install + run

Homebrew (macOS / Linux)

brew tap suhothayan/tap
brew install ireview

Then, from any subdirectory of a git repo:

ireview

It walks up looking for .git, listens on port 3737 by default, and opens your browser automatically. Use --port N to override, or pass an explicit path: ireview /path/to/your/repo.

Windows (PowerShell)

irm https://suhothayan.github.io/iReview/install.ps1 | iex

Downloads the latest release exe to %USERPROFILE%\.ireview\ and adds it to your user PATH. Open a new terminal afterwards, then ireview works from any git repo. You can view the install script before piping.

Other ways to install

Single binary

Download the binary for your OS from the Releases page and run it directly:

./ireview /path/to/your/repo

On macOS, Gatekeeper will block direct downloads — either install via Homebrew (above) or remove the quarantine flag manually:

xattr -d com.apple.quarantine ./ireview-macos-arm64

From source

git clone https://github.com/suhothayan/iReview
cd iReview
npm install
IREVIEW_REPO=/path/to/your/repo npm run dev

Vite dev server on :5173, Express API on :3737.

Requires: git installed on your system.

Features

• Pick any selection. Recent commits + uncommitted (staged / unstaged) all in one picker — combined into a single diff.
• GitHub-style diff viewer. Unified diff with hunk headers, file tree sidebar, sticky per-file headers, comment count badges.
• Click to comment. Click a line for a typed comment (must-fix / suggestion / note). Shift-click another line on the same side to extend a multi-line range. File-level and review-level comments too.
• Mark files reviewed. Per-file checkbox plus tri-state cascading checkboxes on directories.
• Two view modes. Single (one file at a time, with Prev / Next nav) or Scroll all (continuous scroll, sidebar follows scroll position).
• Copy review. Numbered, structured Markdown lands on your clipboard. Paste into your AI agent's chat to apply fixes.
• Light + dark theme. Honours prefers-color-scheme on first load, persists choice.
• Comments survive reloads. Per-repo, no signup, no cloud.
• Quit from the browser. Click the ⏻ in the toolbar to stop the server cleanly.

Command-line options

ireview [REPO_PATH] [--port N] [--no-open] [-f FROM] [-t TO]

Arguments:
  REPO_PATH       Path to a git repository (default: walks up from cwd)

Options:
  -p, --port N    Port to listen on (default: 3737)
      --no-open   Don't auto-open the browser
  -f, --from      One endpoint of the picker range (see below)
  -t, --to        The other endpoint (optional; defaults to "unstaged")
  -v, --version   Show version
  -h, --help      Show this help

Pre-selecting a range on launch

The picker's rows form a single ordered list — unstaged → staged → commits — and --from / --to pick a contiguous range across it. Most useful for AI coding agents that want to say "open iReview on what I just produced".

Either endpoint accepts unstaged, staged, or anything git rev-parse understands (HEAD, HEAD~N, branch names, full or short SHAs). A missing endpoint defaults to unstaged — so --from HEAD~3 means "everything since HEAD~3 including in-flight work", matching how every other git command treats a single ref.

ireview                                # default: all uncommitted work
ireview -f HEAD~3                      # last 3 commits + my dirty edits
ireview -f HEAD                        # last commit + my dirty edits
ireview -f staged                      # staged + unstaged (no commits)
ireview -f HEAD -t HEAD~2              # narrow: just commits, no dirty
ireview -f a1b2c3 -t HEAD              # one specific commit + everything up to HEAD

# Agent-style: review the SHAs you just produced
ireview --from "$(git log -3 --format=%H | tail -1)"

HEAD~N follows git's first-parent convention but the picker shows topo-order log, so --from HEAD~3 --to HEAD may fill in more than four commits if merges sit in between — same as what you'd see if you ticked those two rows in the picker UI.

Output format

The exported review is structured Markdown optimized for pasting into AI agent chats:

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

• Numbered for easy reference (you can ask the agent: "address comment #2")
• File-level comments show as `path`, line as `path:N`, range as `path:N-M`
• Legend only includes types you actually used
• Multi-paragraph bodies indent under the header so they stay inside the list item
• Comments sorted by file (review-level first), then by line

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's no telemetry or analytics. See server/index.js.

Is iReview free? Will it stay free? Yes. MIT licensed, open source, no paid tier, no signup.

Why not just use my agent's built-in review? Letting the agent grade its own homework tends to miss the issues you'd catch. iReview keeps a human in the loop without leaving the terminal flow.

Does it work on Windows? Yes — install via the PowerShell one-liner above.

Can I use it for non-AI code reviews? Yes. iReview is just a local diff reviewer — pre-commit reviews, self-reviews, sanity checks all work fine.

What's new in 0.3.0

• Picker now auto-applies. Every click in the picker writes to the selection immediately — no more Apply / Reset buttons. The diff refetches in the background; closing the picker lands you on an already-loaded diff.
• Unified commit picker. Unstaged, staged, and recent commits are one ordered list. Pick a contiguous range across all of it with calendar-style click rules: click outside the range to extend, click an endpoint to shrink, click inside to reset.
• CLI presets for AI agents. New --from / --to flags (with -f / -t short forms) let an agent launch iReview pointed at exactly the range it just produced. Endpoints accept any git ref plus staged / unstaged. See "Pre-selecting a range on launch" above.
• Refresh button picks up disk changes. Re-pulls /api/repo so a freshly staged file shows up in the picker without restarting the server.

Older releases: see the GitHub Releases page.

Security

iReview binds to 127.0.0.1 only and authenticates the shutdown endpoint with a per-boot token. See SECURITY.md to report a vulnerability.

Contributing

PRs welcome. See CONTRIBUTING.md for architecture notes, build instructions, and the project's npm scripts.

Run npm run typecheck and npm test before submitting; both should be green.

License

MIT — see LICENSE.