Build something.
On your own machine.
The lessons explain what local AI is. These are the things you can make with it — working software, built step by step, that runs entirely on your computer. Each one also carries the Claude Code and Codex instructions that will scaffold it for you, because building with an agent against your own model is the part worth learning twice.
- 4 projects, 23 steps in all
- Roughly 6 hours of building
- Free to read, no account needed
Before you start
The workflow comes first
Twenty minutes on where an agent fits, what to tell it, and how to read what it hands back. Every project after this one uses it.
Building with an agent, against your own model
How to use Claude Code or Codex to write the app, while the app itself runs on the model on your desk.
The projects
Four things to build
In the order they get harder. Each one ends in something that runs on your own machine, and each one carries the instructions that get an agent to scaffold it for you.
Step 1
Start here
your first working thing
Step 2
Once you are comfortable
more moving parts
-
A classifier that sorts a folder full of text
Sort hundreds of messages into your own categories overnight, on your own machine, for nothing.
- Python
- openai
- CSV
- Pip text server
-
Ask questions of your own documents
Point it at a folder of your own files and ask questions answered only from what is in them.
- Python
- openai
- numpy
- Pip text server
- Pip embedding server
Step 3
Ambitious
something you will keep using
Where you are
Your workbench
Nothing built yet
0 of 4 built
Ticking a step is your own note to yourself. Nothing here looks at your machine.
- Projects built
- 0 of 4
- Steps ticked
- 0 of 23
- Time to build the lot
- about 6 hours
Your progress is kept in this browser. Sign in and it moves to your account, added to whatever is already there. Nothing you have ticked is lost.
Building, not reading
How the projects work
A lesson ends in a check you can pass. A project ends in software that runs on your own machine, against your own model, on files you would never upload anywhere.
- Steps you can check
- 23
- Written for both agents
- Every one
- Needs an account
- No
- Costs to run
- Nothing
- A verify step how you know it worked, before you build on it
- Code you can copy named by the file it belongs in
- Claude Code and Codex the instruction that scaffolds it, for each
- A review checklist what goes wrong in what an agent writes
- Somewhere to take it deliberately unfinished
Every project walks you into at least one thing that genuinely goes wrong with a local model — a context window that will not fit, an embedding that needs a prefix, an answer that should have been a refusal. Those are the moments the course exists for.
Free to read and free to build. You never need an account, and nothing you write here leaves your computer.
Start where you are
The first one is 45 minutes
Read the workflow guide first if you have not built with an agent before. Otherwise go straight to it.
Stuck on a step, or found something on this page that is wrong? Email info@redroostertec.com and a person will answer.