Smol NAS AI 🧾🤖

Autonomous AI-Powered NAS Management Platform

Project Goal: Simplify server administration through a natural language interface and a streamlined web UI.

---

📖 What is Smol NAS AI?

Smol NAS AI is an intelligent storage management platform that transforms complex server administration into intuitive interactions. By combining a custom-tuned GPT-2 model with a robust Flask-based backend, Smol allows users to manage Samba shares, RAID arrays, and user permissions using natural language instructions.

Traditional NAS management requires deep CLI knowledge. Smol bridge this gap by providing:

1. AI Command Generation: A 250M parameter GPT-2 model trained specifically for NAS CLI commands.
2. Real-time System Monitoring: Live tracking of CPU, RAM, Temperature (optimized for Raspberry Pi), and Network I/O.
3. Autonomous Share Management: One-click Samba configuration and user delegation.

---

🖼️ Application Screenshots

🖥️ Main Dashboard

_{Centralized monitoring of system health, network traffic, and active Samba sessions.}

---

📁 Management Modules

Storage & RAID Visual disk management: mount/unmount, format, and RAID status tracking via `lsblk`.	User Controls Manage Linux and Samba users. Integration with `chpasswd` and `smbpasswd`.
File Manager Web-based file management with integrated permission controls.	AI Assistant Instruction-based command generation. Ask "Create a Samba share" and get the CLI command instantly.

---

🏗️ System Architecture

graph TD
    Client[Client Browser] -->|HTTPS| Flask[Flask Backend]
    Flask -->|Subprocess| Shell[Linux Shell / CLI]
    Flask -->|Torch| GPT2[NAS-GPT2 Model]
    Shell -->|Config| Samba[Samba / smb.conf]
    Shell -->|Monitoring| psutil[psutil / lsblk]
    Flask -->|Auth| LinuxAuth[Linux System Auth / sshpass]
    Flask -->|VPN| Tailscale[Tailscale IP Discovery]

Loading

---

🧠 AI Pipeline Deep-Dive

Custom NAS-GPT2 Model

The heart of Smol is a specialized GPT-2 model fine-tuned on a high-quality dataset of NAS administrative tasks.

# Inference Logic (test.py)
inputs = tokenizer(prompt, return_tensors="pt").to(device)
outputs = model.generate(
    **inputs,
    max_new_tokens=150,
    temperature=0.2,
    top_p=0.85,
    repetition_penalty=1.1
)

Command Categories

The model supports structured instructions across multiple domains:

• [SAMBA]: Share creation, user access, and global settings.
• [RAID]: Array creation, disk addition, and status checks.
• [NFS]: Export management and mount configurations.
• [PERMISSION]: chown and chmod generation.
• [SERVICE]: Systemd service control and server reboots.

---

🛠️ Full Technology Stack

Layer	Technology	Why
Backend Framework	Flask	Lightweight, extensible, perfect for embedded systems
AI / LLM	GPT-2 (250M)	Local execution, no internet required for core AI
Machine Learning	PyTorch + Transformers	Industry standard for NLP tasks
System Monitoring	psutil	Cross-platform system metrics extraction
Storage Interface	lsblk + mkfs	Direct interaction with Linux block devices
Authentication	Flask-Login + sshpass	Secure session management with system integration
Networking	Tailscale	Easy remote access without port forwarding
Frontend	Jinja2 Templates + HTML/CSS	Minimal overhead for server-side rendering

---

🚀 Local Development Setup

Prerequisites

• Python 3.10+
• Git LFS (for model weights)
• Linux environment (recommended for full system feature support)

Installation

1. Clone and Setup:

   git clone https://github.com/Vector3451/Smol.git
   cd Smol
   pip install -r requirements.txt

2. Pull Model Weights:

   git lfs pull

3. Configure Environment: Create a .env or set environment variables:

   SECRET_KEY=your_secret_key
   NAS_USER=admin
   NAS_PASSWORD=your_password
   NAS_ENV=production  # Use 'development' for mock system calls

4. Run the Application:

   cd NAS
   python app.py

   Access the UI at http://localhost:5000.

---

📁 Project Structure

Smol/
├── assets/screenshots/    # README visual documentation
├── nas_model_250m/        # Full PyTorch model weights (LFS)
├── nas_model_q8.gguf      # Quantized model for efficient inference
├── tokenizer/             # Custom tokenizer configurations
├── NAS/                   # Core Application Logic
│   ├── app.py             # Flask Entrypoint & Route Handlers
│   ├── diagnostic.py      # System Readiness Checker
│   ├── servicenow.py      # Integration for incident reporting
│   └── templates/         # UI View Templates
├── test.py                # Standalone Inference Script
└── README.md              # This file

---

🤝 Authors

Built for modern storage management. For questions, contact the repository owner.