Masih Maafi

Posted on Nov 6 • Originally published at masihmoafi.com

🏰 A-Modular-Kingdom: The Foundation for AI-Powered Multi-Agent Systems

#ai #multiagentsystems #rag #mcp

A comprehensive AI infrastructure providing building blocks for sophisticated multi-agent workflows. Built with modularity and standardization at its core, seamlessly connecting different multi-agent architectures through a unified foundation.

✨ Features

🔗 Seamless Integration - Multi-agent systems connect to host.py for instant access to long-term memory, RAG, and powerful tools
🏗️ Modular Architecture - Build hierarchical or sequential workflows on the same foundation
🛠️ Rich Toolset - Vision, code execution, browser automation, web search, and more
📚 Smart Memory - Persistent memory and RAG systems working across all agents
🌐 MCP Protocol - Model Context Protocol for reliable, structured interactions
🎤 Voice Control - Speech-to-text and text-to-speech capabilities

🏗️ Architecture

The foundation layer (host.py) provides unified access to all capabilities, allowing different multi-agent architectures to share the same powerful toolset and knowledge base.

┌─────────────────────────────────────┐
│         Multi-Agent Layer           │
│  ┌─────────────┐ ┌─────────────┐   │
│  │   Council   │ │     Gym     │   │
│  │   Chamber   │ │             │   │
│  └─────────────┘ └─────────────┘   │
└─────────────────────────────────────┘
              │         │
              ▼         ▼
┌─────────────────────────────────────┐
│         Foundation Layer            │
│  ┌─────────┐ ┌─────────┐ ┌────────┐│
│  │   RAG   │ │ Memory  │ │ Tools  ││
│  └─────────┘ └─────────┘ └────────┘│
│            host.py                  │
└─────────────────────────────────────┘

🛠️ Core Components

📡 Host.py - MCP Server

The central hub providing MCP (Model Context Protocol) access to all capabilities:

save_memory - Direct memory storage
search_memories - Semantic memory search
query_knowledge_base - RAG document search
web_search - Current information retrieval
browser_automation - Web interaction
code_execute - Safe Python execution
analyze_media - Image/video analysis
text_to_speech & speech_to_text - Voice capabilities

💬 Main.py - Interactive Client

Feature-rich chat interface with:

Auto-completion for @ mentions and / commands
Direct memory saving with # prefix
Automatic tool selection (memory vs RAG vs web)
Interactive command interface

📚 RAG System

Three versions with different strategies:

V1 - Basic Chroma + BM25 ensemble
V2 - FAISS + CrossEncoder reranking
V3 - Custom indexes + RRF fusion + LLM reranking

🤖 Multi-Agent Systems

👑 Council Chamber

Hierarchical multi-agent system with defined roles:
👑 King (User) → 👸 Queen Juliette → 🔥 Sexy Teacher → 🤖 Code Agent

Features:

Hierarchical validation
Smart task delegation
MCP tool integration
Code-first solutions with smolagents

💪 Gym

Sequential fitness-focused multi-agent system:
Interviewer → Plan Generator → Nutrition Agent

Features:

CrewAI powered workflows
Specialized fitness agents
Web interface
Flexible LLM support

🔧 Available Tools

Tool	Description	Status
RAG	Document retrieval	✅ 3 versions
Memory	Long-term storage	✅ Integrated
Vision	Image analysis	✅ Ready
Code Exec	Python sandbox	✅ Secure
Browser	Web automation	✅ Playwright
Web Search	Info retrieval	✅ DuckDuckGo
TTS	Text-to-speech	✅ Kokoro
STT	Speech-to-text	✅ Whisper

🚀 Quick Start

# Clone the repository
git clone https://github.com/MasihMoafi/A-Modular-Kingdom.git
cd A-Modular-Kingdom

# Install dependencies
uv sync
# Or with pip
pip install -r requirements.txt

# Start the MCP server
python agent/host.py

# In another terminal, start the interactive client
python agent/main.py

📝 Commands Reference

Memory Management

#message - Save directly to memory
/memory - List and manage memories

Document Access

@filename - Reference documents
/files - Show available documents

RAG Search

/rag "search term"        # Default: v2, current dir
/rag "search term" v3     # Specify version
/rag "search term" v2 /path/to/docs  # Custom path

🎯 CLI Integration

Claude Code and Gemini-cli both connect to host.py as clients.

Compatible with:

Gemini CLI Extension
Claude Code
Any MCP-compatible client

🔗 Links & Resources

Originally published at masihmoafi.com

A-Modular-Kingdom: Where AI agents come together in harmony 🏰✨

DEV Community