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.
Current conversational AIs are amnesic: they forget everything between sessions, force users to repeat themselves, and never truly adapt to their interlocutor.
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.)
| 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 |
┌─────────────────────────────────────────────────────────────┐
│ 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 │
└─────────────────────────────────────────────────────────────┘
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
- ST full (>5 messages) → LLM compression → MT summary
- MT consolidated (every 10 turns) → Heat-based sorting → LT or trash
- Encoding LT memories as vectors (sentence-transformers)
- Cosine similarity search during questions
- Injection of the 2 most relevant memories into context
# Python 3.8+
python --version
# Ollama (for local LLM)
# Download at https://ollama.ai# 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 servepython 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...
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...
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
- Calls to local LLM (Ollama)
- Robust error handling
- Modes:
think()(chat) andsummarize()(compression)
- ST/MT/LT management
- Dynamic heat score with
boost_heat_for_relevant_mt() - Semantic retrieval via sentence-transformers
- Automatic consolidation and pruning
- Conversational loop
- Context injection in prompts
- Triggering compressions and consolidations
| 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 |
📋 Test 5: First truly conclusive test (complete logs)
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.
-
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
- 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.
In brain.py, line 5:
def __init__(self, model_name="phi3:mini"): # Default modelRecommended models:
phi3:mini(1.9GB) → Fast, modest PCs ✅llama3.1:8b(4.7GB) → Better qualitymistral(4.1GB) → Good compromisetinyllama(637MB) → Ultra-lightweight (testing)
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 thresholdIn brain.py:
# Max generation duration
timeout=60 # seconds
# Max response length
"num_predict": 150 # tokens for chat
"num_predict": 50 # tokens for summaries- Streamlit interface for memory visualization
- Session export/import
- Temporal decay of MT memories
- Multi-user support (profiles)
- Multi-document context management
- REST API for integrations
- Image support (vision)
- Audio support (transcription)
- Knowledge graph (Neo4j)
- Production-ready
- Desktop application (Electron)
- Optional encrypted sync
OMEGA is an open-source and evolving project. Contributions are welcome!
- 🧠 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
- 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
MIT License - see LICENSE
- Anthropic for inspirations on memory systems
- Ollama Team for local LLM inference
- Sentence-Transformers for lightweight embeddings
- The ALife community for discussions on emergence
- GitHub: @ModelingSolver
- Issues: Report bugs
- Discussions: Join the conversation
Current version: 0.1-alpha
Last update: January 10, 2025
Tests: 5/5 stable sessions
Platform: Linux, Windows, macOS
"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. 🌌