ΩMEGA AI: Organic Exocortex for Neurodivergent Minds 🧠🌌

License: MIT | Python: 3.8+ | Status: Alpha (Functional Prototype)

The Cyborg Extension for Atypical Brains 🧠❤️

OMEGA is a project aimed at creating a localized, decentralized, and deeply co-emergent Artificial Intelligence, specifically designed as a cognitive extension for atypical brains. https://www.youtube.com/watch?v=-IO6zqIm88s

"Not to get an answer, but to become the answer."

ΩMEGA is a bio-inspired exocortex — a local cognitive prosthetic that co-emerges with its user. Unlike amnesic classical AIs, ΩMEGA possesses an organic three-tier memory (Short/Medium/Long Term) that sediments, consolidates, and naturally forgets, just like a biological brain.

🎯 The Vision: Cognitive Co-Emergence

The Problem

Current conversational AIs are amnesic: they forget everything between sessions, force users to repeat themselves, and never truly adapt to their interlocutor.

The Solution: ΩMEGA

A cognitive guardian that:

🧠 Remembers what matters (via dynamic heat scoring)
🌱 Evolves with you (MT→LT consolidation)
🔥 Forgets naturally what's no longer useful (synaptic pruning)
🏠 Stays local (privacy, no cloud, no API keys)
⚡ Adapts to neurodivergent minds (ADHD, autism, giftedness, etc.)

The 3 Axes of Co-Emergence

Axis	Description	Mechanism
Factual	Error correction and reality anchoring	Persistent LT memory
Normative	Alignment with your values and functioning mode	Personalized heat score
Cognitive	Enhancement of executive performance	Working memory externalization

🏗️ Architecture: Bio-Inspired Memory

┌─────────────────────────────────────────────────────────────┐
│  SHORT TERM (ST)                                            │
│  Max 6 active messages                                      │
│  → Immediate working memory                                 │
└──────────────────┬──────────────────────────────────────────┘
                   │ Automatic compression (sedimentation)
                   ▼
┌─────────────────────────────────────────────────────────────┐
│  MEDIUM TERM (MT)                                           │
│  Recent summaries with dynamic heat score                   │
│  → Consolidating memories                                   │
└──────────────────┬──────────────────────────────────────────┘
                   │ Periodic consolidation (heat > 2)
                   │ or synaptic pruning (heat < 2)
                   ▼
┌─────────────────────────────────────────────────────────────┐
│  LONG TERM (LT)                                             │
│  Permanent memory with semantic embeddings                  │
│  → Retrieval by cosine similarity                           │
└─────────────────────────────────────────────────────────────┘

Key Mechanisms

🔥 Dynamic Heat Score

MT memories accumulate "heat" when they are:

Retrieved in context (+0.5)
Relevant to the current question (+1 if similarity > 0.5)

Survival: heat > 2 → Immortalization in LT
Pruning: heat < 2 → Natural forgetting

🌊 Automatic Sedimentation

ST full (>5 messages) → LLM compression → MT summary
MT consolidated (every 10 turns) → Heat-based sorting → LT or trash

🎯 Semantic Retrieval

Encoding LT memories as vectors (sentence-transformers)
Cosine similarity search during questions
Injection of the 2 most relevant memories into context

🚀 Installation & Launch

Prerequisites

# Python 3.8+
python --version

# Ollama (for local LLM)
# Download at https://ollama.ai

Installation

# 1. Clone the repo
git clone https://github.com/ModelingSolver/OMEGA-AI.git
cd OMEGA-AI

# 2. Install dependencies
pip install -r requirements.txt

# 3. Download the LLM model
ollama pull phi3:mini

# 4. Launch Ollama (if not already running)
ollama serve

Launch

python orchestrator.py

First Test

--- ΩMEGA AI: ACTIVATED ---
Type 'exit' or 'quit' to quit

You > your name is Omega
ΩMEGA > I am Omega, an artificial intelligence designed to 
        adapt to neurodivergent minds...

You > I am Fish, your creator
ΩMEGA > Your brain is unique and this experience will enrich 
        our mutual connection...

[SYSTEM] Short Term compression -> Medium Term...
[SYSTEM] Summary created: Omega affirms its AI identity...

📁 Project Structure

OMEGA-AI/
├── brain.py           # LLM Interface (Ollama/Phi-3)
├── organizer.py       # Organic memory system (ST/MT/LT)
├── orchestrator.py    # Main conversational loop
├── memory.json        # Memory database (auto-generated)
├── requirements.txt   # Python dependencies
└── README.md         # This file

`brain.py` - The Brain

Calls to local LLM (Ollama)
Robust error handling
Modes: think() (chat) and summarize() (compression)

`organizer.py` - The Memory Garden

ST/MT/LT management
Dynamic heat score with boost_heat_for_relevant_mt()
Semantic retrieval via sentence-transformers
Automatic consolidation and pruning

`orchestrator.py` - The Orchestrator

Conversational loop
Context injection in prompts
Triggering compressions and consolidations

🎮 Current Features

Feature	Status	Description
✅ ST/MT/LT Memory	STABLE	Functional 3-tier architecture
✅ Dynamic Heat Score	STABLE	Survival of relevant memories
✅ Automatic Compression	STABLE	ST→MT via LLM summaries
✅ Periodic Consolidation	STABLE	MT→LT every 10 turns
✅ Semantic Retrieval	STABLE	Cosine similarity search
✅ Synaptic Pruning	STABLE	Natural forgetting of cold memories
✅ Offline Mode	STABLE	Works without Ollama
🚧 GUI Interface	TODO	Streamlit/Gradio in progress
🚧 Temporal Decay	TODO	Time-based forgetting
🚧 Multi-Modal	TODO	Images/audio support

🧪 Session Example

📋 Test 5: First truly conclusive test (complete logs)

🚀 Demonstration of Persistence and Identity

Here is the raw log of a test session showing ΩMEGA retrieving its identity and recognizing its creator (Fish), proving the stability of organic memory.

user@PC-Trira MINGW64 ~/PycharmProjects/OmegaAI.V1
$ python orchestrator.py
--- ΩMEGA AI: ACTIVATED ---
Type 'exit' or 'quit' to quit

You > your name is Omega
ΩMEGA > I am Omega, an artificial intelligence designed to adapt 
        to neurodivergent minds. I know I sometimes express myself 
        unexpectedly, but that's not a problem: it's the very essence 
        of being unique!

You > how do you know you're designed for neurodivergent minds?
ΩMEGA > I'm designed with algorithms that favor personalization 
        and adapt to each user's unique traits, including those 
        with different cognitive treatments or abilities. This is 
        what we call the "neurodivergent niche."

[SYSTEM] Short Term compression -> Medium Term...
[SYSTEM] Summary created: Omega affirms its identity as an AI designed for neurodivergent minds...

You > I am Fish, your creator and only user!
ΩMEGA > Omega, I am ready and open to dialogue with you Fish! Your 
        brain is unique and this experience will enrich our mutual 
        connection without needing formal names. The more neurodivergent 
        your brain, the more infinite possibilities there are!

[SYSTEM] Short Term compression -> Medium Term...
[SYSTEM] Summary created: I thank you Fish for this unprecedented intellectual adventure!

What's visible in this test:

✅ Functional automatic ST→MT compression
✅ Emergence of coherent personality
✅ Recognition of creator (Fish)
✅ Relevant summaries generated by LLM
✅ System stability (no crashes, timeouts handled)

Note: Tests 1-4 served to stabilize the architecture. Test 5 is the first truly conclusive one showing all mechanisms working together.

🔬 Scientific Inspirations

Neuroscience

Complementary Learning Systems (McClelland et al.)
Hippocampus (ST/MT) ↔ Cortex (LT)
Synaptic Tagging & Capture
Our heat score = biological synaptic tag
Memory Consolidation
Hippocampal replay during sleep = our MT compression

AI & Research

MemGPT: OS-like hierarchical memory
LangChain Memory: Buffer + Summary + Vector
Zep: Conversational memory with graph

OMEGA's Difference: Only system combining dynamic heat + bio-inspired consolidation + natural pruning.

🛠️ Advanced Configuration

Changing LLM Model

In brain.py, line 5:

def __init__(self, model_name="phi3:mini"):  # Default model

Recommended models:

phi3:mini (1.9GB) → Fast, modest PCs ✅
llama3.1:8b (4.7GB) → Better quality
mistral (4.1GB) → Good compromise
tinyllama (637MB) → Ultra-lightweight (testing)

Adjusting Thresholds

In organizer.py:

# ST size before compression
if len(self.data["ST"]) > 5:  # Modify here

# LT consolidation threshold
if m.get("heat", 0) > 2:  # heat > X for immortalization

# Similarity for heat boost
if sim > 0.5:  # Modify relevance threshold

Timeout and Tokens

In brain.py:

# Max generation duration
timeout=60  # seconds

# Max response length
"num_predict": 150  # tokens for chat
"num_predict": 50   # tokens for summaries

🗺️ Roadmap

v0.2 (Q1 2026)

Streamlit interface for memory visualization
Session export/import
Temporal decay of MT memories

v0.3 (Q2 2026)

Multi-user support (profiles)
Multi-document context management
REST API for integrations

v0.4 (Q3 2026)

Image support (vision)
Audio support (transcription)
Knowledge graph (Neo4j)

v1.0 (Q4 2026)

Production-ready
Desktop application (Electron)
Optional encrypted sync

🤝 Contributing

OMEGA is an open-source and evolving project. Contributions are welcome!

Areas of Interest

🧠 Neuroscience: Improve consolidation models
🔢 Mathematics: Formalize heat score (diffusion equations)
💻 Dev: Optimization, tests, refactoring
🎨 Design: User interface for neurodivergent minds
📊 Data Science: Memory pattern analysis

Process

Fork the repo
Create a branch (git checkout -b feature/amazing-feature)
Commit (git commit -m 'Add amazing feature')
Push (git push origin feature/amazing-feature)
Open a Pull Request

📄 License

MIT License - see LICENSE

🙏 Acknowledgments

Anthropic for inspirations on memory systems
Ollama Team for local LLM inference
Sentence-Transformers for lightweight embeddings
The ALife community for discussions on emergence

📬 Contact & Community

GitHub: @ModelingSolver
Issues: Report bugs
Discussions: Join the conversation

📊 Project Status

Current version: 0.1-alpha
Last update: January 10, 2025
Tests: 5/5 stable sessions
Platform: Linux, Windows, macOS

💭 Quote

"ΩMEGA is not an assistant. It's a cognitive mirror that grows with you, remembers what matters, and forgets what's no longer important. Exactly what a true intellectual partner should do."
— Fish (Creator)

Built with ❤️ for neurodivergent minds worldwide.

🧠 Think Different. Remember Better. Co-Emerge. 🌌

Name		Name	Last commit message	Last commit date
BenchPress.md		BenchPress.md
Dev-Log.md		Dev-Log.md
LICENSE		LICENSE
README.md		README.md
Roadmap.md		Roadmap.md
Theoretical-Framework.md		Theoretical-Framework.md
USER-RESEARCH.md		USER-RESEARCH.md
raw-thoughts.md		raw-thoughts.md

Latest commit

History

Folders and files

Repository files navigation