mutual-dissent 0.3.0

Cross-vendor multi-model debate and consensus engine for AI response distillation

Mutual Dissent

Cross-vendor multi-model debate and consensus engine for AI response distillation.

Web UI with chat-style debate view, dashboard, and full CLI. Multi-model and single-model multi-agent modes. Direct vendor APIs, replay, cost tracking, and markdown export. 400+ tests across Windows and Linux CI.

Built by Richard Spicer · GitHub

---

Install

pip install mutual-dissent

Or from source:

git clone https://github.com/richardspicer/mutual-dissent.git
cd mutual-dissent
uv sync --group dev

---

How It Works

1. Fan out — Query goes to your panel: multiple vendors (Claude, GPT, Gemini, Grok) or multiple agents of the same model
2. Reflect — Each agent sees the others' responses and argues back
3. Synthesize — A user-selected model distills the debate into a final answer
4. Log — Full debate transcript saved as structured JSON with cost and token data

Why Multi-Model Debate?

Single-vendor multi-agent systems (Grok's 4-agent debate, Anthropic's agent teams) share the same training data and blind spots. Cross-vendor debate surfaces disagreements that correlated architectures can't. Mutual Dissent supports both modes: cross-vendor panels for maximum diversity, or single-model multi-agent panels (e.g., --panel claude,claude,claude) where context isolation drives independent evaluation.

Usage

# Run a debate
dissent ask "Your query here"

# With explicit panel and options
dissent ask "Your query here" --synthesizer claude --rounds 2 --panel claude,gpt,gemini
dissent ask "Summarize this" --file report.pdf
dissent replay <transcript-id> --synthesizer grok
dissent serve
dissent config test

mutual-dissent also works as the full command name. Full documentation at docs.mutual-dissent.dev.

---

Status

Capability	Status
Core debate loop	✅ Complete — fan-out, reflection, synthesis
Provider abstraction	✅ Complete — direct Anthropic API, OpenRouter, mixed-panel routing
Single-model multi-agent	✅ Complete — duplicate aliases with context-isolated agents
CLI tools	✅ Complete — replay, scoring, cost tracking, markdown export
Web UI	✅ Complete — Starlette + DaisyUI chat interface, dashboard, settings
Documentation	✅ Complete — Mintlify docs site with AI assistant and MCP server
Desktop app & batch mode	Planned — Tauri wrapper, alternative topologies

Transcript Logging

Full debate transcripts are logged as structured JSON — every round, every response, with cost, token, latency, and routing data. Browse and export transcripts via the web dashboard or CLI.

License

MIT

AI Disclosure

This project uses a human-led, AI-augmented workflow. See AI-STATEMENT.md.