Dialectica

Dialectica is a lightweight MVP that orchestrates a structured debate between two AI agents (Proponent and Critic) and delivers a judged evaluation highlighting the strongest arguments and remaining trade-offs.

Features

• Start a debate on any topic through a single input field.
• Watch the debate unfold turn-by-turn with live status updates.
• Each agent response includes citations backed by web search results.
• Automatic judge evaluation summarizing the strongest pro and con points with outstanding trade-offs.
• Copyable evaluation summary and full list of cited sources.

Prerequisites

• Node.js 18+ (for built-in fetch support).
• An OpenAI API key with access to GPT models.
• A SerpAPI key for real-time web search results (get one at serpapi.com).

Getting Started

1. Clone the repository and navigate into the project directory.
2. Set up your environment variables using one of the methods below.
3. Start the server:

Setting Environment Variables

Option 1: Create a .env file (Recommended)

# Create .env file in the project root
cat > .env << EOF
OPENAI_API_KEY=your_openai_api_key_here
OPENAI_MODEL=gpt-5-nano
OPENAI_MAX_TOKENS=2000
OPENAI_TEMPERATURE=0.7
SERPAPI_KEY=your_serpapi_key_here
PORT=3000
DIALECTICA_TURNS=3
EOF

Option 2: Command line

OPENAI_API_KEY=your_key_here node src/server.js

Option 3: Export in your shell

export OPENAI_API_KEY=your_key_here
export OPENAI_MODEL=gpt-5-nano
node src/server.js

4. Start the server:

npm start

5. Open http://localhost:3000 in your browser, enter a topic, and click Start Debate.

The server streams debate events to the browser using Server-Sent Events (SSE), so you'll see each agent's response and the judge's evaluation as soon as they are generated.

Environment Variables

Variable	Description	Default
OPENAI_API_KEY	Required. OpenAI API key used to call the debate and judge models.	-
OPENAI_MODEL	Optional. OpenAI model to use for debates and judging.	gpt-5-nano
OPENAI_MAX_TOKENS	Optional. Maximum tokens per response.	2000
OPENAI_TEMPERATURE	Optional. Response randomness (0.0-2.0).	0.7
DIALECTICA_TURNS	Optional. Number of turns per agent.	3
PORT	Optional. HTTP port for the server.	3000

Project Structure

├── public
│   ├── app.js          # Frontend logic for managing debates and rendering UI
│   ├── index.html      # Single-page application shell
│   └── styles.css      # Minimalist styling
├── src
│   └── server.js       # Node HTTP server and debate orchestrator
└── README.md

Notes

• The application requires outbound network access to the OpenAI API. Ensure the environment where you run the server permits HTTPS requests to api.openai.com.
• The debate agents expect to cite their sources using placeholders like [S1] that are automatically translated into global citation numbers for display and judging.
• The judge prompt reuses the debate transcript and shared citation list to ensure all conclusions remain grounded in the debate's evidence.
• You can use different OpenAI models by setting the OPENAI_MODEL environment variable (e.g., gpt-5-nano, gpt-4, gpt-3.5-turbo, etc.).
• Note: gpt-5-nano may not be available in all regions or API accounts. If you encounter errors, try using gpt-4o-mini or gpt-3.5-turbo instead.

License

This project is licensed under the ISC License.