Introduction

In today’s economic climate, "cost reduction and efficiency improvement" have become critical priorities. Large Language Models (LLMs) offer powerful support for automating tasks, boosting productivity, and driving innovation. However, reliance on cloud-based services introduces data security risks and recurring API costs. Deploying a local, private LLM within an intranet effectively addresses these challenges.

Why Choose On-Premises Deployment?

Advantages of Local Deployment

1. Data Security & Privacy: Data is processed locally, eliminating cloud uploads—ideal for sensitive industries like finance and legal.
2. Cost Control: One-time hardware investment replaces ongoing API fees, yielding long-term savings.
3. Low Latency: Local execution minimizes network delays for faster responses.
4. Stability: Avoids public network congestion that disrupts business operations.
5. Customization: Tailor models and workflows to enterprise needs.
6. Offline Availability: Operates without internet connectivity.

---

Software & Hardware Requirements

This tutorial uses a demonstration configuration. Adjust based on your needs.

Hardware

• Mac Mini M4: Apple Silicon M4 chip, minimum 16GB RAM for efficient model execution.

Software

• Ollama: Open-source tool for local LLM execution.
• OrbStack: Lightweight containerization platform (Docker Desktop alternative).
• DeepSeek-R1: High-performance open-source model for inference tasks.
• Open-WebUI: Intuitive web interface for model interaction.

---

Deployment Guide

Step 1: Install OrbStack

OrbStack provides optimized containerization for macOS:

1. Download OrbStack:
   • Visit OrbStack’s official site, download the latest .dmg file.
   • Double-click the .dmg and follow installation prompts.
2. Launch OrbStack:
   • Open the OrbStack app. Confirm its menu bar icon appears, indicating the service is active.

---

Step 2: Install Ollama

Ollama simplifies local LLM execution:

1. Download Ollama:
   • Visit Ollama’s official site, click “Download for macOS.”
   • Install via the downloaded package.

---

Step 3: Download the LLM

1. Select a Model:
   • Browse models on Ollama’s Models page.
   • For this demo, use DeepSeek-R1:8B (adjust based on hardware).

2. Pull the Model:
   ollama run deepseek-r1:8b

   • The model (~4.7GB) downloads automatically.
3. Verify Operation:
   • After download, the terminal enters interactive mode. Test with:
     Hello, introduce yourself.

   • Expected response:
     I am DeepSeek-R1, an AI assistant...

   • Exit with /bye or Ctrl+D.

---

Step 4: Install Open-WebUI

1. Pull the Open-WebUI Docker Image:
   docker run -d --network=host -v open-webui:/app/backend/data \  
   -e OLLAMA_BASE_URL=http://127.0.0.1:11434 --name open-webui \  
   --restart always ghcr.io/open-webui/open-webui:main

2. Access Open-WebUI:
   • Open http://localhost:8080 in a browser.
   • Register an administrator account (email required).

---

Step 5: Configure Open-WebUI

1. Add Users:
   • Navigate to Settings → Admin Panel to create internal users.

2. Customize Workspace:
   • Add custom models, knowledge bases (via .txt, .pdf), and prompts.
   • Knowledge bases enable domain-specific model fine-tuning.

---

Step 6: Optimization & Maintenance

1. Monitor Resources:
   • Use OrbStack’s dashboard to track CPU, memory, and disk usage.
   • Customize model storage:
     export OLLAMA_MODELS=/custom/path

2. Update Models:
   ollama pull deepseek-r1:8b

3. Backup Data:
   • Regularly back up the Open-WebUI volume (open-webui) via OrbStack’s volume management.

---

Conclusion

This guide enables secure, efficient on-premises LLM deployment on a Mac Mini M4 using Ollama, OrbStack, DeepSeek-R1, and Open-WebUI. The solution ensures data privacy, reduces long-term costs, and supports flexible customization for enterprise workflows. Adapt and scale this framework to meet evolving business needs.