MAGI-PROTOCOL

A multi-agent AI operating system inspired by the MAGI supercomputer from Neon Genesis Evangelion.

Three AI agents - MELCHIOR (Logic), BALTHASAR (Emotion), and CASPER (Ethics) - debate every user query in real time. Watch them disagree, argue, and vote.

---

How It Works

Every query is sent to three AI subsystems simultaneously. Each agent analyses the question through a different lens and streams its response token by token in real time.Once all three have responded, the fourth ARBITER system then synthesises a final verdict.

The Three Subsystems

You can view the full prompts in /backend/prompts/

Agent	Designation	Reasoning Style
MELCHIOR-1	Logic	Probabilistic, data-driven, outcome focused
BALTHASAR-2	Emotion	Empathetic, intuitive, human centred
CASPER-3	Ethics	Principled, moral, precedent aware

Verdict States

• APPROVE - majority vote in favour
• OPPOSE - majority vote against
• DEADLOCK - no majority reached, further deliberation required

What It Works Best For

MAGI-PROTOCOL is designed for questions where reasonable perspectives can conflict:

• Decisions - "Should I quit my job?"
• Technical tradeoffs - "Should I use PostgreSQL or MongoDB?"
• Ethical dilemmas - "Is it acceptable to surveil employees?"
• Risk assessment - "Should I go bungee jumping?"

---

Architecture

Layer	Technology
Frontend	Next.js
Backend	FastAPI
Real-time	Websockets
Message Broker	Redis
LLM Runtime	Ollama or any LLM API
Deployment	Docker Compose

System Flow

User Query
    ↓
WebSocket /ws/debate
    ↓
Stream MELCHIOR-1 → BALTHASAR-2 → CASPER-3
    ↓
Parse confidence + stance from each response
    ↓
ARBITER synthesises final verdict
    ↓
Redis stores session history
    ↓
Frontend renders live via WebSocket

---

Configuration

Provider	LLM_PROVIDER	API_BASE_URL
Ollama (local)	ollama	—
OpenAI	api	https://api.openai.com/v1
Anthropic	api	https://api.anthropic.com/v1
Mistral	api	https://api.mistral.ai/v1
Groq	api	https://api.groq.com/openai/v1

No API key needed for Ollama. For cloud providers, set API_KEY in your .env.

Use .env.example as a master reference for other .env files.

---

Run Local

Prerequisites

• Python 3.14+
• Node.js 26+
• Redis

Clone the repo

git clone git@github.com:CosmoJelly/MAGI-PROTOCOL.git
cd MAGI-PROTOCOL

Pull a model (if using Ollama)

ollama pull llama3:latest

Configure environment

cp .env.example .env

Run with Docker

docker compose up

Then open http://localhost:3000.

---

Project Structure

MAGI-PROTOCOL/
├── backend/
│   ├── prompts/
│   │   ├── melchior.md      
│   │   ├── balthasar.md     
│   │   ├── casper.md        
│   │   └── arbiter.md       
│   ├── main.py              
│   ├── agents.py            
│   ├── llm.py               
│   ├── arbiter.py          
│   ├── parser.py            
│   ├── redis_client.py  
│   └── requirements.txt
├── frontend/
│   └── app/
│       ├── components/
│       │   ├── MagiTriangle.tsx    
│       │   ├── ArbiterVerdict.tsx   
│       │   └── QueryTerminal.tsx    
│       ├── page.tsx                
│       ├── layout.tsx             
│       └── globals.css 
├── docker-compose.yml
├── .env.example
└── README.md

---

Copyright & Legal

MAGI-PROTOCOL is an independent fan project and is not affiliated with, endorsed by, or connected to Gainax, Khara, or the Neon Genesis Evangelion franchise in any way.

Neon Genesis Evangelion and all related names, characters, and imagery are the intellectual property of Gainax and Khara. No copyright infringement is intended.

All original code in this repository is released under the MIT License. See LICENSE for details.