I Bought a New Mac Mini to Build My Own AI Assistant — And It Was Way Harder Than It Should Have Been

Setting up a local AI assistant on a Mac Mini with OpenClaw and Ollama took me three days of troubleshooting Homebrew, Node.js, memory configuration, and model compatibility. Here is exactly what went wrong and how I fixed each problem — so you can skip the pain.

I’m not gonna lie: I didn’t just “install” OpenClaw. I earned it. I bought a brand new Mac Mini specifically for this. Fresh machine, fresh start. How hard can it be? Spoiler: I was so wrong.

The “Basic Stuff” That Wasn’t Basic

First hurdle: I don’t have a technical background. When people said “just install Homebrew” or “you need Node.js” — I had no idea what any of that meant. Stack Overflow threads assumed I knew things I didn’t. YouTube tutorials skipped the “obvious” steps. It took me ages just to figure out the foundation. Package managers? Command line tools? What even is a terminal? Everything that seemed “basic” to developers was completely new to me.

The Memory & File Structure Rabbit Hole

Before I even got to errors, I had to learn how OpenClaw thinks. What’s a workspace? How do I set up memory files? What’s the difference between MEMORY.md and daily memory files? I spent hours figuring out:

• How to structure my workspace folders
• How to set up persistent memory so the AI actually remembers things between sessions
• How to organize files so I could find anything later

It felt like learning a whole new language just to have a conversation.

If you’re just getting started with AI tools, check out the rest of the blog for more beginner-friendly guides.

But I Got It Running... Sort Of

I pushed through. Got Homebrew installed. Got Node.js working. Figured out the file structure. Thought I was finally there. Still didn’t work. Why? Because I needed a model that could actually see my terminal and edit files. Not just chat — I needed something that could read my screen, run commands, make changes.

That’s when the real problems started.

Who Do You Even Ask?

So here I am, stuck. And who do I ask for help? I tried asking regular AI chatbots. But they didn’t understand what I was going through — or more importantly, I didn’t understand what they were telling me. They’d say things like “just run this npm command” and I’d be like... where? In what terminal? How do I even open one of those?

And I couldn’t even buy my way out of this. No credit card that works for international subscriptions. No way to pay for the “easy” solutions. I was stuck with whatever I could get for free, whatever was available in Hong Kong, and whatever I could figure out on my own.

The Money Problem

I’m in Hong Kong, and I can’t subscribe to paid services. Claude Code? Not available here. Antigravity? I couldn’t even download it — needed extra steps just to get it on my machine. Most free options just couldn’t do what I needed. They could chat, sure. But looking at my terminal files and actually making things work? That required something more. So I cobbled together what I could:

• Antigravity (my only option that could actually see and edit files)
• Gemini Flash (free, for the parts it could handle)
• Claude Code (eventually got working through sheer stubbornness)

Basically: I had to piece together whatever was free, whatever I could access, whatever wouldn’t block me geographically.

The Errors Kept Coming

Then the OpenClaw errors started:

• “Unrecognized key: heartbeat” — cool, let me remove that
• Config reloaded... and reloaded... and reloaded. Gateway restarted 6 times in the first hour.
• “plugins.allow is empty” — okay, noted? Is this a problem? Do I need to fix this?
• “[tools] message failed: Channel is required” — I had Discord connected. But “connected” and “able to send messages” are apparently different things.

Each error was a detour. Each detour took hours of Googling, reading logs, and trying things that might not work.

But It Finally Works

After all that:

• Discord connected
• Web UI running
• Memory and context working
• Custom tools for file management
• AI running locally, reachable from my phone

The Bottom Line

OpenClaw isn’t a “download and go” tool. It’s a builder’s tool. You need to:

• Read logs (they actually help, eventually)
• Figure out file structure and memory
• Tinker with config
• Understand the pieces
• ...and have zero financial resources and live in the wrong geography

But once it’s running? It’s magic. I can text my AI from Discord. From WhatsApp. From anywhere. And it lives on my machine. My data. My rules.

I write about this journey every week — subscribe to the newsletter to follow along, or learn more about me and why I started ThinkingWithJon.

If you’ve ever wanted your own AI that’s actually yours — not a subscription, not a cloud service, not blocked by where you live — it’s worth the struggle.