Council of High Intelligence: How 18 AI Personas Deliberate Your Hardest Decisions
The Council of High Intelligence assembles 18 AI personas—Aristotle, Feynman, Torvalds, Kahneman—across multiple LLM providers to give you structured disagreement instead of confident guesses.
The Council of High Intelligence is an open-source tool that assembles 18 AI personas—Aristotle, Feynman, Torvalds, Kahneman, and more—across multiple LLM providers to deliberate your hardest decisions. Instead of one confident answer, you get structured disagreement, genuine debate, and verdicts that surface what the council doesn't know.
The Problem With Single-Model Advice
Ask any LLM a hard question and you'll get a fluent, structured, confident answer. The problem? It's just one reasoning path dressed up as certainty. You're getting a single advisor's best guess, not a board of experts debating the tradeoffs.
What if you could convene a council instead? Aristotle categorizes your problem. Socrates destroys your assumptions. Feynman rebuilds from first principles. Linus Torvalds tells you to ship it. And they all argue with each other—using genuinely different models, not costume changes on one.
That's exactly what Council of High Intelligence does. It's trending on GitHub with over 2,000 stars and 331 new stars daily, and it's changing how developers approach hard decisions.
How the Council Works
18 Personas, Multiple LLM Providers
The council isn't just prompting tricks. It routes different members across Claude, OpenAI, Gemini, and Ollama to ensure you get actually different reasoning patterns:
- Aristotle (Claude Opus) — Categorization & structure
- Socrates (Claude Opus) — Assumption destruction
- Feynman (Claude Sonnet) — First-principles debugging
- Linus Torvalds (Claude Sonnet) — Pragmatic engineering
- Daniel Kahneman (Claude Opus) — Cognitive bias & decision science
- Charlie Munger (Claude Sonnet) — Multi-model reasoning
- Nassim Taleb (Claude Opus) — Antifragility & tail risk
- Andrej Karpathy (Claude Sonnet) — Neural network intuition
- Ilya Sutskever (Claude Opus) — Scaling frontier & AI safety
And 9 more, each chosen as a deliberate counterweight to others.
Polarity Pairs: Engineered Disagreement
Members are paired as intellectual opposites:
Aristotle vs Lao Tzu — Classifies everything vs structure IS the problem
Torvalds vs Watts — Ships concrete solutions vs questions whether the problem exists
Karpathy vs Sutskever — Build it, observe it, iterate vs pause, research, ensure safety first
This isn't random. If >70% agree too early, two members are forced to steelman the opposing view. Dissent quotas and novelty gates prevent groupthink.
The Problem Restate Gate
Before analysis begins, every member must reframe your question. If three members restate it differently, you know the question itself was the problem. This catches wrong questions early—before you waste deliberation on the wrong thing.
Three Deliberation Modes
Full Mode (Default)
3-round structured deliberation: independent analysis → cross-examination → final positions. Use for high-stakes decisions.
/council Should we open-source our agent framework?
/council --full What is the right pricing model?Quick Mode
2-round rapid analysis. No cross-examination. Use for simpler decisions.
/council --quick Should we add caching here?
/council --quick --triad shipping Should we release today?Duo Mode
2-member dialectic using polarity pairs. Great for exploring specific tensions.
/council --duo Should we use microservices or monolith?
/council --duo --members torvalds,ada Is this abstraction worth it?Pre-Defined Triads: 20 Domain-Specific Combinations
Don't want all 18 members? Use a triad—three experts optimized for specific domains:
| Domain | Triad | Rationale |
|---|---|---|
| architecture | Aristotle + Ada + Feynman | Classify + formalize + simplicity-test |
| debugging | Feynman + Socrates + Ada | Bottom-up + assumption testing + formal verification |
| ai-safety | Sutskever + Aurelius + Socrates | Safety frontier + moral clarity + assumption destruction |
| shipping | Torvalds + Musashi + Feynman | Pragmatism + timing + first-principles |
| decision | Kahneman + Munger + Aurelius | Bias detection + inversion + moral clarity |
| product | Torvalds + Machiavelli + Watts | Ship it + incentives + reframing |
| risk | Sun Tzu + Aurelius + Feynman | Threats + resilience + empirical verification |
Use them like this:
/council --triad strategy What's our competitive moat?
/council --triad ai Should we build or fine-tune?Real-World Example: Architecture Decision
Let's say you're deciding between microservices and a monolith for a new project:
/council --duo Should we use microservices or monolith?What happens:
- Problem Restate Gate: Both members reframe your question. Torvalds asks "What's the team size and shipping timeline?" Ada asks "What are the formal boundaries of your domains?"
- Round 1 (Independent Analysis): Torvalds argues monolith (simplicity, deploy speed). Ada argues microservices (domain isolation, independent scaling).
- Round 2 (Cross-Examination): They challenge each other. Torvalds: "Your microservices add network latency and operational complexity." Ada: "Your monolith creates coupling that slows feature development."
- Final Verdict: Structured synthesis with unresolved questions highlighted.
The verdict doesn't just say "use microservices." It says: "If team <5 people and shipping in 3 months, monolith. If you have clear domain boundaries and 10+ engineers, microservices. Unresolved: How will you handle distributed transactions?"
That unresolved question is more valuable than any definitive answer. It tells you exactly what you need to research before committing.
Key Benefits
- Genuinely different perspectives — Multi-provider routing ensures different reasoning, not just different prompts
- Catch wrong questions early — Problem Restate Gate surfaces when the question itself is flawed
- Know what's unknown — Verdicts lead with unresolved questions, not confident-sounding consensus
- Prevent groupthink — Dissent quotas force genuine disagreement when agreement is too easy
- Works in Claude Code and Codex — Install once, use as a slash command
- Open source (CC0) — Fully open, no restrictions
Installation
For Claude Code:
git clone https://github.com/0xNyk/council-of-high-intelligence.git
cd council-of-high-intelligence
./install.shFor Codex:
git clone https://github.com/0xNyk/council-of-high-intelligence.git
cd council-of-high-intelligence
./install.sh --codexThen use /council in your Claude Code or Codex session.
FAQ
Why not just ask Claude directly?
A single prompt gives you one model's confident best guess. The council gives you 3-18 independent analyses from different intellectual traditions, forces them to challenge each other, and synthesizes a verdict that surfaces disagreement rather than hiding it. It's the difference between asking one advisor and convening a board.
What LLM providers does it support?
The council routes members across Claude (Opus and Sonnet), OpenAI, Gemini, and Ollama. Multi-provider routing ensures you get actually different reasoning patterns, not just different prompts on the same model.
How long does a deliberation take?
Quick mode: 2-3 minutes. Full mode: 5-10 minutes depending on complexity. The council runs members in parallel where possible, so it's faster than you'd expect for 18 personas.
Can I use it for coding decisions?
Absolutely. Use the --triad architecture or --triad debugging modes. Linus Torvalds provides pragmatic engineering advice, Ada Lovelace focuses on formal systems, and Feynman rebuilds from first principles.
What if I want specific members instead of all 18?
Use the --members flag: /council --members torvalds,kahneman,taleb. Or use a pre-defined triad for domain-specific combinations.
Is it expensive to run?
A full 18-member deliberation uses roughly 50k-100k tokens across all providers. Quick mode and triads are cheaper. You control which providers to use in your config.