Not a course about artificial intelligence. A course where you build one — in an afternoon, for nothing, with no account to make and nothing to install. You click through five steps in a program, and nine short films walk beside you.
Create Your AI, and Own It.
Free · no sign-up · no programming · runs on an ordinary Mac · nothing leaves your machine
What you will actually have
Every figure on this page was measured on the model this course teaches you to build — not estimated, not borrowed from a paper. You will produce the same numbers, and you should check them.
Why we are doing this
In 1811 the weavers of Nottingham did the arithmetic and it said their trade was finished. It was. Their craft died, the cloth got cheaper, the country got richer — and not one of them was ever made whole.
Their mistake was reaching for the only lever they had, which happened to be a hammer. The larger mistake belonged to everybody else.
Same human spirit. Different tools.
We would rather not repeat it. Retrain people before the job goes. Share what the machine makes. Protect the person rather than the job title. Those take governments and employers — but there is a fourth answer that needs nobody's permission, and it is this course.
The course · nine films · 13 minutes
This is the way in, and it is the one almost everybody should take. Get the text. Turn its characters into numbers. Build the model. Train it, and watch the number fall in front of you. Then let it write. Nine short films walk you through the program a step at a time, with the camera close enough to read every number it produces.
You do not need to program, and there is nothing to install. You will still finish holding a real language model that you made.
Lesson 1
What You Are Going To Build
Five steps, and at the end of them a language model that is yours.
Lesson 2
A Million Characters Of Shakespeare
A model cannot be taught English. It can only be shown a great deal of it.
Lesson 3
What The Text Is Made Of
The commonest character in English is the space, and that is the first thing any model learns.
Lesson 4
Characters Into Numbers
Sort the characters, number them in order, and nothing is lost.
Lesson 5
Keeping Ourselves Honest
A score on text the model has already seen is not a score, it is a memory test.
Lesson 6
What The Model Is Made Of
Training changes the numbers and nothing else — not the code, not the shape.
Lesson 7
Attention, Without The Mystery
Attention is one question asked three ways, and the answer is a weighted average.
Lesson 8
Watching It Learn
Loss is one number: how surprised the model was by the character that actually came next.
Lesson 9
Letting It Write
It never writes a word — only the next character, six hundred times over.
Genesis for Geeks · eleven films · 28 minutes
If you would rather see the machinery, take the original course: the whole thing built from an empty folder at the command line, in one file of Python you type yourself. Four commands to set the machine up, a megabyte of Shakespeare, a tokeniser, attention, four blocks, three minutes of a number falling, and a page of verse your own laptop wrote.
Nothing here is harder than the program — it is the same five steps with the lid off, and every command is on a page you can copy. The two courses finish in exactly the same place.
Start here
Create Your First Large Language Model
You are going to build a mind that writes Shakespeare, and it will fit in one file you typed yourself.
Lesson 1
Setting Up Your Machine
Four commands, and your machine is ready. Nothing here is difficult; it is only unfamiliar.
Lesson 2
A Million Characters of Shakespeare
The model will never know anything that is not in this file. So open the file.
Lesson 3
Characters Into Numbers
A model cannot read. It can only do arithmetic. So first we turn every character into a number.
Lesson 4
Attention, Without the Mystery
Attention is just this: every position looks back at what came before, and decides how much each of them matters.
Lesson 5
Four Blocks Deep
There is no clever part. It is one small block, repeated four times, and that is the whole model.
Lesson 6
Watching It Learn
One number falls for three minutes. When it stops, nonsense has turned into something that reads like a play.
Lesson 7
Where To Go From Here
You have already built it. Everything from here is the same idea, made larger.
Lesson 8
Seeing Is Believing
Every part of it can be drawn, and none of the drawings are metaphors.
Lesson 9
Talking To It
Chat is not a conversation. It is one document that stops mid-sentence, and the model finishes it.
Lesson 10
Create Your AI, and Own It
An intelligence you built is one you can answer for.
The application
Five steps down the side of one window: get the text, turn it into numbers, build the model, train it, let it write. Three Python services doing the work and a small Rust gateway holding them together — the same architecture the lessons teach, running on your own graphics chip rather than somebody's server.
It is a prototype the Webspinner Foundation expects to release later this year. The nine Deep Dive films are recordings of it working.
Who is behind this
Project Genesis is published by the Webspinner Foundation, a 501(c)(3) not-for-profit serving the Silicon Forest from Portland, Oregon. The Foundation exists to democratise artificial intelligence — on the view that the surest way to see the technology used well is for ordinary people to own theirs outright rather than rent it.
We are funded by our community. Not by advertising, not by selling what you do here, and not by a subscription that turns this course into a hostage. There is no account to make on this site, because we do not want your details.
Every film, every command sheet and the application itself, at no charge. There is no upgrade tier behind any of it.
The corpus is fetched once and kept. Training runs on your own hardware. We could not read your work if we wanted to.
Supported by patrons and by the Foundation's own work, so the teaching answers to the people learning from it.
How the software is licensed
Open source solved freedom. Four decades of it gave the world a commons that nearly everything now stands on, this course included — Python, PyTorch, the corpus, the tooling. We are standing on that work and we are grateful for it.
What open source never solved is sustainability: it protects your freedoms and explicitly permits charging for the software, which left maintainers holding the bill. nosource is the Webspinner movement's answer. Software is released free of charge — zero licence fee, ever — under the nosource agreement, and the paid value moves to people: registered providers whose hourly rates are published before you engage them.
To be precise about it, because the distinction matters: the nosource agreement is a novel licence, currently in draft. It is not an OSI-approved open-source licence and we do not describe it as open source. Open source permits charging for the software; nosource fixes the price of the software at zero and pays the humans instead.
The movement
Every description below is the property's own, read from its live page on 20 September 2026. Each placard is a screenshot taken the same day — not artwork, not a mock-up. If a page looks unfinished in the picture, that is because it is.
In the open
Webspinner is in prelaunch. Every property below is marked MVP in our own register — not one of them is a proven production release, and several are a good idea with a real page in front of it. We know the difference. We would rather tell you than let you find out.
We are doing it the way the open source movement did it: build in the open, show the work at whatever stage it has actually reached, and let people judge for themselves. A movement does not bootstrap from a press release. It bootstraps from something you can click on, poke at, and disagree with.
What you can rely on today is this course. It is finished, it is free, and every number it puts on screen was measured rather than asserted. The rest is on its way — and you are welcome to watch it arrive, and to tell us when we get it wrong.
Who stands behind this course, and under what terms.
webspinner.foundationMVP
The Webspinner Foundation
Serving the Silicon Forest
A 501(c)(3) cultivating technology, community and a sustainable future in Portland — and the governance that keeps the rest of this list honest.
nosource.aiMVP
nosource
The code is free. The service is fair.
Software released free of charge — zero licence fee, ever — under the nosource agreement, with the paid value moved to people whose rates are published before you engage them.
webspinner.comMVP
Webspinner
AI-native products for the smarter web
One company, eighteen properties. Sealed units of compute, studios that turn a sentence into finished work, and a library that cites its sources.
The sovereign substrate underneath everything on this page.
nephonous.comMVP
NephoNous
Cloud Mind — a sovereign AI solution
A federation of individually complete nodes on owned Apple silicon, each retaining nothing once a turn ends. A node that never persists patron data cannot disclose it.
webspinner.aiMVP
Spinners
The sealed, signed unit of AI-native compute
What containers did for workloads, Spinners do for the work itself: content-addressable digests, publisher signatures, and audit by default.
agathogen.comMVP
Agathogen
Agents for a brighter tomorrow
The structural inverse of a pathogen — a thing whose spreading produces good. A pathogen propagates because that serves the pathogen; an agathogen propagates because that serves the host.
The same architecture you will build in this course, grown up and put to work.
inferencevideo.aiMVP
InferenceVideo
Make powerful videos with a sentence
Describe a video and it writes the script, generates the imagery, narrates it and renders it. Every film on this page was made with it.
inferenceoffice.comMVP
InferenceOffice
Your intelligence, our inference
AI-native applications where you talk and the finished work product is revealed. Blackboard is the first one, doing exactly that.
inferencedesktop.comMVP
Inference Desktop
A more human way to compute
A vision for a free, AI-native desktop: create in natural language, bring intelligence closer to home, and build a computing canvas that grows with you.
inferenceestimator.comMVP
Inference Estimator
Countertop estimates from a photograph
Photograph a kitchen and get a fabricator-grade estimate: square feet, finished-edge linear feet, and every sink and cooktop cutout, priced.
Free, and given away rather than sold.
webspinner.blogMVP
Webspinner Blog
Writing from inside the machine
Long-form writing on sovereign AI, governance, and what changes when software stops suggesting and starts doing. Everything there says who wrote it.
ainoculate.orgMVP
AInoculate
Immunity, not ignorance
A not-for-profit developing hands-on AI literacy for K-12 schools, on the view that banning generative AI relocates risk rather than removing it.
newsspinners.comMVP
Webspinner News
A daily paper for the smarter web
Headlines threaded from across the open press with paywalls flagged — the day's signal lifted clear of the noise.
The record
The Webspinner Journey is a chronology of every project in the estate on one timeline — swim lanes by what a thing is, colour by how far it got. Assembled from git history, the repository index and a live probe of every address. Measured, not remembered.
Get in touch
We read everything. If you finished the course, we would genuinely like to know — what you trained it on, what it wrote, and where it fell over. If you teach, or you want to help, or you think we have something wrong, that is just as welcome.
You will get a one-click confirmation email, so we know the note really came from you. Nothing else is done with your address.