Local AI

AI doesn't have to live in someone else's cloud. You can have it on your own computer, with no subscription and without your data leaving home. I work with four platforms (Ollama, LM Studio, llama.cpp and Unsloth) and I've tested more than 50 models. In this section I share that experience from the very beginning, with what worked and with the money and hours I lost on what didn't.

It isn't just about models. Everything you need to make AI work on your own machines belongs here: the computer and the graphics card (GPU), the models, the programs that run them and what surrounds them so they're actually useful. I explain why the card's memory matters more than anything else, what quantization is and why it lets you run a big model on a modest machine, and what model sizes mean, without unnecessary jargon. The servers themselves, which machine does what and the VPS, I cover in VPS vs in-house servers.

And the underlying question: what is it actually good for? There are tasks where local AI replaces the paid kind and others where it falls short, and it's worth knowing that before spending money on a GPU. I'll tell you what I run locally, what I still do in the cloud, and who it pays off for to set up their own AI at home, and who it doesn't.