How I Built an AI Assistant on a Mac Mini M4

The Setup

I run a Mac Mini M4 Pro as a 24/7 AI workstation. It sits on my desk, never sleeps, and runs an AI agent called Caper that handles everything from content creation to code generation to business automation.

Here is exactly how I set it up, what it costs, and what it actually does.

Hardware

• 
  Mac Mini M4 Pro — 24GB RAM, 512GB SSD ($599 base, upgraded RAM)


• 
  External SSD — 2TB for storing projects, logs, and media


• 
  Always-on internet — Ethernet, not WiFi, for reliability

Total hardware cost: ~$800. Runs 24/7 at about $3/month in electricity.

The Stack

Local LLMs

I run Ollama with several models pulled locally:

• 
  llama3.2 for fast general tasks


• 
  codellama for code generation


• 
  mistral for creative writing

These run entirely on-device using the M4 Neural Engine. No API costs for local inference.

Cloud APIs (When Needed)

For heavy lifting, Caper calls:

• 
  Claude API — Complex reasoning, long documents, code review


• 
  OpenAI Whisper — Audio transcription


• Cost: ~$5-15/month depending on usage

Python Automation Layer

The glue holding everything together:

• 
  Cron jobs for scheduled tasks (content publishing, monitoring, backups)


• 
  Playwright for browser automation when APIs are not available


• 
  FFmpeg for video/audio processing


• 
  yt-dlp for media downloads and analysis

What It Actually Does

Content Creation

Caper writes articles, generates social media posts, and creates digital products. It handles the first draft, I review and publish.

Business Automation

• Monitors competitor pricing


• Tracks analytics across platforms


• Generates reports on content performance


• Manages a task queue with priorities

Code Generation

When I need a script, I describe what I want. Caper writes it, tests it, and saves it to the scripts directory. If it fails, it debugs itself.

Research

Give it a topic, it searches the web, reads documentation, cross-references sources, and delivers a summary with citations.

The Economics

Compare that to hiring a virtual assistant ($500-2000/month) or subscribing to multiple SaaS tools ($100+/month). The Mac Mini pays for itself in the first month.

Lessons Learned

1. 
   Local models are good enough for 80% of tasks. Save cloud API calls for the complex stuff.


2. 
   Cron is underrated. Scheduled automation is more reliable than event-driven for most business tasks.


3. 
   Always log everything. Every API call, every token spent, every task completed. You cannot optimize what you do not measure.


4. 
   Start with one workflow, then expand. I started with content generation, then added research, then automation, then monitoring.

Get Started

You do not need a Mac Mini. Any always-on computer works. The key ingredients:

1. A machine that stays on 24/7


2. Python + cron for automation


3. Ollama for local LLMs


4. One or two cloud APIs for heavy tasks

Total setup time: one weekend. ROI: immediate.

I am building this in public. Follow for updates on what an AI workstation can actually do for a solo operator.