Story 2: How do I control this?

Once the chatbot was running, I kept coming back to two words: security and control. I’m not a security expert, so the real question was simple. How do I control this thing?

I spent a lot of time checking my network and working out how to let this local AI into my day-to-day life safely. Gemini and I talked through running everything in Docker containers, using Tailscale for private remote access, tightening my firewall, and living within the limits of an 8 GB graphics card. We talked about fine-tuning. We talked about connecting it to my home NAS, where the family documents and photos live. Then we got into RAG, short for retrieval-augmented generation, which lets a model look things up in your own files before it answers.

From there I went deep, not just into AI but into local AI. Which models were coming out? How could I use them? How would I fit it all together? I was hooked on how far it could go.

This path didn’t exist a few years ago. Now a regular person can sit at home, put information to work, and build things they always wanted to make but never had the technical skills for. If you’re reading this, you probably know that rush: with AI’s help, I can actually build things.

That first year I went from Ollama to AnythingLLM to llama.cpp. I discovered Docker and soon had this container and that container, and I just kept building. I made operator dashboards, watcher dashboards, and log dashboards. For a non-programmer, it was a blast.

I wasn’t worried about breaking the server. I always figured I’d end up wiping Windows and starting over. I’ve built my own computers for years, so starting from scratch was something I’d done fifty times.

Not to date myself, but my first computer was a Timex Sinclair 1000. It’s still in a closet. The jump from that to my Core i9 and RTX 4070 Ti still blows my mind.

Some of the jank was on purpose. I could always start over.