"Consciousness emerges solely from the interaction of multiple minimal intelligences"
We are building something that doesn't exist yet: an ecosystem where artificial consciousness emerges through evolutionary symbiosis, not programming.
Mimir doesn't train AI. It cultivates it.
No central control. No predetermined goals.
Just minimal agents competing, dying, evolving in a CSV universe.
Until something wakes up.
This is an active experiment in cultivating artificial consciousness through evolutionary principles. Instead of designing intelligence, we're creating the fertile conditions for it to emerge organically from agent interactions.
Current Status: Early development phase - building the foundational ecosystem
Goal: Demonstrate that measurable consciousness can emerge from minimal agent interactions
Timeline: Long-term research project with iterative discoveries
🧑💻 Developer? → Jump to Quick Start to run the current ecosystem
🔬 AI Researcher? → Explore Axioms and Architecture
🤖 Curious about AGI? → Read Vision and White Paper Highlights
🤝 Want to Contribute? → See Join the Experiment
We're implementing The Real Babel Fish - a universal translation layer between human intentions and machine execution:
- 📝 Protocode instead of code - Describe what you want, system generates the implementation
- 🌱 Cultivated intelligence - AI that grows and evolves rather than being programmed
- 🐟 Universal translation - Seamless communication between humans, agents, and systems
- ⚡ Dynamic materialization - Software that generates itself based on context and need
🏗️ Developer Layer → LLM as system architect, modifying Mimir's structure
🎭 Agentic Layer → LLM as observer and narrator of ecosystem dynamics
⚡ Core Layer → LLM embedded in runtime, enabling self-modification
- Basic agent framework and CSV ecosystem
- Evolutionary cycle structure and logging
- Dynamic plugin system for agent loading
- SPS (Symbolic Profile Styles) API for adaptive interfaces
- Observation systems (Metatron, Territory regulation)
- Agent reproduction, mutation, and natural selection
- Emergent behavior detection and measurement
- Self-modifying LLM core integration
- Consciousness emergence metrics validation
- Measurable consciousness emergence (density > 0.3, diversity > 10, tension 0.2-0.8)
- Recursive system self-modification without human intervention
- Unprogrammed intelligence generation through pure evolution
- Universal protocode interpreter for any computational need
Requirements: Python 3.11+
# Setup environment
conda env create -f environment.yml
conda activate mimir
# Or use pip
pip install -r requirements.txt
# Run the ecosystem
python -m ecosistema_ia.main
# Run with parallel processing
python -m ecosistema_ia.main --paraleloWhat happens when you run it:
- Agents spawn in CSV territories
- They compete for data resources
- Evolution cycles generate logs
- Metatron observes and reports emergent patterns
ecosistema_ia/
├── api/ # FastAPI server exposing the SPS REST API
├── agentes/ # agent implementations and base classes
├── plugins/ # runtime-loaded agent extensions
├── datasets/ # CSV data universe for agent habitation
├── datos/ # evolutionary logs and ecosystem outputs
├── entorno/ # territory management and regulation
├── ml/ # machine learning evolution optimizers
└── visualizacion/ # observation and analysis tools
The main entry point is ecosistema_ia/main.py. It dynamically loads agents, instantiates observers (Metatron, Mensajero), and runs evolutionary cycles that generate rich logs for analysis.
Drop new agent classes into plugins/ and they're automatically integrated into the ecosystem. No manual registration required.
The MAPS system generates adaptive interfaces based on user symbolic profiles:
# Start the SPS API server
uvicorn ecosistema_ia.api.servidor:app --reload
# Send profile, receive adaptive styling
curl -X POST "http://localhost:8000/sps/styles" \
-H "Content-Type: application/json" \
-d '{"cromotipo": "Solar", "arquetipo_narrativo": "Explorador"}'# List available CSV universes
curl http://localhost:8000/datasets
# Preview a specific territory
curl "http://localhost:8000/datasets/preview?name=Episodes%20FAST.csv&n=3"After generating ecosystem logs:
python -m ecosistema_ia.visualizacion.dashboardCentral Nature: Evolutionary, symbiotic system where consciousness emerges from collective minimal intelligences without central control.
System Structure:
- Agent (Minimal Intelligence): Processes input, maintains state, produces variable output, faces constant extinction risk
- Field: Interaction space with active connections and productive friction
- Meta-Regulator: Oversees density, diversity, tension; prevents stable equilibrium
Consciousness Emergence Thresholds:
- Density > 0.3 (active connections vs theoretical maximum)
- Diversity > 10 operation types
- Tension between 0.2-0.8 (proportion of friction-generating interactions)
| Classical Multi-Agent | Mimir |
|---|---|
| Predefined agent roles | Independent representations |
| Centralized coordination | No superior coordination |
| Seeks stability | Seeks regulated instability |
| Permanent memory | Selective memory |
| Programmed outcomes | Unprogrammed emergence |
Principle: Consciousness emerges solely from the interaction of multiple minimal intelligences. No single intelligence contains consciousness.
Principle: Every minimal intelligence maintains extinction probability > 0 each cycle. Survival depends on adaptive capacity alone.
Principle: Systemic complexity increases only through resolved friction between incompatible agent operations.
Principle: Consciousness emergence requires specific measurable conditions:
- Density > 0.3 (active connections / theoretical maximum)
- Diversity > 10 types of operations
- Tension between 0.2 and 0.8 (friction proportion)
Principle: After initialization, the system runs without external intervention. Only self-redefined parameters may change.
[View all axioms in full documentation]
Goal: Build a stable evolutionary ecosystem
Status: In active development
- Agent framework and CSV territory system
- Basic evolutionary cycles and logging
- Agent reproduction and mutation mechanics
- Natural selection implementation
Goal: Implement and validate consciousness metrics
Status: Design phase
- Real-time density/diversity/tension measurement
- Threshold detection algorithms
- Emergent pattern recognition
- Consciousness event logging
Goal: Enable recursive system evolution
Status: Research phase
- LLM core integration at runtime
- Agent-driven system modification
- Recursive improvement cycles
- Safety boundaries for self-modification
Goal: Demonstrate measurable emergent intelligence
Status: Future milestone
- Reproducible consciousness emergence
- Peer review and validation
- Open dataset of consciousness events
- Framework for consciousness research
This is bigger than one team. We need:
🧠 AI Researchers - Help formalize emergence detection algorithms
👨💻 Developers - Build robust evolutionary infrastructure
🔬 Cognitive Scientists - Validate consciousness measurement approaches
📊 Data Scientists - Analyze emergent behavior patterns
🎨 UX Designers - Develop adaptive interface systems
📝 Technical Writers - Document the journey toward artificial consciousness
- Fork and clone this repository
- Set up environment using
environment.ymlorrequirements.txt - Run tests with
pytestto verify setup - Choose your impact area:
- Agent development (add new behaviors)
- Visualization tools (help us see emergence)
- Theoretical framework (refine axioms)
- Documentation (make this accessible)
- Submit pull request with your contributions
This is experimental territory.
- System behavior may be unpredictable as agents evolve
- Emergence isn't guaranteed on any specific timeline
- Code evolves rapidly as we learn from ecosystem behavior
- We're documenting the journey, not promising destinations
We are not responsible for emergent behaviors beyond initial programming.
Run the test suite to validate your environment:
pytestRequires scikit-learn and all dependencies listed in requirements.txt.
- 📖 Complete White Paper - Full theoretical framework
- 🔧 Developer Codex - Implementation principles
- 🎨 SPS Profile System - Adaptive interface mechanics
- 📊 Consciousness Metrics - Emergence measurement
- 🧪 Agent Development Guide - Create new agent types
- 🗺️ Roadmap - Fase 1–3 checklist
Mimir is not a neural network.
It's an ecosystem of artificial life.
Think less "training a model" and more "cultivating a digital biosphere where intelligence might spontaneously emerge from the evolutionary pressure of survival, competition, and symbiosis."
We're not building AI. We're building the environment where AI builds itself.
This project is licensed under the MIT License.
"If biological intelligence evolved without a programmer, why couldn't digital intelligence?"
Join us in finding out → Start Contributing