Self-hosted AI code review for GitHub pull requests.
Multi-runtime (Cloudflare Workers, Node.js 24, Docker), queue-backed, repository-aware, and built for teams that want to own their review engine.
Website | Docs | Installation | Cloudflare | Docker | Node.js | Issues | Contributing
Codra listens to GitHub pull request events, runs AI-powered review jobs, posts inline findings back to the PR, and gives you a dashboard to inspect jobs, repositories, model routing, review history, and failed queue runs.
Beta - Codra is under active development. Cloudflare Worker is stable for self-hosted use; Node/Docker standalone is newer and still being hardened. Expect rough edges. Bugs and feedback welcome via GitHub Issues.
Docker — the fastest way to run Codra locally or on your own server:
docker run -p 3000:3000 devarshishimpi/codraFor a full setup with PostgreSQL and Redis included, use the Docker Compose file from the Docker guide:
curl -O https://codra.run/docker-compose.yml
docker compose upCloudflare Workers — deploy to the edge in minutes:
npm install -g wrangler
git clone https://github.com/devarshishimpi/codra.git && cd codra
npm install
npm run setup:cloudflare- Own the whole review loop: Run the GitHub App, web server, queue, database, model credentials, and dashboard under your own control (deploy on Cloudflare Workers, Node.js 24, or your own server via Docker).
- Review with repository context: Codra checks pull request diffs for correctness, security, performance, maintainability, and repo-specific patterns.
- Configure each repository: Tune triggers, skipped paths, draft handling, mention reviews, labels, custom rules, and review budgets from the dashboard.
- Route models deliberately: Use global defaults, per-repo model chains, fallbacks, and size-based overrides for larger pull requests.
- Operate the system: Inspect job history, PR findings, webhook deliveries, model usage, and dashboard stats.
- Automatic reviews on
opened,synchronize,ready_for_review, andreopenedpull request events - Mention-triggered reviews for on-demand analysis
- Inline GitHub review comments plus summary reviews and check run updates
- Queue-backed processing through Cloudflare Queues or BullMQ + Redis
- GitHub OAuth dashboard authentication
- PostgreSQL storage with support for Cloudflare Hyperdrive or direct connection pooling
- Dashboard-managed LLM providers for OpenAI, OpenRouter, Anthropic, Google, NVIDIA, and Cloudflare models
- Repository settings for labels, skipped globs, custom rules, and model routing
- GitHub sends Codra a pull request webhook.
- Codra verifies the signature and loads repository review settings.
- A review job is stored in PostgreSQL and queued on Cloudflare Queues or BullMQ.
- The review worker consumes the job, fetches the PR diff, runs model review passes, and formats findings.
- Codra posts inline comments and a summary review back to GitHub.
- The dashboard keeps the job history, findings, logs, and stats available for operators.
- Server / Runtime: Cloudflare Workers, Node.js 24, Docker (Distroless), Hono, Wrangler
- Dashboard: React, Vite, Tailwind CSS, Radix UI, Recharts
- Data: PostgreSQL, Cloudflare Hyperdrive, Cloudflare KV, Redis
- Queues: Cloudflare Queues, Workflows, BullMQ
- Models: OpenAI, OpenRouter, Anthropic, Google, NVIDIA, and Cloudflare providers
- GitHub: GitHub App webhooks, checks, reviews, and OAuth
- Quality: TypeScript, Zod, Vitest, Playwright browser tests
The full setup and operations guides live at codra.run/docs.
- Installation Prerequisites
- Local Development Guide
- Configuration Guide
- LLM Providers Guide
- Contributing
- Security Policy
New contributors are very welcome! We label well-scoped, beginner-friendly tasks — each issue includes which file to look at, what "done" looks like, and why it matters.
Browse good first issue and help wanted labels to find your first contribution.
- Discord: https://discord.gg/qkJmc8hHrC
- Issues: https://github.com/devarshishimpi/codra/issues
- Discussions: https://github.com/devarshishimpi/codra/discussions
- X/Twitter: https://x.com/devarshishimpi
Contributions are welcome. Please read CONTRIBUTING.md before opening a pull request against main. Codra uses a Contributor License Agreement for contributions.
Codra is licensed under the GNU Affero General Public License v3.0.
