I built a CLI tool to orchestrate structured debates between LLMs. The goal is to reduce hallucinations and bias by forcing models to critique each other before reaching a consensus.
It supports a hybrid setup: you can have your local Ollama models debate against GPT-5.2 or Claude 4.5.
Key Features: - Local-first: Auto-discovers Ollama models (Llama 3.3, Mistral, etc) with zero config. - Method Advisor: Press Tab to analyze your prompt. A meta-agent will recommend the best debate method (e.g. "Oxford Debate" for binary choices or "Delphi" for forecasting). - 7 Built-in Methods: Includes Devil's Advocate, Brainstorming, and Socratic dialogue.
It uses a React-based TUI for a modern terminal experience.
License: BSL 1.1 (free for personal/internal use).