OpenReview

OpenReview is an open-source, self-hosted AI code review bot built by Vercel Labs. It deploys to Vercel, connects via a GitHub App, and triggers on-demand pull request reviews powered by Anthropic's Claude Sonnet 4.6. The project was built as an internal tool to help the Vercel team test their technologies together and is currently in public beta.

What It Is

OpenReview sits in the AI-assisted code review category. Developers mention @openreview in any GitHub PR comment to trigger a review — the bot then spins up an isolated sandbox, clones the repository on the PR branch, runs the Claude-powered agent against the diff, and posts inline suggestions directly in the pull request. It can also commit and push fixes (formatting, lint errors, simple bugs) back to the branch automatically.

How the Workflow Operates

The end-to-end flow is driven by GitHub webhooks and Vercel's durable execution primitives:

1. A developer comments @openreview on a PR (optionally with specific instructions like "check for security vulnerabilities" or "run the linter and fix any issues").
2. The webhook handler receives the event and starts a Vercel Workflow run.
3. A Vercel Sandbox is created, the repo is cloned on the PR branch, and dependencies are installed.
4. The Claude agent reviews the diff, reads files, runs linters and tests, and posts line-level comments with GitHub suggestion blocks.
5. If the agent makes changes, they are committed and pushed to the PR branch automatically.
6. The sandbox is torn down, and developers can react with 👍 or ❤️ to apply suggestions or 👎 / 😕 to skip them.

Progressive Skill System

OpenReview ships with a built-in skill registry and supports custom skills via .agents/skills/. The agent loads only the specialized instructions relevant to a given review, keeping the context window focused:

• 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 for component refactoring
• vercel-react-best-practices — React and Next.js performance optimization guidelines
• vercel-react-native-skills — React Native and Expo best practices
• web-design-guidelines — UI code review for Web Interface Guidelines and accessibility compliance

Custom skills are defined as Markdown files with YAML frontmatter in .agents/skills/, and the agent discovers them at runtime.

Tech Stack and Architecture

OpenReview is built on a modern TypeScript/Next.js stack with Vercel-native infrastructure:

• Next.js — App framework and route handlers
• Vercel Workflow — Durable, resumable execution
• Vercel Sandbox — Isolated code execution environment with full repo access
• AI SDK — AI model integration (Claude Sonnet 4.6 via Anthropic)
• Chat SDK — GitHub webhook handling
• Octokit — GitHub API client
• Redis (optional) — Persistent state; falls back to in-memory

Current Status

OpenReview is currently in beta. The project page explicitly notes it was built as an internal Vercel Labs project and that users should "expect rough edges and breaking changes." The GitHub repository under vercel-labs was created in early 2026 and has accumulated over 1,500 stars, indicating meaningful community interest. The MIT license applies to the source code, and the project is freely self-hostable by anyone with a Vercel account and an Anthropic API key.