⚔ WARROOM

Multi-Agent AI Debate Arena

Tell it anything. Four AI agents debate it. A judge gives you a verdict.

---

🔗 Live Links

Service	URL	Status
🌐 Frontend	warroom-frontend.vercel.app
⚙️ Backend API	warroom-1.onrender.com
📡 Health Check	/health

Note: Backend runs on Render free tier — first load may take 30–50s to wake up (cold start).

---

What it does

You type any topic. Four AI agents take over and debate it in real time:

Agent	Name	Role
▲ Proponent	AXIOM	Argues as strongly as possible for your topic
▼ Opponent	REFUTE	Finds every weakness and argues against
◆ Fact-Checker	VERITAS	Neutral auditor — detects fallacies, verifies claims
◉ Judge	ARBITER	Scores every turn, delivers a final verdict

Works for any topic

🐕  "Should I buy a dog?"
💼  "Should I quit my job and freelance?"
🏠  "Should I rent or buy a house right now?"
⚡  "Nuclear energy is essential for net-zero"
🤖  "AI development should be paused immediately"
🌍  "Social media does more harm than good"

---

✨ Key Features

• ⚡ Real-time streaming — Every agent's response streams token-by-token via WebSocket. Watch agents think and speak live.
• ⏸ Pause & interrupt — Pause any debate mid-turn. Inject evidence, challenge a claim, or redirect the conversation.
• 🔍 Live web search — Fact-Checker uses Tavily to search the web in real time, verifying claims as they're made.
• ⚠ Fallacy detection — Every turn scanned for 10+ fallacy types (Ad Hominem, Straw Man, False Dichotomy, etc.).
• 📊 Live metrics — Animated score bars, head-to-head battle view, and consensus arc update after every turn.
• 🧬 Consensus engine — Vector embeddings measure stance convergence. Debate ends when agents agree enough.
• ⏱ Time-travel & forking — Every round checkpointed. Jump back and fork a new debate branch from any point.
• 🏆 Judge verdict — Full verdict with winner, key arguments, fallacy summary, and concrete recommendation.
• 👤 Personal topic support — Detects personal decisions and adjusts agents to discuss practical, real-world impact.

---

Tech Stack

Frontend Next.js 14 (App Router) + TypeScript Pure CSS custom properties WebSocket real-time streaming Supabase Auth Deployed on Vercel

Backend FastAPI + Python LangGraph state machine Groq API (Llama 3.3 70B) Supabase (PostgreSQL + Auth) Redis (pause/interrupt state) Deployed on Render

---

Architecture

LangGraph State Machine

flowchart TD
    START --> round_start
    round_start --> proponent
    round_start --> opponent
    round_start --> fact_checker
    proponent --> moderator
    opponent --> moderator
    fact_checker --> moderator
    moderator --> consensus
    consensus -->|continue| round_start
    consensus -->|end| judge
    judge --> END

Loading

How it works

flowchart TD
    A[User enters topic] --> B[Next.js frontend\nPOST /debates]
    B --> C[FastAPI backend\nstarts LangGraph task]
    C --> D[4 agents debate in sequence\nPro → Opp → Fact → Mod]
    D --> E[Groq streams tokens\nLlama 3.3 70B]
    D --> F[Scoring runs parallel\nlogic + fallacies]
    E --> G[WebSocket pushes events live]
    F --> G
    G --> H[User sees live debate + verdict]

Loading

`` `

Project Structure

WarRoom/
├── frontend/                  # Next.js app
│   ├── app/
│   │   ├── page.tsx           # Landing page
│   │   ├── auth/              # Login + Signup
│   │   ├── debate/
│   │   │   ├── new/           # Create debate form
│   │   │   └── [id]/          # Live debate arena
│   │   └── history/           # Past debates
│   ├── components/
│   │   ├── agents/            # AgentCard
│   │   ├── dashboard/         # LiveMetricsPanel, ConsensusGauge
│   │   ├── debate/            # DebateFeed, InterruptModal, DebateControls
│   │   ├── graph/             # GraphCanvas (React Flow)
│   │   ├── timeline/          # DebateTimeline, ForkModal
│   │   └── ui/                # Button, Badge, Tooltip
│   ├── hooks/
│   │   ├── useAuth.ts         # Supabase session
│   │   ├── useDebate.ts       # Debate state + WS events
│   │   └── useCheckpoints.ts  # Time-travel / fork
│   └── lib/
│       ├── api.ts             # FastAPI client
│       ├── supabase.ts        # Supabase client
│       ├── types.ts           # TypeScript interfaces
│       └── constants.ts       # Shared constants
│
└── backend/                   # FastAPI app
    ├── main.py                # Entry point + CORS
    ├── models/schemas.py      # Pydantic models
    ├── routers/
    │   ├── debates.py         # REST endpoints
    │   └── ws.py              # WebSocket manager
    ├── agents/
    │   ├── graph.py           # LangGraph state machine
    │   ├── consensus.py       # Vector embedding engine
    │   ├── prompts.py         # System prompts
    │   └── scoring.py         # Per-turn scoring
    └── services/
        ├── groq_client.py     # Groq API
        ├── supabase_client.py # DB operations
        └── redis_client.py    # Ephemeral state

---

Getting Started

Prerequisites

• Node.js 18+
• Python 3.11+
• Supabase project (free)
• Groq API key (free)
• Tavily API key (optional — enables web search)

---

1. Clone the repo

git clone https://github.com/harshitayadavv/warroom.git
cd warroom

2. Database setup

1. Go to your Supabase project → SQL Editor → New query
2. Paste the contents of backend/supabase_schema.sql
3. Click Run

3. Frontend

cd frontend
npm install

Create frontend/.env.local:

NEXT_PUBLIC_SUPABASE_URL=https://your-project.supabase.co
NEXT_PUBLIC_SUPABASE_ANON_KEY=your-anon-key
NEXT_PUBLIC_API_URL=http://localhost:8000

npm run dev   # http://localhost:3000

4. Backend

cd backend
python -m venv venv
source venv/bin/activate   # Windows: venv\Scripts\activate
pip install -r requirements.txt

Create backend/.env:

GROQ_API_KEY=gsk_your_key
SUPABASE_URL=https://your-project.supabase.co
SUPABASE_SERVICE_KEY=your-service-role-key
REDIS_URL=redis://localhost:6379
TAVILY_API_KEY=tvly-your-key
CORS_ORIGINS=http://localhost:3000,https://warroom-frontend.vercel.app
DEBUG=true

uvicorn main:app --reload --port 8000   # http://localhost:8000

---

Deployment

Frontend → Vercel

Variable	Value
NEXT_PUBLIC_API_URL	https://warroom-1.onrender.com
NEXT_PUBLIC_SUPABASE_URL	Your Supabase URL
NEXT_PUBLIC_SUPABASE_ANON_KEY	Your anon key

Backend → Render

Variable	Value
GROQ_API_KEY	Your Groq key
SUPABASE_URL	Your Supabase URL
SUPABASE_SERVICE_KEY	Your service role key
CORS_ORIGINS	http://localhost:3000,https://warroom-frontend.vercel.app
TAVILY_API_KEY	Your Tavily key (optional)

Start command: uvicorn main:app --host 0.0.0.0 --port 8000

---

Groq Models

Agent	Model
Proponent, Opponent, Judge	llama-3.3-70b-versatile
Fact-Checker	llama-3.3-70b-versatile
Scoring	gemma2-9b-it (separate quota)

---

License

MIT — do whatever you want with it.

---

Built with ⚔ by Harshita Yadav