Council of High Intelligence tackles the challenge of getting multiple AI personas to deliberate genuinely on complex questions. Instead of just roleplaying different characters on the same model, it enforces structured disagreement and routes agents across multiple LLM providers to ensure diverse reasoning paths. This repo is a Shell-based orchestration tool that runs 18 AI personas modeled after historical and modern thinkers to debate decisions through multi-round protocols.
What council of high intelligence orchestrates and how it works
Under the hood, this repo is a Shell script toolset that integrates with Claude Code and Codex CLI environments via slash commands. It installs 18 distinct council agents—thinkers with distinct personas—and skill files that let you run deliberations with three modes: full, quick, and duo. Each mode corresponds to a different interaction pattern:
- Full mode runs three rounds including cross-examination to deepen reasoning.
- Quick mode is a two-round rapid deliberation.
- Duo mode pairs two members with opposing views (polarity pairs) for dialectic exchange.
A critical architectural choice is the multi-provider routing system. Instead of running all personas on the same LLM with different prompts, the tool distributes agents across providers like Claude, OpenAI (Codex), Gemini, and Ollama. This approach surfaces genuinely diverse reasoning rather than prompt-level variations on a single model.
The system also enforces disagreement through dissent quotas, novelty gates, and counterfactual prompts, mechanisms designed to avoid groupthink and scripted consensus. This means the personas are not just echo chambers but are pushed to challenge each other authentically.
The repo comes with 20 pre-defined triads tailored for different domains or problem types, making it easier to deploy domain-specific deliberations.
Why the structured disagreement model stands out
Most multi-agent AI frameworks rely on prompt engineering to simulate different viewpoints on the same underlying model. Council of High Intelligence takes a more robust approach by pushing for actual disagreement with explicit dissent enforcement. The use of polarity pairs—agents assigned opposite stances—creates a dialectic tension that leads to richer debate.
The multi-round design ensures that deliberation is not a single-shot output but an evolving conversation. This allows agents to critique and respond, mimicking human debate more closely.
From a code quality standpoint, the repo is surprisingly clean for a Shell-based orchestration tool. The install script handles multiple configuration paths and supports optional multi-provider skill installs. It’s opinionated but practical, favoring integration with existing CLI tools rather than building a heavyweight framework.
The tradeoff is that this approach relies heavily on the availability and latency of multiple LLM providers, which can introduce complexity and potential delays. The shell scripting environment is lightweight but may limit extensibility or advanced error handling compared to a full programming language.
Still, the design decisions emphasize genuine multi-model reasoning diversity and structured disagreement over simpler multi-agent prompt orchestration.
Quickstart for Claude Code and Codex CLI
To get started with Council of High Intelligence in Claude Code, clone the repo and run the install script:
git clone https://github.com/0xNyk/council-of-high-intelligence.git
cd council-of-high-intelligence
./install.sh
After installation, within Claude Code, you can invoke the council with slash commands:
/council Should we open-source our agent framework?
/council --quick Should we add caching here?
/council --duo Should we use microservices or monolith?
For Codex CLI, the installation adds Codex skill support as well:
git clone https://github.com/0xNyk/council-of-high-intelligence.git
cd council-of-high-intelligence
./install.sh --codex
Then use the same slash commands in Codex.
The install script supports various options to customize your setup, including installing only Codex skills, specifying config directories, or running a dry run to preview setup without writing files. After installation, restarting your client and running the provided checklist script helps validate your environment.
verdict: a specialized tool for multi-LLM deliberation with enforced disagreement
Council of High Intelligence is suited for AI researchers, developers, or enthusiasts looking to explore multi-agent reasoning beyond prompt-level roleplay. Its strength lies in enforcing genuine disagreement and achieving model diversity by distributing personas across multiple LLM providers.
The shell-based orchestration favors CLI users already invested in Claude Code or Codex environments, but the tradeoff is less extensibility compared to more programmatic multi-agent frameworks. Latency and provider availability are practical considerations when running multi-round deliberations.
It’s not a general-purpose chatbot or a plug-and-play ensemble but a thoughtful experiment in structured AI debate worth understanding for anyone interested in multi-LLM collaboration patterns or AI decision support systems.
Related Articles
- OpenAI Codex CLI: local-first AI coding assistant with ChatGPT integration — OpenAI Codex CLI brings AI coding assistance local to your terminal, integrating with ChatGPT plans for powerful hybrid
- LobeHub: An extensible AI agent playground with MCP plugin architecture — LobeHub offers a TypeScript-based AI agent platform with a unique MCP plugin system for integrating 10,000+ skills and c
→ GitHub Repo: 0xNyk/council-of-high-intelligence ⭐ 612 · Shell