Learn how to run AI locally.
Then plan with it.
You have heard that you can run AI on your own computer. This course tells you what that means, what it costs you, and when to use it instead of the cloud. It is written for somebody who is not technical and has fifteen minutes at a time.
- 25 lessons in 5 tracks
- About 466 minutes of reading in all
- Free to read, no account needed
Where you are
Your progress
Curious
0 points
500 more to reach Getting it.
- Lessons finished
- 0 of 25
- Checks passed
- 0
- Tracks finished
- 0 of 5
Your progress is kept in this browser. Sign in and it moves to your account, added to whatever is already there. Nothing you have read is lost.
The course
Five tracks, in order
Each track builds on the one before it. You can read them in any order you like, and the first track is the one that makes the rest make sense.
Track 1
Foundations
what this actually is
- What "local AI" means Local AI runs on the computer in front of you, so what you type never has to leave it.
- What a model actually is A model is a file of learned numbers. It guesses the next piece of text, over and over.
- Why size is the whole story A bigger model knows more, needs more room and runs slower. Choosing where to sit on that line is the job.
- What your computer needs Memory sets the ceiling on what will run. Storage is the shelf. The chip decides how fast the words appear.
- What "private" really buys you Local AI gives you a few real, checkable guarantees. It does not give you all of them, and the gaps are worth knowing.
Track 2
Options
what is out there
- The families of models Most models you will meet come from a short list of makers. Knowing the families beats knowing this month's winner.
- Open weights is not open source Being able to download a model tells you nothing about what you may do with it. Only the licence tells you that.
- Quantization, in plain words Quantization rounds a model's numbers so the file fits your computer. The cost is real, and at the usual settings it is small.
- The tools people use Four ways to run a model on your own machine, what each one is really for, and what each asks of you in return.
- How to read a model card A model card is the page beside a model file. Five facts on it decide whether the download is worth starting.
Track 3
Practice
using it well
- Choosing your first model Pick the model your machine can hold with room to spare, then judge it on your own work rather than on a leaderboard.
- Asking a local model well A small model acts on clear instructions, not forceful ones. Say what you want back, show its shape, and hand over the facts.
- Context: the model's short memory The context window is everything a model can see at once. Long documents strain it, and a big number on the card does not settle it.
- When it is slow or wrong Four things explain almost every bad session, and there is a cheaper check to make before you start on them.
- Keeping it tidy Model files pile up, two separate things need updating, and you should always know which model answered you.
Track 4
Strategy
local and cloud together
- What each one is genuinely good at An honest list in both directions, including the four things cloud AI does better than any machine on your desk.
- A decision rule you can actually use Four questions in a fixed order, and the first hard answer decides. Sometimes the answer it gives you is the cloud.
- Designing a two-lane system Give both lanes real work, write down where each thing is allowed to go, and decide now what happens when a lane goes dark.
- Cost, risk and control The cloud sells you power and reach for a bill that keeps coming. Local sells you certainty for money paid up front.
- Writing your own plan Answer seven questions about one real piece of your own work, and finish holding a one-page plan.
Track 5
Performance
making it fast, and knowing that it is
- The four numbers that matter Tokens a second, the wait before the first word, context and memory. What each one means, which one really decides, and the words around them.
- Where the time actually goes Reading your prompt and writing the answer are two different jobs with different limits. That one fact explains most of the waiting.
- The settings that move the needle Five settings are worth knowing about, and most people should still change none of them. Each one is named and priced here.
- Measuring it honestly How to compare two runs without fooling yourself, and why a speed posted on a forum tells you almost nothing.
- When one machine is not enough Several people, several requests, and the three real options when one machine runs out. Sometimes the answer is to stop.
Points and levels
How the points work
Points mark progress. They are not a score, nothing is ever taken away, and a check can be retried as often as you like.
- Each section you read
- 10
- A closing check, all right
- 40
- Finishing a lesson
- 50
- Finishing a track
- 150
- Curious 0 points You are here
- Getting it 500 points
- Comfortable 1400 points
- Planning with it 2500 points
- Ready to build 3800 points
The top of the ladder means intermediate. It does not mean expert, and this course will not tell you that it does.
Everything written so far is worth 4890 points in all.
Start where you are
The first lesson is fifteen minutes
Read one a day and you are through the course inside a month.
Something on this page not working, or a lesson that looks wrong? Email info@redroostertec.com and a person will answer.