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.

---

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

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.

---

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.