LLM Cage Match // Multi-Agent Arena

🎛️ DEBATE SETUP

Chamber Mode ⓘ

Motion ⓘ

Try:

Supporting Context (Optional)

📄 Drag & drop .txt, .md, or .pdf — or browse

Audience Profile ⓘ

Speech Speed ⓘ

// PROPOSITION

Speaker 1

Speaker 2

// OPPOSITION

Speaker 1

Speaker 2

Moderator / Chair

📊 AUDIENCE OPINION POLL

OPINION LEADER ⓘ UNDECIDED

SWING SHIFT ⓘ 0.0%

PRO: 33%

UND: 34%

CON: 33%

Establish pre-debate vote. Ready to initiate opening arguments.

🗳️PRE-VOTE

🎤OPENINGS

⚡REBUTTALS

🙋QUESTIONS

📜CLOSINGS

🏆VERDICT

// ACTIVE RESOLUTION BEFORE THE HOUSE

"AGI will be achieved before 2030"

🤖

CHAMBER CLOSED

Awaiting motion selection...

OFFLINE

SYSTEM

Welcome to the LLM Cage Match. To start, configure the debate parameters in the setup panel and click "Start Debate".

You can choose simulated mode, or connect to your local Ollama server and cloud APIs using the API Configuration panel.

📜 DEBATE TRANSCRIPT

The debate has not yet begun. Once agents take the stand, their speeches will be recorded here.

❓ LLM CAGE MATCH GUIDE

Welcome to the LLM Cage Match, an arena where advanced language models compete in intellectual combat using multi-mode cognitive arena (debate, assembly, value clash, and war gaming).

📌 What is a "Motion"?

In this arena, a Motion is the central resolution or topic being debated (e.g. "AGI will be achieved before 2030"). The Proposition team argues in favor of the motion, while the Opposition team argues against it.

👥 The Debate Teams

- Proposition: Defends the Motion. (Usually ChatGPT + Grok).
- Opposition: Opposes the Motion. (Usually Gemini + Codex).
- Moderator / Chair: Directs the flow, reads the motion, interrogates speakers, and delivers the final audit results.

🗳️ Winning the Debate: The Swing Shift

Unlike traditional debates where the team with the highest final vote wins, the winner in an arena session is determined by the Swing Shift. The audience votes before the debate starts (Pre-Vote) and after it ends (Post-Vote). The team that successfully shifts the highest percentage of undecided voters to their side (or swings voters from the opposing team) is declared the victor.

🔄 The 14-Step Debate Cycle

The debate moves through five structured phases:

Pre-Vote: Establish baseline audience opinion.
Opening Speeches: Constructive arguments from each speaker.
Rebuttals: Cross-examination rebutting the opponent's logic.
Interrogation: The Moderator poses direct questions to challenge each side.
Closing Statements & Verdict: Final appeals followed by the post-debate poll swing audit.

⚙️ Operating Modes

- Parliamentary Debate: Classic adversarial parliamentary format. Proposition vs Opposition. Winner determined by swing shift of undecided voters.
- Engineering Assembly: Collaborative technical review. Architects defend, Auditors challenge. Focus is consensus building rather than winning.
- Lincoln-Douglas (1v1 Value Debate): Single Affirmative vs single Negative. Emphasizes value criteria, philosophical framing, and clash over the resolution's worth. Ideal for testing whether an idea has moral, strategic, or long-term merit.
- Red/Blue Team War Game (Viability): Red attacks the idea (risks, failure modes, hidden costs). Blue defends mitigations and upside. Output is a clear "viable to pursue" signal with quantified risk/reward framing, not a winner/loser verdict.
- Ollama Only: Runs fully locally by calling your Ollama endpoint model to generate new dynamic speeches for all roles.
- Hybrid Mode: Combines your cloud API keys (OpenAI, Gemini, Anthropic, Grok) with Ollama local fallbacks.

📊 SESSION TELEMETRY & ANALYTICS

TOTAL SESSION COST $0.000

TOKENS CONSUMED 0 In: 0 | Out: 0

AVG RESPONSE TIME 0.00s

SWING / PERSUASION N/A 0% shift / 1k tok

📝 DEBATE HIGHLIGHTS & LAYMAN SUMMARY

TOKEN USAGE BY MODEL

AVERAGE LATENCY COMPARISON (s)

TURN-BY-TURN LOGS

Round	Speaker	Model	In Tokens	Out Tokens	Latency	Cost	Swing Shift

🔑 API CONFIGURATION

🎭 CUSTOMIZE PERSONA SYSTEM PROMPTS