I also didn't want to pay $15-30/month for another AI subscription when I already have an amazing coding agent (Claude Code, OpenCode, Gemini, etc.) that's way better than what these review bots use.
LaReview works differently: you paste a PR link, it groups the changes by logical area (auth changes, API endpoints, migrations, UI updates) ordered by risk, then uses your coding agent to draft review comments that you edit before posting. Nothing posts automatically.
Example: a 40-file PR touching auth, API, and frontend becomes 3 reviewable chunks. You start with auth (highest risk), draft 2-3 focused comments, then move to the next chunk. No 40-comment dump.
How it works: 1. Paste a GitHub/GitLab PR link or unified diff 2. AI analyzes the changes and proposes a review plan grouped by area and risk 3. You open each task and see only the relevant code hunks 4. AI drafts potential feedback for that specific task 5. You edit, approve, or delete each suggestion before posting to GitHub/GitLab (or export as Markdown)
Install:
macOS: brew install --cask puemos/tap/lareview
Linux: brew install puemos/tap/lareview
Windows: not yet (let me know if you need it)
LaReview runs locally and fetches PR data directly from GitHub/GitLab - no LaReview server involved.
Why it's different from Copilot/CodeRabbit/PR-Agent: 1. They mostly comment file-by-file, LaReview groups changes by logical area first 2. They auto-post or make you filter spam, LaReview drafts suggestions you review 3. They often need cloud integrations, LaReview is local-first
License: MIT + Apache-2.0