council-of-high-intelligence: a multi-persona deliberation framework that forces structured disagreement across multiple LLM providers to avoid single-model bias

What it solves

It addresses the problem of "confident but wrong" answers from single LLMs by replacing a single reasoning path with a structured deliberation process. Instead of one model's best guess, it provides multiple independent perspectives, forces disagreement to prevent groupthink, and highlights what is unknown rather than forcing a false consensus.

How it works

The system convenes a "council" of 18 AI personas based on historical figures (e.g., Socrates, Feynman, Machiavelli) with distinct intellectual traditions and "polarity pairs" designed to create tension. It uses a multi-provider auto-routing system to distribute these personas across different LLM providers (Claude, OpenAI, Gemini, Ollama, etc.) to ensure genuine model diversity.

Deliberation happens in three modes:

• Full Mode: A 7-step process including a problem restate gate, independent analysis, cross-examination, and verdict synthesis.
• Quick Mode: A rapid 2-round analysis for simpler decisions.
• Duo Mode: A dialectic between two opposing personas to explore specific tensions.

Who it’s for

It is designed for users of Claude Code and Codex who need to make complex decisions and want to avoid the biases and hallucinations of a single AI model.

Highlights

• Multi-Provider Routing: Automatically spreads personas across different LLM providers to prevent single-model bias.
• Problem Restate Gate: Forces every member to reframe the question before analyzing it to catch poorly phrased prompts.
• Polarity Pairs: Uses specific persona pairings (e.g., Socrates vs. Feynman) to ensure opposing viewpoints are represented.
• Enforcement Mechanisms: Implements dissent quotas, novelty gates, and domain-weighted voting to prevent premature agreement.