OpenReview

An open-source, self-hosted AI code review bot. Deploy to Vercel, connect a GitHub App, and get on-demand PR reviews powered by Claude.

Beta: OpenReview is currently in beta. It was built as an internal project to help the Vercel team test their technologies together. Expect rough edges and breaking changes.

Features

On-demand reviews — Mention @openreview in any PR comment to trigger a review. Powered by Chat SDK
Sandboxed execution — Runs in an isolated Vercel Sandbox with full repo access, including the ability to run linters, formatters, and tests
Inline suggestions — Posts line-level comments with GitHub suggestion blocks for one-click fixes
Code changes — Can directly fix formatting, lint errors, and simple bugs, then commit and push to your PR branch
Reactions — React with 👍 or ❤️ to approve suggestions, or 👎 or 😕 to skip
Durable workflows — Built on Vercel Workflow for reliable, resumable execution
Extensible skills — Ships with built-in review skills and supports custom skills via .agents/skills/
Powered by Claude — Uses Claude Sonnet 4.6 via the AI SDK for high-quality code analysis
Simple route handler — Easily define route handlers using Next.js Route Handlers for custom API endpoints and webhooks

How it works

sequenceDiagram
    participant U as Developer
    participant GH as GitHub
    participant WH as Webhook Handler
    participant WF as Vercel Workflow
    participant SB as Vercel Sandbox
    participant AI as Claude Agent

    U->>GH: @openreview in PR comment
    GH->>WH: Webhook event
    WH->>WF: Start workflow

    WF->>GH: Check push access
    WF->>SB: Create sandbox
    SB->>SB: Clone repo on PR branch
    SB->>SB: Install dependencies
    SB->>SB: Configure git

    WF->>AI: Run agent with PR context
    AI->>SB: Read files, run linters, explore code
    SB-->>AI: Command output
    AI->>GH: Post inline comments & suggestions
    AI-->>WF: Agent complete

    WF->>SB: Check for uncommitted changes
    alt Changes made
        WF->>SB: Commit & push to PR branch
        SB->>GH: Push changes
    end
    WF->>SB: Stop sandbox

    U->>GH: React 👍 or ❤️ on suggestion
    GH->>WH: Reaction event
    WH->>WF: Start new workflow run

Mention @openreview in a PR comment (optionally with specific instructions)
OpenReview spins up a sandboxed environment and clones the repo on the PR branch
A Claude-powered agent reviews the diff, explores the codebase, and runs project tooling
The agent posts its findings as PR comments with inline suggestions
If changes are made (formatting fixes, lint fixes, etc.), they're committed and pushed to the branch
The sandbox is cleaned up

Setup

1. Deploy to Vercel

Click the button above or clone this repo and deploy it to your Vercel account.

2. Create a GitHub App

Create a new GitHub App with the following configuration:

Webhook URL: https://your-deployment.vercel.app/api/webhooks

Repository permissions:

Contents: Read & write
Issues: Read & write
Pull requests: Read & write
Metadata: Read-only

Subscribe to events:

Issue comment
Pull request review comment

Generate a private key and webhook secret, then note your App ID and Installation ID.

3. Configure environment variables

Add the following environment variables to your Vercel project:

Variable	Description
`ANTHROPIC_API_KEY`	API key for Claude
`GITHUB_APP_ID`	The ID of your GitHub App
`GITHUB_APP_INSTALLATION_ID`	The installation ID for your repository
`GITHUB_APP_PRIVATE_KEY`	The private key generated for your GitHub App (with `\n` for newlines)
`GITHUB_APP_WEBHOOK_SECRET`	The webhook secret you configured
`REDIS_URL`	(Optional) Redis URL for persistent state, falls back to in-memory

4. Install the GitHub App

Install the GitHub App on the repositories you want OpenReview to monitor. Once installed, mention @openreview in any PR comment to trigger a review.

Usage

Trigger a review: Comment @openreview on any PR. You can include specific instructions:

@openreview check for security vulnerabilities
@openreview run the linter and fix any issues
@openreview explain how the authentication flow works

Reactions: React with 👍 or ❤️ on an OpenReview comment to approve and apply its suggestions. React with 👎 or 😕 to skip.

Skills

OpenReview uses a progressive skill system — the agent only loads specialized instructions when relevant, keeping context focused and reviews thorough. Skills are discovered from .agents/skills/ at runtime.

Built-in skills

Skill	Description
`next-best-practices`	File conventions, RSC boundaries, data patterns, async APIs, metadata, error handling
`next-cache-components`	PPR, `use cache` directive, `cacheLife`, `cacheTag`, `updateTag`
`next-upgrade`	Upgrade Next.js following official migration guides and codemods
`vercel-composition-patterns`	React composition patterns that scale for component refactoring
`vercel-react-best-practices`	React and Next.js performance optimization guidelines
`vercel-react-native-skills`	React Native and Expo best practices for performant mobile apps
`web-design-guidelines`	Review UI code for Web Interface Guidelines and accessibility compliance

Adding custom skills

Create a folder in .agents/skills/ with a SKILL.md file containing YAML frontmatter:

.agents/skills/
└── my-custom-skill/
    └── SKILL.md

---
name: my-custom-skill
description: When to use this skill — the agent reads this to decide whether to load it.
---

# My Custom Skill

Your specialized review instructions here...

The agent sees only skill names and descriptions in its system prompt. When a request matches a skill, it calls loadSkill to get the full instructions — keeping the context window clean.

Tech stack

Next.js — App framework
Vercel Workflow — Durable execution
Vercel Sandbox — Isolated code execution
AI SDK — AI model integration
Chat SDK — GitHub webhook handling
Octokit — GitHub API client

Development

bun install
bun dev

License

MIT