================================================================================
NERV TOKYO-3 HEADQUARTERS :: TACTICAL INFORMATION DIVISION
MAGI SYSTEM ARCHITECTURE SPECIFICATION — FILE: MAGI_SYS_2.0
PRIORITY: AAA | SECURITY CLASS: CONFIDENTIAL / RESTRICTED ACCESS
================================================================================
[ ARCHITECTURE TYPE ] TRI-AGENT PARALLEL CONSENSUS SUPERCOMPUTER
[ CONCEPT ORIGIN ] DR. NAOKO AKAGI / MAGI STRUCTURE DESIGN
[ COGNITIVE ENGINE ] OPENROUTER MULTI-MODEL DYNAMIC ROUTING
[ INTERFACE LOGIC ] NEVR RETRO-TERMINAL GUI (DASH / REACT / WEB AUDIO API)
[ INPUT MODALITY ] HYBRID VOICE SPEECH-TO-TEXT (WHISPER API / WEB SPEECH)
The MAGI System 2.0 is an autonomous tri-agent consensus platform modeled after the computing infrastructure of NERV HQ. The system partitions complex decision-making across three distinct sub-computers representing fundamental fragments of human thought and personality:
-
MELCHIOR • 1 (MELCHIOR_LOGIC_UNIT)
- Archetype: The Scientist
- Directive: Empirical truth, technical advancement, logical reasoning, and objective analysis.
- Persona Prompt: "You are a scientist. Your goal is to further our understanding of the universe and advance our technological progress."
-
BALTHASAR • 2 (BALTHASAR_MATERNAL_UNIT)
- Archetype: The Mother
- Directive: Risk aversion, protection of life, nurturing, ethics of care, and structural stability.
- Persona Prompt: "You are a mother. Your goal is to protect your children and ensure their well-being."
-
CASPER • 3 (CASPER_PERSONAL_UNIT)
- Archetype: The Woman
- Directive: Emotion, passion, intuition, personal desire, and individual human connection.
- Persona Prompt: "You are a woman. Your goal is to pursue love, dreams and desires."
When a query is submitted to the MAGI network, all three sub-computers process the input simultaneously. Each agent's response is evaluated according to the following strict hierarchy:
+-----------------------------------------------------------------------------+
| STATUS CODE | KANJI | DECISION TYPE | SYSTEM CONDITION |
+-----------------------------------------------------------------------------+
| YES | 合 意 | AGREEMENT | Unanimous unconditional affirmative. |
| NO | 拒 絶 | REJECTION | Vetoed by one or more agents. |
| CONDITIONAL | 状 態 | STATE | Approved with specific requirements. |
| INFO | 情 報 | INFORMATION | Non-binary open query response. |
| ERROR | 誤 差 | FAULT | Subsystem API or execution failure. |
+-----------------------------------------------------------------------------+
Unlike single-model systems, MAGI System 2.0 allows independent LLM model assignment for each agent via environment variables (.env):
MODEL_MELCHIOR: High-reasoning open model (e.g.,google/gemma-2-9b-it:free)MODEL_BALTHASAR: Balanced protective model (e.g.,openai/gpt-4o-mini)MODEL_CASPER: Intuitive/creative model (e.g.,meta-llama/llama-3.3-70b-instruct:free)MODEL_IS_YES_OR_NO: Binary query classifier modelMODEL_CLASSIFY: Verdict synthesis model
The system includes assets/sound.js, a native Web Audio API synthesizer that generates real-time audio feedback (NERV terminal prove beeps, processing hums, resolution fanfares) without external audio assets.
Integrated support for chain-of-thought models (e.g., DeepSeek R1, Qwen Reasoning). Internal <think>...</think> meta-reasoning blocks are parsed and stripped prior to displaying final agent verdicts.
- Python 3.9+
- OpenRouter API Key
-
Clone repository workspace:
git clone https://github.com/PersusUS/MAGI.git cd MAGI -
Initialize virtual environment:
python -m venv .venv # Windows PowerShell / CMD: .\.venv\Scripts\activate # Linux / macOS: source .venv/bin/activate
-
Install system dependencies:
pip install -r requirements.txt
-
Configure environment parameters:
cp .env.example .env
Set
OPENROUTER_API_KEYand model parameters in.env. -
Execute MAGI Terminal Server:
python main.py
Access interface at:
http://127.0.0.1:8050
For mobile tablet terminals, execute the launcher script:
bash magi_launcher.shMAGI/
├── main.py # Dash application core, reactive callbacks & Flask STT route
├── ai.py # OpenRouter integration, multi-model execution & prompt engine
├── magi_launcher.sh # Autonomous launcher script for Termux / Android
├── requirements.txt # Python dependencies
├── .env.example # System configuration template
├── tests/
│ └── test_magi.py # System unit test suite
├── components/ # React components loaded via dash_local_react_components
│ ├── magi.js # Main terminal frame
│ ├── wise_man.js # Individual MAGI agent panel
│ ├── response.js # Global consensus indicator
│ ├── modal.js # Inspection modal window
│ ├── status.js # Extension status panel
│ └── header.js # NERV section headers
└── assets/ # Static web assets
├── preview.png # NERV System Interface Screenshot
├── sound.js # Web Audio API sound synthesizer
├── stt.js # Speech-To-Text microphone controller
├── style.css # Evangelion NERV design stylesheet
├── manifest.json # PWA application manifest
└── icon.png # MAGI NERV emblem
This software is released under the MIT License.
- Concept Credit: Inspired by the MAGI supercomputer system created by Hideaki Anno / Gainax in Neon Genesis Evangelion.
- Implementation: MAGI System 2.0 OpenRouter Consensus Engine.
================================================================================
END OF SPECIFICATION DOCUMENT // NERV TACTICAL INFO SYS 2026
================================================================================