entropic-core 3.0.0

Privacy-first entropy monitoring with active LLM intervention for multi-agent systems

Entropic Core v1.0

100% FREE & OPEN SOURCE entropy regulation for multi-agent AI systems.

Entropic Core monitors and automatically regulates chaos in multi-agent systems, preventing both system collapse and stagnation. Think of it as a thermostat for your AI agents.

What Makes This Special?

• Completely FREE - No tiers, no limits, no paywalls
• Open Source - MIT License, use it anywhere
• No API Keys Required - Works entirely locally
• Production Ready - Battle-tested algorithms
• Framework Agnostic - Works with AutoGen, LangChain, CrewAI, or custom agents

What Problem Does It Solve?

Multi-agent systems have a critical problem: they either become too chaotic (agents conflict, system crashes) or too ordered (agents stagnate, no innovation). Entropic Core solves this by:

1. Measuring entropy across 3 dimensions (decisions, state dispersion, communication)
2. Automatically regulating the system to maintain optimal chaos/order balance
3. Learning patterns to prevent future failures
4. Predicting problems before they happen
5. Diagnosing root causes when issues occur

Quick Start

Installation

# Basic installation (core features)
pip install entropic-core

# Full installation (all features)
pip install entropic-core[full]

Basic Usage

from entropic_core import EntropyBrain

# Initialize
brain = EntropyBrain()

# Connect your agents (unlimited)
brain.connect([agent1, agent2, agent3])

# Monitor and regulate automatically
brain.run(cycles=100)

Integration with AutoGen

from autogen import AssistantAgent
from entropic_core import EntropyBrain

# Your existing AutoGen agents
writer = AssistantAgent("writer", llm_config={...})
critic = AssistantAgent("critic", llm_config={...})

# Add entropy monitoring
brain = EntropyBrain()
brain.connect([writer, critic])

# Run with entropy awareness
for i in range(10):
    writer.generate_reply(messages)
    
    # Measure and regulate
    metrics = brain.measure()
    if metrics['combined'] > 0.8:
        brain.regulate()  # Automatically stabilizes

All Features Included (100% Free)

Core Features

• Real-time entropy monitoring (3 metrics)
• Automatic chaos/order regulation
• SQLite/PostgreSQL memory storage
• Universal agent adapter
• Pattern learning and recognition
• Unlimited agents
• Unlimited measurements

Advanced Analytics

• Causal Analysis - Diagnoses WHY entropy spiked
• Predictive Forecasting - Predicts failures before they happen
• Anomaly Detection - Catches unusual patterns
• Pattern Recognition - Learns from history
• Time Series Analysis - Tracks entropy trends

Enterprise Features

• Real-time web dashboard
• Automated PDF/HTML reports
• Multi-system orchestration
• Compliance logging & audit trails
• REST API
• Slack/Email/Webhook alerts

Examples

Example 1: Basic Monitoring

from entropic_core import EntropyBrain
from entropic_core.core.agent_adapter import AgentAdapter

# Create mock agents
agents = [
    AgentAdapter.create_mock_agent(f"agent_{i}", behavior='balanced')
    for i in range(5)
]

# Initialize brain
brain = EntropyBrain(auto_regulate=True)
brain.connect(agents)

# Run for 10 cycles
brain.run(cycles=10)

# Get status
status = brain.get_status()
print(f"Current entropy: {status['current_entropy']:.3f}")

Example 2: Causal Diagnosis

from entropic_core import EntropyBrain

brain = EntropyBrain()
brain.connect(my_agents)

# Build history
for i in range(20):
    my_agents[i % len(my_agents)].act(observation)
    brain.measure()

# Diagnose problems
diagnosis = brain.diagnose()
print(f"Root cause: {diagnosis['primary_cause']}")
print(f"Confidence: {diagnosis['confidence']:.1%}")
print(f"Fix: {diagnosis['suggested_fix']}")

Example 3: Predictive Monitoring

brain = EntropyBrain()
brain.connect(my_agents)

# Get forecast
forecast = brain.forecast(steps=10)
print(f"Risk level: {forecast['risk_level']}")

if forecast['time_to_collapse']:
    print(f"WARNING: Collapse in {forecast['time_to_collapse']} steps!")
    print("Recommended actions:")
    for action in forecast['recommended_preventive_actions']:
        print(f"  - {action}")

How It Works

1. Entropy Measurement

Entropic Core calculates entropy across three dimensions:

• Decision Entropy: How unpredictable agent decisions are (Shannon entropy)
• State Dispersion: How different agent states are from each other
• Communication Complexity: How much inter-agent communication is happening

These combine into a single "combined entropy" score from 0 to 1.

2. Automatic Regulation

Based on entropy levels, the system takes action:

• High entropy (>0.8): System too chaotic → Reduce exploration, merge similar agents, enforce protocols
• Low entropy (<0.2): System too ordered → Inject randomness, create explorer agents, relax constraints
• Optimal (0.4-0.6): Maintain homeostasis → Fine-tune parameters

3. Learning & Memory

Every decision and outcome is stored in SQLite/PostgreSQL:

• Events: What happened and when
• Patterns: What worked in similar situations
• Rules: Which regulations were effective
• Metrics: Time-series entropy data

The system learns from history to make better decisions over time.

4. Causal Analysis

When entropy spikes, the causal analyzer:

1. Analyzes correlations in recent history
2. Identifies root causes with confidence scores
3. Searches for similar past events
4. Generates specific fix recommendations

5. Predictive Engine

The predictive engine:

1. Forecasts entropy using time-series analysis
2. Predicts time until system collapse or stagnation
3. Detects anomalies using statistical methods
4. Recommends preventive actions

Architecture

entropic-core/
├── core/                    # Core modules (always included)
│   ├── entropy_monitor.py   # Measures entropy
│   ├── entropy_regulator.py # Takes regulatory action
│   ├── evolutionary_memory.py # Persistent storage
│   └── agent_adapter.py     # Universal agent wrapper
│
├── advanced/               # Advanced analytics (free)
│   ├── causal_analyzer.py  # Root cause diagnosis
│   ├── predictive_engine.py # Forecasting & anomalies
│   ├── simulation_mode.py  # Scenario simulation
│   └── security_layer.py   # Attack detection
│
├── integrations/           # Framework adapters (free)
│   ├── autogen_adapter.py  # AutoGen integration
│   ├── langchain_adapter.py # LangChain integration
│   └── custom_builder.py   # Custom adapter builder
│
├── visualization/          # Dashboards & reports (free)
│   ├── dashboard.py        # Real-time web dashboard
│   ├── report_generator.py # Automated reports
│   └── alert_system.py     # Multi-channel alerts
│
├── enterprise/            # Enterprise features (free)
│   ├── orchestrator.py    # Multi-system coordination
│   ├── compliance.py      # Audit & compliance
│   └── marketplace.py     # Pattern sharing
│
└── brain.py               # Main orchestrator

Framework Integrations

Entropic Core works with any agent framework:

• AutoGen: Full integration with conversation agents
• LangChain: Compatible with chains and agents
• CrewAI: Works with crew-based systems
• Custom: Universal adapter for any agent architecture

Performance

• Overhead: <5ms per measurement cycle
• Memory: ~10MB base + 1KB per agent
• Storage: ~1MB per 10,000 cycles
• Scalability: Tested with 1,000+ agents

Use Cases

Financial Trading

Monitor multi-agent trading systems to prevent both runaway risk and missed opportunities.

Game Development

Keep NPC behaviors interesting but not chaotic, preventing both boredom and frustration.

Research Labs

Coordinate multiple research agents exploring solution spaces without getting stuck or diverging.

Production Systems

Ensure agent-based microservices maintain healthy communication patterns without overload.

Installation Options

# Minimal (core only)
pip install entropic-core

# With advanced analytics
pip install entropic-core scipy scikit-learn

# With visualization
pip install entropic-core flask plotly pandas

# Everything (recommended)
pip install entropic-core[full]

Why 100% Free?

We believe that fundamental infrastructure for AI safety should be accessible to everyone. Entropic Core is:

• MIT Licensed - Use it anywhere, including commercial projects
• No Hidden Costs - No API calls, no cloud services required
• Community Driven - Contributions welcome
• Research Friendly - Perfect for academic use

Support & Community

• Documentation: https://github.com/entropic-core/entropic-core/wiki
• Issues: https://github.com/entropic-core/entropic-core/issues
• Discussions: https://github.com/entropic-core/entropic-core/discussions
• Discord: Join our community

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

License

MIT License - Use it anywhere, no restrictions.

Citation

If you use Entropic Core in research, please cite:

@software{entropic_core_2025,
  title={Entropic Core: Entropy Regulation for Multi-Agent Systems},
  author={Entropic Core Team},
  year={2025},
  url={https://github.com/entropic-core/entropic-core},
  license={MIT}
}