nan-council — Multi-AI debate system

Why use multiple AI models?

Every AI model has blind spots. GPT-4o might confidently recommend a library that Claude knows has a critical security vulnerability. Llama might suggest an approach that Gemini identifies as deprecated. When you rely on a single model for technical decisions, you inherit that model's biases and knowledge gaps without any cross-check.

nan-council eliminates single-model risk by querying multiple AI models in parallel and analyzing their agreement. When all models converge on the same answer, you can proceed with high confidence. When they diverge, you know exactly where the uncertainty lies and can investigate further. This is especially valuable for architecture decisions, library selection, and security-sensitive code review.

How does nan-council work?

nan-council supports three modes. Ask sends your question to all configured models simultaneously and returns their responses with an agreement analysis. Debateruns multiple rounds where each model sees the others' responses and can rebut — this naturally converges toward the strongest arguments. Vote presents predefined options and collects a structured ballot with confidence scores from each model. Auto-detect finds AI tools already installed on your machine (Ollama, Claude CLI, Codex CLI, Copilot) so setup takes seconds.