What "Local AI" Actually Means (It's More Like Owning a Car Than Calling a Taxi)
Running AI on your own computer sounds either trivial or impossible depending on who you ask. It's neither. Here's the plain-English version of what local AI is, why a gaming PC is often enough, and what you're not signing up for.
In the last post I made a promise: there’s a whole world of capable AI you can run on your own computer, for free, and it’s worth learning before the cheap frontier subscriptions get expensive.
Now let me explain what that actually means, because “running AI on your own computer” sounds either trivial or impossible depending on who you ask, and it’s neither.
Taxis and cars
When you use ChatGPT, Claude, or Gemini, you’re calling a taxi.
You open an app, you say where you want to go, and a vehicle you don’t own, maintained by someone else, driven by someone else, takes you there. It’s wonderfully convenient. You don’t think about fuel or insurance or oil changes. You just pay per ride, or a monthly pass, and someone else handles everything.
The catch is the one every taxi rider knows. You don’t control the car. You don’t set the price. If the company raises fares, changes the rules, or decides your neighbourhood isn’t worth serving, that’s their call, not yours. And every trip you take, they know about.
Local AI is owning a car.
There’s more upfront effort. You have to get the thing, learn to drive it, deal with the occasional maintenance. But once it’s in your driveway, it’s yours. No fare meter. No company watching where you go. You drive it as much as you like for the cost of fuel, and nobody can reprice your daily commute out from under you.
Neither option is “better.” I own a car and I still take taxis. The point is that having both is a much stronger position than depending entirely on the taxi company’s goodwill.
So where does the AI live?
When people say a model runs “locally,” they mean the actual AI (the enormous file full of everything it learned) sits on your own computer’s hard drive and runs on your own hardware. Nothing leaves your machine. You ask it a question, your computer does the thinking, and the answer appears. No internet required once it’s set up.
Compare that to the taxi version, where your question travels over the internet to a giant warehouse full of computers owned by a tech company, gets answered there, and travels back. That warehouse is why the frontier tools are so capable, and also why they cost what they cost to run.
The trade is real. The model in the warehouse is bigger and smarter than anything you can fit at home. But the model at home is private, free to run, and entirely under your control. For a surprising amount of everyday work, the home model is more than good enough.
”But I don’t have a supercomputer”
You probably don’t need one. This is the part that genuinely surprised me.
The hardware that runs AI well is the same hardware that runs video games well. It’s called a graphics card, or GPU, and it’s the expensive chip that gamers obsess over. If you or someone in your house has a gaming PC bought in the last few years, you may already own a perfectly capable AI machine and not know it.
I’ll go deeper on the hardware question in the next post, because it’s the thing that decides which models you can actually run. For now, just hold onto this: a normal gaming computer is often enough. You are not necessarily looking at a big purchase.
The “app store” that makes this possible
A few years ago, running a model at home meant typing cryptic commands into a black screen and praying. It was genuinely a programmer’s task. That’s the version of this story most people still have in their heads, and it’s out of date.
Today there’s a free program called LM Studio that I can only describe as an app store for AI models. You install it like any normal application. It shows you a searchable library of available models with friendly descriptions. You click to download the one you want, the way you’d install an app on your phone. Then you chat with it in a window that looks a lot like the AI tools you already use.
That’s the whole shift. The thing that used to require real technical skill now looks like browsing an app store and clicking download. Is there still some fiddling involved? Yes, and I’ll be honest about it when we get there. But the gap between “curious non-technical person” and “running my own AI” has collapsed in a way that most people haven’t noticed yet.
What you’re not signing up for
Let me head off the two biggest worries.
You are not becoming a programmer. Nothing in this series requires you to write code. If you can install an application and follow a recipe, you can do this.
And you are not committing to abandoning the tools you love. I run a local model and I still pay for the frontier ones. This is about adding an option, not replacing your whole setup. The goal is to understand the alternative well enough that you’re choosing the taxi rather than depending on it.
Where we’re headed next
Next post is the one that gives this whole project its purpose. A private chatbot on your own machine is nice, but on its own it’s just a slightly more private version of what you already have. The real prize is turning that model into an assistant with a job: one that remembers you, can act on your behalf, and answers to nobody else. That’s where we’re going next, before we get our hands dirty with the build.