Virtual-Debate-Room

# Virtual Debate Room

Open-source multi-agent AI debate engine. Submit a topic and 4 AI agents — Architect, Skeptic, Economist, Risk Officer — debate it in structured rounds. A Judge synthesizes the result with adversarial scoring and a confidence rating.

Built with FastAPI + React + OpenRouter.

---

## How it works

1. You submit a topic, goal, and constraints
2. 4 agents receive the same problem but think through different roles
3. Each agent uses a different LLM model via OpenRouter
4. The debate runs in 4 structured rounds in parallel
5. A Judge scores the debate quality and produces a final report

---

## Screenshots

### New Debate Form
![New Debate](examples/screenshots/new_debate.png)

### Live Debate Progress
![Live Debate](examples/screenshots/live_debate.png)

### Final Report
![Report 1](examples/screenshots/report_1.png)

![Report 2](examples/screenshots/report_2.png)

---

## Quick Start
```bash
git clone https://github.com/StanimirTenev/Virtual-Debate-Room.git
cd Virtual-Debate-Room

# Configure
cp .env.example .env
# Add your OPENROUTER_API_KEY to .env

# Install
cd backend && python3 -m venv .venv && source .venv/bin/activate && pip install -r requirements.txt && cd ..
cd frontend && npm install && cd ..

# Start
./scripts/dev.sh
```

Opens at **http://localhost:5173**

Full installation guide: [INSTALL.md](INSTALL.md)

---

## Model Presets

| Preset | Cost per debate | Models |
|--------|----------------|--------|
| Economy | ~$0.08–$0.17 | Gemini 2.5 Pro, Claude 3.7, Llama 3.3 70B, GPT-4.1 |
| Standard | ~$3–$6 | Top-tier models for all agents |
| Premium | ~$8–$15 | Highest-tier everywhere |

---

## Agents

| Agent | Role | Temperature |
|-------|------|-------------|
| 🏗 Architect | Proposes the solution | 0.30 |
| 🔍 Skeptic | Attacks with specific arguments | 0.65 |
| 💰 Economist | Evaluates cost and ROI | 0.25 |
| ⚠️ Risk Officer | Identifies failure scenarios | 0.35 |
| ⚖️ Judge | Scores debate quality and synthesizes | 0.10 |

---

## Tech Stack

**Backend:** Python 3.11, FastAPI, SQLAlchemy, httpx
**Frontend:** React 18, Vite, Tailwind CSS
**Models:** OpenRouter API (single key, multiple providers)

---

## Report Output

Each debate produces:
- **Confidence score** (0–100) — how certain the recommendation is
- **Adversarial score** (1–10) — how genuine the debate was
- **Strongest argument** — which claim survived all scrutiny
- **Weak agents** — who failed their role and why
- **Export** — Markdown, HTML, and JSON download

---

## Run Tests
```bash
cd backend
source .venv/bin/activate
python -m pytest tests/ -v
# 86 tests
```

---

## License

AGPL-3.0 — see [LICENSE](LICENSE)