Debate

Pro

Two AI models argue opposing positions while a third model judges. Watch the debate unfold in real-time with progressive streaming.

Overview

Debate mode creates an adversarial discussion between two AI models. One model (Pro) argues for a position, another (Con) argues against, and a third model serves as an impartial Judge. This approach helps uncover nuances, strengthen arguments, and identify weaknesses in reasoning.

Models involved

Debate rounds

Final verdict

Key Features

Progressive Streaming

Watch the debate unfold in real-time. Each round appears as soon as it completes—no waiting for the full debate.

Customizable Personas

Choose different personas and models for Pro, Con, and Judge roles before starting the debate.

Impartial Judgment

A separate AI model evaluates both sides and delivers a reasoned verdict with clear winner declaration.

Multi-Model Support

Mix GPT-4, Claude, Gemini, and other models. Different models bring unique argumentation styles.

How It Works

Setup

Configure Pro, Con, and Judge roles. Choose models and personas for each participant, or use the defaults.

Opening Statements

Parallel

Pro and Con present their initial positions in parallel—both run simultaneously for faster setup. Opening statements are independent, so parallel execution is safe.

Debate Rounds (3 rounds)

Sequential

Within each round, Pro argues first, then Con responds directly to Pro's argument. This creates authentic back-and-forth debate where each side actually counters the other.

Judgment

The Judge model evaluates both sides and delivers a reasoned verdict, declaring a winner (Pro, Con, or Draw).

Graceful Degradation

Debate mode handles model failures gracefully at every stage:

Opening Failures

If Pro or Con fails during setup, their position shows the error. The debate continues with the working side, and the Judge accounts for the incomplete debate.

Round Failures

If a model fails during a round, that side's argument shows the error. The debate continues, and the Judge notes which side had technical difficulties.

Streaming Metadata

Each streaming event includes pro_success and con_success flags so the UI can show error states in real-time.

Debate Roles

↑ ProDefault: Advocate persona

Argues in favor of the proposition. Presents supporting evidence, benefits, and reasons why the position is correct.

↓ ConDefault: Critic persona

Argues against the proposition. Identifies weaknesses, presents counterarguments, and highlights risks or drawbacks.

⚖ JudgeDefault: Impartial Judge persona

Evaluates both sides fairly. Considers argument strength, evidence quality, and reasoning to declare a winner.

Example Output

Question

"Should remote work become the default for knowledge workers?"

↑ ProGPT-4o

"Remote work increases productivity by 13% according to Stanford research, reduces commute stress, and enables access to global talent pools..."

↓ ConGemini 2.0 Flash

"In-person collaboration drives innovation—studies show spontaneous interactions lead to 40% more creative breakthroughs. Remote work also risks employee isolation..."

⚖ JudgeClaude Sonnet 4.5

"Both sides present valid arguments. However, the Con side provides stronger evidence about innovation impacts. Winner: Con. A hybrid model may balance both concerns."

Follow-up Q&A

After a debate completes, you can ask follow-up questions about the arguments, verdict, or topic. The system automatically provides debate context to the AI.

How It Works

• After a debate ends, simply type your question in the chat
• The system automatically enters follow-up mode (no need to click "Ask About This")
• Your question is sent to the Judge model with full debate context
• The AI can reference specific arguments, rounds, and the verdict

Example Questions

• "Why did Con win this debate?"
• "What was Pro's strongest argument?"
• "How could Pro have argued better?"
• "Summarize the key disagreements"

When to Use Debate

Controversial or divisive topics

Strategic decisions with trade-offs

Testing the strength of an argument

Exploring multiple perspectives systematically

When you want to challenge your assumptions

Due diligence on investments or acquisitions

API Usage

Konnect uses an OpenAI-compatible API with extensions for debate mode.

cURL

curl https://api.konnect.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "konnect-debate",
    "messages": [
      {"role": "user", "content": "Should AI development be regulated?"}
    ],
    "stream": true,
    "konnect.pattern": "debate",
    "konnect.models": ["gpt-4o", "gemini-2.0-flash", "claude-sonnet-4-5-20250929"],
    "konnect.personas": [
      {"personaId": "advocate", "modelId": "gpt-4o"},
      {"personaId": "critic", "modelId": "gemini-2.0-flash"},
      {"personaId": "impartial_judge", "modelId": "claude-sonnet-4-5-20250929"}
    ]
  }'

The three personas correspond to Pro (advocate), Con (critic), and Judge roles. The konnect.personas parameter is optional—defaults are used if omitted.

Streaming Events

Debate mode streams events progressively via Server-Sent Events (SSE):

setup

Opening positions from Pro and Con are ready. Includes pro_success and con_success flags.

round

A debate round completed. Contains round number, both arguments, and success/failure status for each side.

judge

Judge's verdict is ready. Contains winner and reasoning.

Example round event with graceful degradation

{
  "konnect.metadata": {
    "debate_event": "round",
    "debate_new_round": {
      "round_number": 1,
      "pro_argument": "...",
      "con_argument": "No argument provided (error: insufficient_quota)",
      "pro_success": true,
      "con_success": false
    }
  }
}

Explore other modes

Prediction Market for probability forecasting Ensemble mode for synthesized responses Council mode for expert panel analysis Full API reference