pr-review-me 1.0.0

Automated GitHub PR code review agent powered by LangGraph and Dough.id (Llama 3.3)

🤖 AI Code Review Agent

Automated GitHub PR analysis using 3 parallel AI agents.

Architecture • Features • Quick Start • API • Demo

---

⚡ Overview

The AI Code Review Agent is a full-stack application that acts as an automated, multi-disciplinary code reviewer. By pasting a GitHub Pull Request URL, the system fetches the code changes and processes them through three specialized AI agents in parallel.

Each agent has a specific focus:

• 🔒 Security Agent: Hardcoded secrets, SQL injection, XSS, insecure dependencies.
• 🚀 Performance Agent: N+1 queries, memory leaks, blocking I/O, O(n²) loops.
• ✨ Code Quality Agent: Code smells, SOLID violations, naming conventions, missing error handling.

The agents parse the PR diff, generate a structured JSON report, and synthesize a comprehensive markdown review that can be automatically posted directly to the GitHub PR.

---

🏗️ Architecture

The backend utilizes FastAPI for high-performance async routing, while LangGraph orchestrates the multi-agent LLM workflow. Requests are routed through the Dough.id API (OpenAI compatible) to process using Meta's llama-3.3-70b-versatile model.

graph TD
    %% Styling
    classDef frontend fill:#2d3748,stroke:#4a5568,stroke-width:2px,color:#fff
    classDef backend fill:#2c5282,stroke:#4299e1,stroke-width:2px,color:#fff
    classDef agent fill:#702459,stroke:#d53f8c,stroke-width:2px,color:#fff
    classDef external fill:#276749,stroke:#48bb78,stroke-width:2px,color:#fff

    %% Nodes
    UI[🖥️ React SPA Frontend]:::frontend
    API[⚡ FastAPI Backend]:::backend
    GH_API[🐙 GitHub API]:::external
    DOUGH[🧠 Dough.id / Groq LLM]:::external
    
    subgraph LangGraph Orchestration
        SEC[🔒 Security Agent]:::agent
        PERF[🚀 Performance Agent]:::agent
        QUAL[✨ Code Quality Agent]:::agent
        SYNTH[📝 Markdown Synthesizer]:::agent
    end

    %% Connections
    UI -- "POST /review (PR URL)" --> API
    API -- "Fetch Diff & Meta" --> GH_API
    GH_API -- "Raw Git Diff" --> API
    API -- "asyncio.gather()" --> SEC & PERF & QUAL
    
    SEC -. "JSON Review" .-> DOUGH
    PERF -. "JSON Review" .-> DOUGH
    QUAL -. "JSON Review" .-> DOUGH
    
    SEC & PERF & QUAL --> SYNTH
    SYNTH -- "Combined Report" --> API
    API -- "Post Comment (Optional)" --> GH_API
    API -- "JSON + Markdown" --> UI

---

✨ Features

• Parallel Execution: All three AI agents run simultaneously using asyncio.gather() to minimize latency.
• Rate-Limit Resilient: Employs tenacity for exponential backoff retries and staggers agent starts by 2 seconds to avoid 429 errors.
• Robust Parsing: Diff truncation ensures LLM token limits are respected (~8k char limit). Robust JSON extraction ignores markdown code fences.
• Dark Glassmorphism UI: A premium, responsive React dashboard built with Vite.
• Cloudflare Bypass: Configured with custom User-Agent headers to securely route through Dough.id's protected API gateway.

---

🚀 Quick Start

You can run this project in two ways: as a Standalone CLI or as a Full-Stack Web App.

1. Prerequisites

• Python ≥ 3.11
• Dough.id API Key (or Groq/OpenAI compatible key)
• GitHub Personal Access Token (Optional for public repos, required to post comments)
• Node.js ≥ 18 (Only required for the Web App)

---

Option A: Use as a CLI Package

The fastest way to use the code reviewer is directly from your terminal. You do not need to clone this repository.

# 1. Install globally via pip
pip install pr-review-me

# 2. Set your API key (or pass it directly via --dough-api-key)
export DOUGH_API_KEY="sk-your-key-here"

# 3. Run the review against any public PR!
pr-review-me https://github.com/django/django/pull/21523

# (Add --post-comment to automatically post the review back to GitHub)
pr-review-me https://github.com/django/django/pull/21523 --post-comment

---

Option B: Run the Web Dashboard

If you prefer a visual interface, you can boot up the FastAPI backend and React frontend.

1. Backend Setup

# Enter the directory
cd pr_rev

# Create and activate a virtual environment
python -m venv .venv
.venv\Scripts\activate      # Windows
source .venv/bin/activate   # macOS / Linux

# Install dependencies
pip install -r requirements.txt

3. Environment Variables

Create a .env file in the root directory:

# API Keys
DOUGH_API_KEY=sk-your-dough-api-key
GITHUB_TOKEN=ghp_your_github_token_here

4. Frontend Setup

cd frontend
npm install
cd ..

5. Running the Application

Open two terminals to run the services concurrently:

Terminal 1 — Backend (FastAPI)

.venv\Scripts\activate
uvicorn main:app --reload --port 8000

Terminal 2 — Frontend (Vite)

cd frontend
npm run dev

Navigate to http://localhost:5173 to access the dashboard.

---

📖 API Reference

GET /health

Liveness probe to verify the backend is running.

{ "status": "ok", "version": "1.0.0" }

POST /review

Executes the full multi-agent review pipeline.

Request:

{
  "pr_url": "https://github.com/owner/repo/pull/123",
  "post_comment": true
}

Response:

{
  "success": true,
  "pr_title": "Add user authentication",
  "pr_author": "octocat",
  "security": { "agent_name": "Security Agent", "issues": [], "summary": "..." },
  "performance": { "agent_name": "Performance Agent", "issues": [], "summary": "..." },
  "code_quality": { "agent_name": "Code Quality Agent", "issues": [], "summary": "..." },
  "markdown_comment": "# 🤖 AI Code Review Report\n...",
  "comment_posted": true,
  "comment_url": "https://github.com/owner/repo/pull/123#issuecomment-12345",
  "total_issues": 4,
  "critical_count": 0,
  "duration_seconds": 12.4
}

---

🛠️ Project Structure

pr_rev/
├── main.py                # FastAPI Application & Endpoints
├── agents.py              # LangGraph Agents & Dough.id Integration
├── github_utils.py        # GitHub REST API interactions
├── requirements.txt       # Python dependencies
├── .env                   # Environment variables
└── frontend/
    ├── src/
    │   ├── App.jsx        # Main React Dashboard Component
    │   ├── App.css        # Glassmorphism Styling
    │   └── main.jsx       # React Entry Point
    ├── package.json       # Node dependencies
    └── vite.config.js     # Vite configuration & proxy

---

🔧 Troubleshooting

Issue	Resolution
401 Unauthorized (GitHub)	Verify GITHUB_TOKEN is set in .env and has repo scope.
404 Not Found (GitHub)	Ensure the PR URL is formatted correctly: https://github.com/owner/repo/pull/NNN.
403 / Blocked (Dough.id)	Make sure User-Agent is configured in ChatOpenAI headers to bypass Cloudflare.
CORS Errors	Ensure the backend is running on port 8000. Vite automatically proxies /review to it.
Empty Diff	The PR might only contain merge commits without code changes.