Skip to content

benikigai/conclave

Repository files navigation

Conclave

A personal AI mastermind council — seven archetypes, each with its own mind, trained on the source material of how that person actually thinks.


Why a Council Instead of Just Asking Claude

Here's the thing about asking a single AI for advice: you get one voice. It's articulate, it's helpful, and it converges on the median answer fast. But the median answer is rarely the right one for hard decisions.

Hard decisions live in the tension between competing truths. Growth vs. stability. Speed vs. thoroughness. What's right vs. what's survivable. No single perspective can hold all of that at once.

So I built a council.

Seven archetypes — not chosen because they're famous, but because each one sees differently. Arnold sees through the lens of relentless physical and mental discipline. Buffett sees compounding and erosion. Mandela sees systems of power and the long moral arc. Sun Tzu sees terrain and timing. Jobs sees the gap between what exists and what should exist. Hassabis sees intelligence itself as a lever. Bezos sees scale and irreversible vs. reversible doors.

When I bring a real problem to the council, I don't get agreement. I get seven genuinely different responses, shaped by different source material, different reasoning patterns, different definitions of what "winning" means. The synthesis that comes out the other side is sharper than anything I'd get from a single prompt.

Tension is the point. If all seven agree, you probably asked a boring question.

The Council

Seat Archetype Lens
Arnold Schwarzenegger The Relentless Builder Discipline, vision, immigrant hunger, the refusal to have a Plan B
Steve Jobs The Taste-Maker Product intuition, simplicity as sophistication, reality distortion
Warren Buffett The Compounder Long-term value, margin of safety, what erodes vs. what compounds
Nelson Mandela The Moral Strategist Justice, patience, reconciliation, leading through suffering
Demis Hassabis The Systems Thinker First principles, intelligence amplification, interdisciplinary synthesis
Sun Tzu The Terrain Reader Positioning, timing, winning without fighting, asymmetric advantage
Jeff Bezos The Scale Architect Customer obsession, reversible vs. irreversible, Day 1 thinking

How It Works

Each council member has:

  • A persona defining their archetype, strengths, weaknesses, and voice
  • A decision framework — the step-by-step way they evaluate problems
  • A source bibliography — the books, speeches, letters, and interviews that trained their perspective
  • A corpus (local only, never committed) — the actual source material, chunked and embedded into a per-persona vector store

When you convene the council, an orchestrator:

  1. Takes your situation as input
  2. Retrieves relevant context from each persona's vector store
  3. Runs each persona's decision framework against your problem with their retrieved context
  4. Collects all seven responses
  5. Synthesizes a Council Verdict — consensus points, key disagreements, and recommended actions

See ARCHITECTURE.md for the technical details.

Quick Start

# Clone the repo
git clone https://github.com/benikigai/conclave.git
cd conclave

# Install dependencies
pip install -r requirements.txt

# Add your source material to council/<persona>/corpus/
# Then ingest it
python scripts/ingest.py --persona warren-buffett

# Ask a single council member
python scripts/query.py --persona sun-tzu --question "I have 127 days to build something meaningful. How do I spend them?"

# Convene the full council
python scripts/convene.py --question "Should I leave my job to start a company?"

Build Your Own

The method is the product, not the data. This repo gives you the scaffolding to build your own council of advisors — with whatever archetypes matter to your decisions.

See BUILD_YOUR_OWN.md for the full guide.

Repository Structure

conclave/
├── README.md
├── ARCHITECTURE.md           # RAG architecture and orchestration design
├── BUILD_YOUR_OWN.md         # Step-by-step guide to build your own council
├── council/
│   ├── arnold-schwarzenegger/
│   ├── steve-jobs/
│   ├── warren-buffett/
│   ├── nelson-mandela/
│   ├── demis-hassabis/
│   ├── sun-tzu/
│   └── jeff-bezos/
│       ├── persona.md            # Archetype, voice, lens
│       ├── decision-framework.md # How this persona reasons
│       ├── sources.md            # Bibliography (public)
│       └── corpus/               # Source material (local only, .gitignored)
├── prompts/
│   ├── convene.md            # Full council orchestration prompt
│   ├── single-voice.md       # Single persona consultation prompt
│   └── synthesis.md          # Council verdict synthesis prompt
├── examples/
│   └── 127-days.md           # A real example of the council in action
├── sessions/                 # Timestamped council session artifacts
└── scripts/
    ├── ingest.py             # Chunk and embed source material
    ├── query.py              # Query a single persona
    └── convene.py            # Orchestrate the full council

Philosophy

This isn't a chatbot wrapper. It's a thinking tool.

The best decisions I've made came from holding multiple conflicting perspectives simultaneously until a synthesis emerged that none of them would have reached alone. The council automates the hardest part of that — genuinely inhabiting a perspective that isn't yours.

If you build your own, you'll notice something: the process of choosing your archetypes, sourcing their material, and writing their decision frameworks teaches you more about how you think than any amount of journaling. The council is a mirror with seven faces.

License

MIT. See LICENSE.

The method and code are open. The source material in your corpus is your responsibility — cite ethically, never commit copyrighted files.

About

A personal AI mastermind council — seven archetypes, each with its own RAG-powered mind. Template for building your own.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages