CAI - Context-Aware Infrastructure

---

Table of Contents

• Overview
• Architecture
• Components
• Quick Start
• Development
• Testing
• Configuration
• License

---

Overview

CAI is a modular framework combining:

• Vector database (LanceDB, 768-dim nomic-embed-text embeddings)
• Filesystem archive (Mycelium graph, branching 0.85/mutation 0.15)
• MCP server ecosystem (5 servers: memory, filesystem, orchestration, messaging)
• Batch consolidation (scheduled pattern extraction and tool generation)
• 4-layer storage hierarchy (global > vector > filesystem > workspace)

The system provides persistent cross-session storage, automatic context loading on workspace open, and scheduled pattern analysis.

---

Architecture

┌─────────────────────────────────────────────────────────────┐
│                     CAI System                              │
├─────────────────────────────────────────────────────────────┤
│                                                              │
│  ┌─────────────────────────────────────────────────────┐   │
│  │                  Storage Hierarchy                    │   │
│  │  ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌────────┐ │   │
│  │  │ GLOBAL   │ │ VECTOR   │ │ FILESYSTEM│ WORKSPACE│ │   │
│  │  │ config   │ │ LanceDB  │ │ Mycelium │ │ .clin. │ │   │
│  │  └──────────┘ └──────────┘ └──────────┘ └────────┘ │   │
│  └─────────────────────────────────────────────────────┘   │
│                                                              │
│  ┌─────────────────────────────────────────────────────┐   │
│  │                    MCP Servers                       │   │
│  │  ┌────────────┐ ┌──────────────┐ ┌──────────────┐  │   │
│  │  │ NeuralMem  │ │ MyceliumHuman│ │ Orchestrator  │  │   │
│  │  │ (LanceDB)  │ │ (Markdown FS)│ │ (Watchdog)    │  │   │
│  │  └────────────┘ └──────────────┘ └──────────────┘  │   │
│  └─────────────────────────────────────────────────────┘   │
│                                                              │
│  ┌─────────────────────────────────────────────────────┐   │
│  │              Batch Consolidation Pipeline            │   │
│  │  ┌────────┐ ┌────────┐ ┌──────────┐ ┌────────────┐ │   │
│  │  │ Ingest │→│Analyze │→│ Generate │→│   Deploy   │ │   │
│  │  │ (100   │ │(Pattern│ │ (FastMCP │ │ (Dynamic   │ │   │
│  │  │entries)│ │Detect) │ │  Tool)   │ │  Skills)   │ │   │
│  │  └────────┘ └────────┘ └──────────┘ └────────────┘ │   │
│  └─────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────┘

---

Components

Storage Layer

Component	Stack	Description
Vector Store	LanceDB + nomic-embed-text	768-dim persistent storage with semantic search
File Archive	Mycelium graph	Branching filesystem (factor 0.85) with mutation (factor 0.15)
Bridge	memory_bridge.py	Unified read/write across all storage layers
Auto-Load	recall() on init	Hydrates context from vector store on workspace open
Hierarchy	4 layers	Config > Vector > Filesystem > Workspace

MCP Servers

Server	Function
NeuralMemory	Vector database CRUD, semantic search
MyceliumHumanized	Markdown-format filesystem operations
Orchestrator	Server lifecycle management, hot-reload watchdog
ThreemaGateway	Message relay via Threema API

Batch Consolidation

Stage	Details
Schedule	Daily via Windows Task Scheduler
Input	Last 100 LanceDB entries
Analysis	Pattern detection via qwen2.5-coder:32b
Output	FastMCP tool for repetitive workflows
Deploy	%LOCALAPPDATA%\mcp-dynamic-skills hot-reload

---

Quick Start

Prerequisites

python --version              # 3.10+
ollama pull nomic-embed-text  # embeddings model
ollama pull qwen2.5-coder:32b # analysis model

Installation

git clone https://github.com/piot5/CAI.git
cd CAI

python -m venv .venv
.venv\Scripts\activate          # Windows
# source .venv/bin/activate     # Linux/Mac

pip install -r requirements.txt

Project Structure

CAI/
├── .clinerules           # Global configuration (4-layer hierarchy)
├── mcp.json              # MCP server registration
├── memory/               # Core storage and MCP servers
│   ├── mcp_neural_memory.py      # LanceDB vector store server
│   ├── mcp_mycelium.py           # Mycelium filesystem server
│   ├── mcp_categorizer.py        # Auto-categorization MCP
│   ├── mycelium_engine.py        # Mycelium filesystem engine
│   ├── memory_bridge.py          # Unified 4-layer memory bridge
│   ├── orchestrator.py           # MCP server lifecycle manager
│   ├── dream_cycle.py            # Batch consolidation pipeline
│   ├── path_config.py            # Path resolution
│   ├── validation.py             # Input validation
│   ├── categorizer.py            # Category engine
│   ├── mcp_status_check.py       # Health check utility
│   ├── cai_server.py             # Development server
│   ├── global_mcp_manager.py     # Global MCP activation
│   ├── autostart/                # Windows autostart components
│   │   └── autostart_orchestrator.py
│   ├── demos/                    # Integration demos
│   │   ├── tool_integration_demo.py
│   │   ├── complete_mcp_integration_demo.py
│   │   ├── practical_mcp_usage.py
│   │   └── mcp_learning_demonstration.py
│   ├── tests/                    # Test suite (36 tests)
│   │   └── test_mcps_comprehensive.py
│   └── docs/                     # Reports and summaries
├── requirements.txt
└── README.md

Development Server

python -m memory.cai_server              # Start all MCP servers
python -m memory.memory_bridge           # Storage layer only
python -m memory.orchestrator            # MCP server manager
python -m memory.mcp_status_check        # Health check
python -m memory.dream_cycle --dry-run   # Consolidation dry-run

---

Development

VSCode

Pre-configured in .vscode/:

File	Contents
settings.json	Python (Black, Ruff, isort), Markdown, Git
extensions.json	10 recommended extensions
launch.json	10 launch configs + 1 compound

Code Quality

black --line-length=100 memory/
isort --profile black --line-length=100 memory/
ruff check --line-length=100 memory/
mypy memory/

---

Testing

python -m memory.tests.test_mcps_comprehensive.py  # Full suite (36 tests)
python -m memory.mcp_status_check                 # Live system check

Component	Tests	Status
Neural Memory Server	8	✅
Mycelium Server	4	✅
Mycelium Humanized	6	✅
Memory Bridge	8	✅
Dynamic Skills	4	✅
Threema Gateway	2	✅
Orchestrator	4	✅
Total	36	✅

---

Configuration

MCP Server Registration

Global path (IDE-managed):

%APPDATA%\Code\User\globalStorage\rooveterinaryinc.roo-cline\settings\mcp.json

Local reference: c:\Software\CAI\mcp.json

Runtime Config

File: .clinerules (JSON at project root)

• Defines server paths, model names, consolidation schedule
• Scanned on workspace open
• Supports workspace-specific overrides

---

License

MIT License - see LICENSE for details.

---

v1.1.0 · github.com/piot5/CAI