Debate Crew

AI-powered debate system using crewAI. Two agents (debater and judge) engage in a structured debate on any given motion, followed by a final decision from the judge.

Features

• Debater Agent: Presents persuasive arguments for or against a given motion
• Judge Agent: Evaluates arguments and decides which side is more compelling
• Sequential Process: Propose → Oppose → Decide workflow
• Spanish Language: Fully localized interface
• Configurable: Easy YAML configuration for agents and tasks
• Multiple Modes: Run, train, test, and replay crew executions

Installation

Ensure you have Python >=3.10 <3.14 installed on your system. This project uses UV for dependency management.

# Install UV
pip install uv

# Install dependencies
uv sync

Configuration

1. Add your OPENAI_API_KEY to the .env file
2. Configure agents in src/debate/config/agents.yaml
3. Configure tasks in src/debate/config/tasks.yaml

Usage

Run a Debate

# Run with default motion
crewai run

# Run with custom motion
crewai run --motion "¿Deberían las IA ser reguladas por el gobierno?"

The system executes three sequential tasks:

1. Propose (propose.md) - Arguments in favor of the motion
2. Oppose (oppose.md) - Arguments against the motion
3. Decide (decide.md) - Judge's final decision

Output files are saved to the output/ directory.

Additional Commands

# Train the crew (n_iterations, filename)
crewai train 10 results.json

# Test the crew (n_iterations, eval_llm)
crewai test 5 gpt-4

# Replay a specific task
crewai replay task_id

# Run with custom trigger payload
crewai run --trigger '{"motion": "custom topic here"}'

Project Structure

debate/
├── src/debate/
│   ├── config/
│   │   ├── agents.yaml      # Agent definitions (debater, judge)
│   │   └── tasks.yaml       # Task definitions (propose, oppose, decide)
│   ├── crew.py              # Crew orchestration class
│   └── main.py              # Entry point with run/train/replay/test functions
├── output/                  # Generated debate reports
├── .env                     # Environment variables (API keys)
├── pyproject.toml           # Project dependencies
└── README.md               # This file

Agent Configuration

Debater Agent

• Role: Convincing debater
• Goal: Present clear arguments for or against the motion
• LLM: openai/gpt-4o-mini

Judge Agent

• Role: Decide the winner based on arguments
• Goal: Evaluate both sides and determine which is more compelling
• LLM: openai/gpt-4o-mini

Workflow

┌─────────────────┐
│   Propose       │  Debater presents arguments IN FAVOR
└────────┬────────┘
         ↓
┌─────────────────┐
│   Oppose        │  Debater presents arguments AGAINST
└────────┬────────┘
         ↓
┌─────────────────┐
│   Decide        │  Judge evaluates and makes final decision
└─────────────────┘
         ↓
    output/*.md

License

MIT