duo-review-agent

AI-powered code reviews on every merge request –automatically

Comment

Inspiration

Every developer knows the feeling — you open a pull request, and it sits there for hours waiting for someone to review it. Code review is one of the most critical parts of the software development lifecycle, yet it's also one of the most time-consuming and inconsistent. We wanted to eliminate that bottleneck entirely by building an agent that never sleeps, never gets tired, and reviews every merge request the moment it's submitted.

What it does

duo-review-agent is an AI-powered code review agent built on the GitLab Duo Agent Platform. The moment a developer opens a Merge Request, the agent automatically analyzes code changes for bugs and logic errors, scans for security vulnerabilities, identifies bad practices, posts inline comments on problematic lines, labels the MR based on findings severity, and assigns the appropriate reviewer — all without any human intervention.

How we built it

The core flow is registered in the GitLab AI Catalog using GitLab Duo's native tools. The backend is powered by Python, FastAPI, and a multi-agent system using LangChain and CrewAI — with three specialized agents handling bug detection, security scanning, and code style review. Results are stored in PostgreSQL with Redis for caching. The frontend dashboard is built with React and TypeScript, deployed on Vercel.

Challenges we ran into

The GitLab Duo Agent Platform schema validation was strict and undocumented in places — we went through several failed pipelines before getting the flow structure right. Integrating the multi-agent CrewAI system with GitLab's API also required careful orchestration to avoid duplicate comments on MRs.

Accomplishments that we're proud of

  • Published a working flow to the GitLab AI Catalog
  • Built a full multi-agent review pipeline handling bug detection, security scanning, and style review in parallel
  • Delivered a clean, responsive frontend dashboard visualizing agent activity in real time

What we learned

Building on GitLab Duo taught us how powerful ambient AI agents can be when deeply integrated into a developer's existing workflow. The difference between an AI that answers questions and an AI that takes action is enormous.

What's next for duo-review-agent

  • Auto-fix simple issues directly in the codebase
  • Support for more languages and frameworks
  • Team analytics showing code quality trends over time

Built With

Share this project:

Updates