Skip to main content
Mutual Dissent CLI demo Mutual Dissent sends a query to multiple AI models simultaneously, shares competing responses back to each model for reflection and critique, then synthesizes a final answer through a user-selected model. Works with cross-vendor panels (Claude, GPT, Gemini, Grok) or single-model multi-agent panels where context isolation drives independent evaluation.

​
Why Multi-Model Debate?

Single-vendor multi-agent systems share training data, safety postures, and blind spots. Cross-vendor debate surfaces disagreements that correlated architectures can’t. Mutual Dissent supports both approaches: cross-vendor panels for maximum diversity, or single-model panels (e.g., --panel claude,claude,claude) where each agent evaluates independently through context isolation.

​
The Debate Loop

  1. Fan out — the query goes to all panel agents in parallel
  2. Reflect — each agent sees the others’ responses and argues back (1–3 rounds)
  3. Synthesize — a designated model distills the full debate into a final answer
  4. Log — the complete transcript is saved as structured JSON with cost and token data

​
Transcript Logging

Every debate produces a structured JSON transcript with cost, token, latency, and routing data. Browse and export transcripts via the web dashboard or CLI. The --ground-truth flag scores synthesis against a reference answer using a judge model.

​
Get Started

Quickstart

Install and run your first debate in under 5 minutes

Core Concepts

Panels, rounds, synthesis, and transcripts explained

CLI Reference

Full command and flag reference

Configuration

API keys, routing, and defaults