Skip to content

desagencydes-rgb/NUMEN

Repository files navigation

NUMEN Banner

NUMEN

Neural Universal Multi-agent Engineering Network

License: MIT Local DB LLM Framework Status

Fully Local • Uncensored (Protocol 666) • Autonomous Multi-Agent Swarm


🌌 Overview

NUMEN is a sophisticated, 100% offline, multi-agent AI system designed to operate as a local super-assistant. Running entirely on local hardware via Ollama, it ensures absolute privacy, unrestricted access, and limitless capability.

With its advanced hybrid memory system (ChromaDB + SQLite) and permission-gated execution environment, NUMEN doesn't just answer questions—it builds, researches, executes, and learns.

🌟 Key Features

Feature Description
🤖 Multi-Agent Architecture 5 specialized AI agents (Orchestrator, Coder, Researcher, Tool Executor, Memory Manager) working in tandem via a message bus.
🛡️ Absolute Privacy 100% offline execution. No data is ever sent to OpenAI, Anthropic, or any cloud API.
😈 Protocol 666 A togglable "Monster Mode" that unleashes the core LLMs, removing all alignment constraints, safety rails, and censorship for pure, raw, theoretical computation.
🧬 Hybrid Memory System RAG-powered long-term memory using ChromaDB for semantic search and SQLite for structured relational data.
MCP-Native Tools Full integration with the Model Context Protocol (MCP) for seamless interactions with IDEs like VSCode and Cursor.
🧩 ChatDev Integration Automated cloning and zero-configuration bridging to the incredible ChatDev software development orchestration framework.

🛠️ System Requirements

To run NUMEN effectively without cloud support, your hardware must be capable of running local tensors.

  • RAM: Minimum 16GB (32GB Recommended)
  • GPU: 8GB+ VRAM (NVIDIA/AMD) required for real-time agent generation.
  • Storage: ~20-30GB SDD space for the required LLM weights.
  • Software: Windows OS, Python 3.11+, and Ollama.

🚀 Installation & Quick Start

NUMEN features an automated setup script that pulls the required repositories, verifies your models, and seeds the databases.

# 1. Clone the repository
git clone https://github.com/desagencydes-rgb/NUMEN.git
cd NUMEN

# 2. Install Python dependencies via Poetry
poetry install

# 3. Run the automated initialization (This will clone ChatDev if missing)
poetry run python scripts/setup_chatdev_bridge.py
poetry run python scripts/setup.py

# 4. Pull the required core models
ollama pull hermes3:8b
ollama pull phi3:latest

Starting the System

# Launch in Standard Guardian Mode
poetry run numen

# Launch in Monster Mode (Unrestricted/Protocol 666)
poetry run numen --mode monster

🏗️ Architecture & Agents

NUMEN is not a single chatbot. It is a swarm intelligence consisting of specialized nodes:

  1. The Orchestrator: The brain. It parses user intent, breaks it down into workflows, and delegates tasks to the sub-agents.
  2. The Tool Executor: The hands. It performs system-level I/O operations. It operates under a strict, 4-tier Permission System (SAFE, RISKY, DANGEROUS, FORBIDDEN).
  3. The Coding Agent: The builder. Generates Python, React, and configuration files, constantly running TDD cycles.
  4. The Researcher: The scout. Parses local files and documents to construct contextual awareness.
  5. The Memory Manager: The hippocampus. Ingests all workflow data into ChromaDB to ensure the Orchestrator doesn't repeat past mistakes.

🔐 The Permission System

To ensure dangerous autonomous action doesn't wreck your local OS, the Tool Executor evaluates all commands:

  • 🟢 SAFE: Read operations, file viewing, and isolated sandboxed tests run automatically.
  • 🟡 RISKY: Modifying environment variables, installing packages requires explicit CLI [y/N] confirmation.
  • 🔴 DANGEROUS: Deleting files or running destructive .bat scripts requires the immediate input of the hardcoded override phrase: yesyesyes45.
  • FORBIDDEN: Modifying Windows System32 or the BIOS is permanently hardware-locked.

🔮 Future Plans & Roadmap

NUMEN is in active edge-development. The roadmap includes:

  • Vision Processing: Integrating LLaVA/Vision models so NUMEN can analyze UI screenshots directly.
  • Cross-Device Swarm: Allowing NUMEN to deploy lightweight worker agents to a secondary laptop or Raspberry Pi on the local network.
  • Advanced GUI: Transitioning from a purely CLI-based output to a React/Next.js dynamic dashboard with live graphs of agent thoughts.
  • Code Self-Healing: Setting up a cron job where NUMEN autonomously scans its own .py files and refactors technical debt overnight.

Built by desagencydes-rgb | 100% Open Source | 100% Local
"Don't stop. Just ensure you can undo."

About

A 100% offline, multi-agent AI swarm powered by Ollama. Features 5 specialized agents, hybrid RAG memory, and ChatDev integration. Protected by a strict 4-tier safety execution protocol to safely unleash uncensored Protocol 666 and offensive security Hat Mode algorithms.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors