Running AI Locally
A model on your own machine, and nothing leaves it.

Every other course in this school raises the same question in its first lesson: what if I cannot paste this anywhere? This is the answer. A model that runs on your own computer, that you can hand your real notes, your real files, and your real questions, because nothing leaves the machine. It is slower and smaller than the hosted tools, and this course is honest about that ceiling. It is also free after the download, private by construction, and enough for a surprising share of the work.

What it is

Six lessons that take you from an empty machine to a working assistant that runs entirely on it. You pick a model that fits your memory, install one tool, download one model, feed it your own documents, and wire it into one job you do every day. Every lesson has real commands and real prompt text you can copy, and ends with one thing to do before the next one.

What it is not

It is not a claim that local beats hosted. It does not, and lesson 1 says exactly where the ceiling is so you are never surprised by it. It is not a tour of every tool — you pick one and use it. And it is not a purity exercise: lesson 5 builds a routing rule that sends the hard, public work to the big hosted models on purpose, because the boundary is about what is in the text, not about principle.

What you need

• A computer you use every day, with 16 GB of memory as the working floor. Lesson 2 does the arithmetic: your operating system, browser, and normal apps take 6 to 8 GB before a model loads, and the context window costs more on top, so 16 GB is what leaves room for the 7B–9B band this course is built around. 8 GB still works, but only for the 3B–4B band, and only with the browser and everything else shut — useful for rewriting, reformatting, and extraction, and not for much else. Lesson 2 shows you how to find your real number and what it will hold.
• Several gigabytes of free disk, and a little patience for a one-time download. You will pull two candidate models and keep one.
• One document you will build as you go. Plain text or markdown, started in lesson 1, added to in every lesson: your memory budget, your model, your routing rule, your daily job. By the end it is the whole setup on one page.
• Five real documents of your own — notes, statements, letters, whatever your work is made of. Real ones, with the real contents, because being able to use them is the entire point.
• A text editor you can save a file without an extension in, and a terminal you are willing to type six commands into — ollama pull, ollama list, ollama run, ollama create, ollama rm, and one source to reload your shell. That is the whole command set the course requires. If a terminal is a wall, lesson 3 names a windowed alternative that does the same job.
• About four hours, splittable — two sittings of decisions, one afternoon of setup, then short sessions.

How to work through it

In order, and with your hands on the keyboard. Lessons 1 and 2 are decisions you make before downloading anything, and they are quick — an hour between them, mostly spent looking at your own week rather than reading. Lesson 3 is the install, and it is the one to give a real afternoon: the download runs while you do other things, but the auditioning and the configuring want your attention. Lessons 4 through 6 are short sessions on your own material, best done on separate days so each one has real work to point at.

Keep three things open the whole way through. Your course document, because every lesson adds a line to it and by lesson 6 it is the setup. A terminal or your model's window, so that a prompt you read is a prompt you run rather than a prompt you mean to try. And your actual work — email, files, notes — because every exercise in this course is done on your own material, and the ones done on invented examples teach you nothing you will keep.

Prompts belong in files, not in your head or your scrollback. From lesson 3 on, keep a folder — ~/prompts/ is the suggestion — with one plain-text file per prompt, beside your Modelfile and your course document. That folder is what you copy to your next machine instead of rebuilding from memory.

The toolkit is the page you keep after the lessons are done: every prompt from the six lessons finished rather than sketched, each with a line on when to use it, alongside the files the course tells you to build. Read the lessons for the argument; keep the toolkit open while you work.