CodeAir

AI-Powered Code Review Platform for GitLab

Overview

CodeAir is a web application that brings AI-powered automation to your GitLab merge request workflow. It enables teams to configure intelligent agents that automatically generate merge request descriptions and provide code review suggestions using Anthropic Claude models.

Key Features

🔍 Project Discovery

Search and connect your GitLab projects to CodeAir. Simply add the CodeAir bot user as a Maintainer to your repository to enable AI capabilities.

🤖 AI Agents

Configure specialized agents for your projects:

• 📝 MR Description Writer — Automatically generates comprehensive descriptions for merge requests, summarizing changes, impact, and context.

• 🔍 MR Code Reviewer — Analyzes code changes and provides inline suggestions, identifying potential bugs, improvements, and best practices.

⚙️ Flexible Configuration

• Customize agent behavior with custom prompts
• Enable/disable agents per project
• Support for external webhook engines for custom integrations

📊 Job Monitoring

• Track agent execution history with detailed run logs
• View stdout/stderr output for debugging
• Monitor execution duration and status

Architecture

Component	Technology
Frontend	React, TypeScript, Bulma CSS
Backend API	Python, Litestar
Worker	Async Python
AI Engine	PR-Agent v0.29 (last Apache 2.0 licensed version)
Authentication	GitLab OAuth
AI Providers	Anthropic Claude

Setup

Prerequisites

1. GitLab Bot Account — Create a dedicated GitLab user for CodeAir and generate a Personal Access Token with api scope
2. GitLab OAuth Application — Register an OAuth application in GitLab with read_user scope
3. Anthropic API Key — Obtain from console.anthropic.com (configured per-agent in the UI)

Quick Start with Docker Compose

1. Clone the repository:

git clone https://github.com/tsv1/codeair.git
cd codeair

2. Create a .env file:

# Application
APP_ENCRYPTION_KEY=<32-character-random-string>
APP_WEBHOOK_BASE_URL=https://your-codeair-domain.com
JWT_SECRET_KEY=<64-character-random-string>

# Database
DATABASE_URL=postgresql://codeair:dev_aQUCVlDkUk@postgres:5432/codeair

# GitLab Bot
CODEAIR_BOT_TOKEN=glpat-xxxxxxxxxxxxxxxxxxxx

# GitLab OAuth
GITLAB_API_BASE_URL=https://gitlab.com  # or your self-hosted GitLab URL
GITLAB_OAUTH_CLIENT_ID=<your-oauth-client-id>
GITLAB_OAUTH_CLIENT_SECRET=gloas-<your-oauth-client-secret>
GITLAB_OAUTH_REDIRECT_URI=https://your-codeair-domain.com/auth/callback

3. Start the services:

docker-compose up -d

4. Access CodeAir at http://localhost:8080

Alternative Deployment

The CodeAir image (ghcr.io/tsv1/codeair:0.3.0) can be deployed anywhere (Kubernetes, VM, etc.) by running two containers with the same environment variables.

How It Works

1. Authenticate with your GitLab account
2. Search for projects where the bot has access
3. Configure AI agents with your API keys and preferences
4. Receive automatic descriptions and reviews on new merge requests

---

CodeAir streamlines your code review process, helping teams ship better code faster.