CodeReview Agent 🤖

AI-powered real-time collaborative code review with Fix-It Mode. Paste code or a GitHub PR URL → a LangGraph agent reviews it for bugs, security, style, and performance — streaming tokens live via SSE. When the review is done, hit Fix-It and the AI generates a fully corrected version of your code with a live diff, per-issue explanations, and a one-click download.

100% free to run. No Docker. No AWS. No Kafka. Runs fully offline with Ollama.

---

Live Demo

🌐 codereview-xi-sepia.vercel.app

Powered by Groq free tier — 50 reviews/day per user. To run unlimited and offline, use Ollama (see below).

---

Features

• 🤖 LangGraph agent — 4 parallel review nodes (bugs, security, style, performance)
• ⚡ Live SSE streaming — see each analysis node stream tokens in real time
• 🔁 Fix-It Mode — AI generates corrected code + streams a per-issue explanation with live diff, side-by-side, and fixed code views
• 📥 Download fixed file — one-click download of the fixed code as the correct file extension (.py, .ts, .go, etc.)
• 🔗 GitHub PR diff review — paste any public PR URL and the agent fetches + reviews the diff
• 📊 Analytics dashboard — issue trends by category, severity, and language
• 👥 Real-time collaboration — multiple users in the same review room via WebSocket + Redis Pub/Sub
• 🛡️ Rate limiting — 50 reviews/day per IP, protects Groq quota from abuse
• 🔐 JWT auth — register/login, all endpoints protected
• 🔄 Retry with backoff — automatic retry on Groq rate limits (5s → 10s → 20s) instead of failing
• 🦙 Ollama support — run fully offline with any local model, zero data leaves your machine

---

Stack

Layer	Tech	Cost
Frontend	Next.js 15	Vercel free tier
Backend	FastAPI + WebSockets	Railway free tier
AI Agent	LangGraph + Groq (llama-3.3-70b)	Groq free tier
Local AI	Ollama (optional)	Free / offline
Realtime	Redis Pub/Sub	Upstash free tier
Streaming	Server-Sent Events (SSE)	—
Database	PostgreSQL	Neon.tech free tier
CI/CD	GitHub Actions	Free

---

Free Services You Need

1. neon.tech → New Project → copy connection string → DATABASE_URL
2. upstash.com → Create Redis → copy Redis URL → REDIS_URL
3. console.groq.com/keys → New API Key → GROQ_API_KEY

---

LangGraph Agent Architecture

START
  └─► analyze_structure
        ├─► bug_check        ──┐
        ├─► security_check   ──┤  (parallel + SSE-streamed)
        ├─► style_check      ──┤
        └─► performance_check──┘
              └─► aggregate
                    └─► END

Fix-It Mode (separate flow, triggered manually):
  START → plan_fixes → generate_fixed_code (streamed) → explain_changes (per-issue, streamed) → END

---

Local Development

1. Backend

cd backend
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
# Fill in DATABASE_URL, REDIS_URL, GROQ_API_KEY
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

2. Frontend

cd frontend
npm install
echo "NEXT_PUBLIC_API_URL=http://localhost:8000" > .env.local
echo "NEXT_PUBLIC_WS_URL=ws://localhost:8000" >> .env.local
npm run dev

Open http://localhost:3000

---

Running Locally with Ollama (Zero API Cost)

The agent supports Ollama as a drop-in replacement for Groq. No internet required, no API key, no rate limits.

1. Install Ollama

# macOS / Linux
curl -fsSL https://ollama.com/install.sh | sh

# Windows: download from https://ollama.com/download

2. Pull a model

ollama pull llama3.2        # recommended — fast and accurate
# or
ollama pull codellama       # code-specialized alternative

3. Configure the backend

Add to backend/.env:

OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_MODEL=llama3.2
# Leave GROQ_API_KEY blank or remove it — Ollama takes priority

4. Install the Ollama LangChain package

cd backend
pip install langchain-ollama

Start the backend normally — it will automatically use Ollama instead of Groq.

How it works: get_llm() checks for OLLAMA_BASE_URL first. If set, uses ChatOllama. If not, falls back to ChatGroq. No other code changes needed.

---

Deployment (Free)

Backend → Railway

1. railway.app → New Project → Deploy from GitHub → select backend/
2. Add env vars (see .env.example)
3. Get your URL: https://codereview-xxx.up.railway.app

Frontend → Vercel

cd frontend
npx vercel
# Set env vars:
# NEXT_PUBLIC_API_URL=https://your-backend.up.railway.app
# NEXT_PUBLIC_WS_URL=wss://your-backend.up.railway.app

CI/CD — GitHub Actions

Add to Settings → Secrets:

Secret	Source
RAILWAY_TOKEN	Railway → Account Settings → Tokens
VERCEL_TOKEN	Vercel → Account Settings → Tokens
VERCEL_ORG_ID	vercel link → .vercel/project.json
VERCEL_PROJECT_ID	same file

---

API Reference

Method	Endpoint	Description
POST	/api/auth/register	Register
POST	/api/auth/login	Login → JWT
POST	/api/reviews	Create review (code paste)
POST	/api/reviews/from-pr	Create review from GitHub PR URL
GET	/api/reviews	List your reviews
GET	/api/reviews/stats	Trend metrics
GET	/api/reviews/{id}	Get review + issues
GET	/api/reviews/{id}/stream	SSE — stream live agent tokens
POST	/api/reviews/{id}/comments	Add comment
WS	/ws/review/{id}?token=JWT	Join real-time room
GET	/api/fix/{id}	Get review info for Fix-It
GET	/api/fix/{id}/stream	SSE — stream Fix-It progress

---

Environment Variables

See backend/.env.example. Required:

DATABASE_URL=postgresql+asyncpg://...neon.tech/...
REDIS_URL=rediss://...upstash.io:6379
GROQ_API_KEY=gsk_...
SECRET_KEY=<random 32-char string>
FRONTEND_URL=https://your-app.vercel.app

Optional:

GITHUB_TOKEN=ghp_...         # avoids GitHub rate limits; needed for private repos
OLLAMA_BASE_URL=http://localhost:11434   # enables local Ollama backend
OLLAMA_MODEL=llama3.2        # which Ollama model to use (default: llama3.2)