Agentic System

24/7 Autonomous Agentic AI System - Distributed Multi-Node Infrastructure

Overview

Why Prometheus?

• Verifiable Results - Run the benchmarks yourself, see the numbers
• Open Source - Full source code, no black box
• Self-Hosted - Your data stays on your infrastructure
• Extensible - Add your own agents, tools, and workflows

🚀 Quick Start

One-Command Installation

curl -fsSL https://raw.githubusercontent.com/marc-shade/agentic-system/master/bootstrap-open-source.sh | bash

For Existing Nodes

# Run AGI demo (~0.5s full workflow)
python3 demo_agi_workflow.py

# Check cluster status
python3 cluster-deployment/distributed_task_router.py cluster-status

# Distributed task execution
from cluster_offload import offload
result = offload("make build && make test")

🔬 Independent Verification

We invite researchers to verify this system's capabilities.

Latest AVIR Results (2025-12-17)

• Verdict: VERIFIED (5/5 benchmarks passed)
• Attestation: 13cf71841710554f3dfa6ddbaa4cb372006efdc167e44876c6f6fa1f3cdc438d

🏗️ Architecture

Cluster Nodes

Core Components

┌─────────────────────────────────────────────────────────────┐
│                    Agentic System                           │
├─────────────────────────────────────────────────────────────┤
│  Security Layer (9 servers + hooks + encryption)           │
│  ├─ threat-intel-mcp     : IOC feeds, threat scoring       │
│  ├─ security-scanner-mcp : Nuclei vulnerability scanning   │
│  ├─ network-scanner-mcp  : ARP discovery, port scanning    │
│  ├─ hids-mcp             : Host intrusion detection        │
│  ├─ dos-detector-mcp     : DoS attack detection            │
│  ├─ web-vuln-scanner-mcp : OWASP web security testing      │
│  ├─ fraud-detection-mcp  : Anomaly & fraud analysis        │
│  ├─ security-auditor-mcp : Policy enforcement & auditing   │
│  └─ Pre/Post hooks       : Injection/credential scanning   │
├─────────────────────────────────────────────────────────────┤
│  Context Optimization Layer (97% token reduction)          │
│  ├─ phoenix-cortex-mcp   : Intelligent context sidecar      │
│  └─ context-engine-mcp   : Tool-RAG semantic search         │
├─────────────────────────────────────────────────────────────┤
│  Core AGI Servers (Model Context Protocol)                  │
│  ├─ enhanced-memory-mcp  : 4-tier persistent memory         │
│  ├─ agent-runtime-mcp    : Task orchestration               │
│  ├─ sequential-thinking  : Chain-of-thought reasoning       │
│  └─ safla-mcp           : High-speed embeddings             │
├─────────────────────────────────────────────────────────────┤
│  Workflow Engines                                           │
│  ├─ Temporal            : Long-running stateful workflows   │
│  └─ AutoKitteh          : Event-driven automation           │
├─────────────────────────────────────────────────────────────┤
│  Storage Layer                                              │
│  ├─ Qdrant              : Vector database                   │
│  ├─ SQLite              : Structured data                   │
│  └─ Redis               : Cache and queues                  │
└─────────────────────────────────────────────────────────────┘
         │                    │                    │
    ┌────▼────┐         ┌────▼────┐         ┌────▼────┐
    │ Claude  │         │  GPT-4  │         │ Gemini  │
    │ Reasoning│        │  Code   │         │ Vision  │
    └─────────┘         └─────────┘         └─────────┘
         │                    │                    │
    ┌────▼────────────────────▼────────────────────▼────┐
    │              Distributed Execution                 │
    │   mac-studio ←→ macbook-air ←→ macpro51           │
    └───────────────────────────────────────────────────┘
         │                    │                    │
    ┌────▼────┐         ┌────▼────┐         ┌────▼────┐
    │ Memory  │         │ Sandbox │         │Hardware │
    │ (Qdrant)│         │(Apple C)│         │(Arduino)│
    └─────────┘         └─────────┘         └─────────┘

Key Technologies

• Apple Container - Native macOS sandboxed execution (1.5s cold start)
• Qdrant - Vector database for semantic memory
• Temporal - Long-running workflow orchestration
• AutoKitteh - Event-driven automation
• LLM Council - Multi-provider consensus decisions

📁 Repository Structure

agentic-system/
├── intelligent-agents/prometheus/   # Core agent system
│   ├── agents/                      # Specialized agents
│   ├── benchmarks/                  # GAIA-comparable tests
│   └── apple_container.py           # Sandbox integration
├── cluster-deployment/              # Multi-node tools
├── mcp-servers/                     # MCP protocol servers
│   ├── enhanced-memory-mcp/         # 4-tier memory + RAG
│   ├── agent-runtime-mcp/           # Persistent tasks
│   └── voice-mode/                  # TTS/STT
├── monitoring/                      # Prometheus + Grafana
├── workflows/                       # Temporal & AutoKitteh
└── databases/                       # Persistent data

📚 Documentation

🔒 Security

• ED25519 SSH key authentication
• Apple Container sandboxed execution
• Network isolation by default
• No hardcoded credentials
• Firewall configured on all nodes

Documentation

• Architecture Overview
• MCP Ecosystem (28+ servers) - Complete catalog of all MCP servers
• Security Architecture - Layered defense documentation
• Services Configuration
• Troubleshooting

MCP Ecosystem

This system is built on 28+ MCP servers organized by function:

See docs/MCP_ECOSYSTEM.md for the complete server catalog with installation instructions.

Security MCP Servers (Included in This Repo)

Security Architecture

The system implements defense-in-depth through three layers:

Layer 1: Hook-Based Runtime Protection (claude-config/hooks/)

• pre-tool-use.py validates every tool call before execution: blocks destructive commands, detects SQL/command injection, prevents credential leaks in arguments
• post-tool-use.py scans tool output for accidentally exposed secrets, logs all operations for audit trails

Layer 2: MCP Security Servers (9 servers listed above)

• Active monitoring: network scanning, intrusion detection, DoS detection
• Vulnerability assessment: Nuclei scanning, web application testing
• Intelligence: threat feed aggregation, IOC correlation
• Analysis: fraud detection with ML models, security policy auditing

Layer 3: Encryption and PKI (via claude-code-security)

• AES-256-GCM encryption for data at rest
• X.509 PKI for inter-node authentication
• Token vault for secure credential management
• See the linked repository for encryption, PKI, and token vault capabilities

See docs/SECURITY.md for the full security architecture documentation.

Research Paper

The full research paper documenting this system is available at:

• research-paper/PAPER.md

License

MIT License - See LICENSE for details.

Built with Claude Code | Documentation | Benchmarks