Why the RAPP Brainstem?
It is a local frontier learning and rapid prototyping tool. It is where you learn the agent pattern and prove a prototype on your own machine, before the result graduates into the Microsoft tools that carry it to real users.
Looking for something to show a customer? Browse the solutions or get one-pagers and demo videos.
What it is
A private practice space where you describe real work in ordinary language and watch GitHub Copilot turn it into a working AI skill. You see the files appear, watch the skill run, review the result, and ask Copilot to improve anything that does not work.
There is no cloud environment to prepare before learning begins. The library's industry solutions and Academy workshops give you realistic starting points, synthetic examples, and visible tests so you can learn by doing rather than by sitting through another presentation.
Your AI's companion, so you don't have to be
Your Brainstem is your AI's companion so you don't have to be. It can do the babysitting across any AI system you use: it holds your agents, your memory and your conversation history, and any AI you hand its one plain skill.md can pick them up and use them, so you are never the one keeping each AI up to speed. That is the design for Claude, GitHub Copilot, Codex or whichever assistant you use; we name an assistant as supported once we have shown it working. And it is an AI you have full control and ownership over: it runs on your machine, as files you own, and every skill you build in it is yours to keep.
Your AI stays home. Any AI is just the plug.
Whichever AI you already use, GitHub Copilot, Claude, another assistant, plugs into your Brainstem through one door (/chat) and gets your skills and your memory. Swap the plug any time; nothing is lost, because everything it learned is files on your computer. The AI you plug in stays in charge with the rules it already has. The Brainstem is its reach into your computer: one set of rules, not two.
One plain skill.md, any AI
Hand your AI assistant one plain skill.md and it can use, bring and build agents on your Brainstem, with no change to the Brainstem itself. We name an assistant as supported once we have shown it working; others are expected to work. See the guide.
The hard questions
- Why not just use Copilot or Claude by itself? On its own, the AI forgets what it built. Through the Brainstem, a skill is built once and kept, and the next AI you open picks it up.
- How does it learn something new? You ask in plain words. It writes the new skill as one file in
agents/and uses it on the next message, with no restart. - Does it cost extra? It thinks using your existing GitHub Copilot sign-in, so it uses your Copilot seat. There is no separate Brainstem bill and no key to manage.
- Where is my data? On your computer, as plain files you can open and read. Not in someone else's cloud memory.
- Is it always running? No. It runs only when you start it and goes to sleep when you close it. You decide whether it starts on its own.
- How do I take a skill away? Move its file out of the
agents/folder. The Brainstem stops using it on the next message. Move it back and the skill returns. - What if I switch AIs next year? Nothing is lost. Skills and memory are files. Hand the new AI the same
skill.mdand it picks up the same Brainstem. - Isn't full access to my computer the risk? Full access is the point: the AI you already trust keeps its normal rules. If you want it walled off, run it in a container. It will do less, but it is your call.
What it is not
- Not a replacement for GitHub Copilot. Copilot is the work surface and the model. The Brainstem borrows Copilot's access so there are no keys to manage, and it hands the work back to Copilot the moment you are done learning.
- Not a competitor to Copilot Studio or Microsoft 365 Copilot. Those are where a finished agent lives. The Brainstem is where the agent is first understood, and the library ships the Copilot Studio solution for every workshop precisely so the prototype can leave.
- Not a production runtime. Nothing here is built to serve users. It is built to teach one person at a time, quickly.
- Not another platform to learn. The learner talks to Copilot, watches the work happen, and leaves with a reusable skill. The technical plumbing stays in the background.
The graduation path
| Stage | Where | What you have at the end |
|---|---|---|
| Learn | The Brainstem on your machine | A working single-file agent you understand line by line |
| Prove | Academy workshops and the production guide | Locked test cases passing, evidence captured, a reviewed draft |
| Ship | GitHub Copilot, Copilot Studio, Microsoft 365 Copilot | The same learned workflow, adapted for the place your users already work |
The library calls these the Brainstem, the Hippocampus, and the Nervous System. The names are a mnemonic for the same idea: start with the survival loop, add memory and a cloud body when you need them, then connect to the enterprise nervous system. You advance when you choose to.
How it stays out of the way
- No separate model key. The Brainstem uses your existing GitHub Copilot access instead of asking you to create and manage another model-provider credential.
- Your work is visible. The skill and its local memory are files you can inspect, back up, keep, or delete.
- Skills stay small. One portable agent file is easier to understand, review, share, and improve.
- Copilot does the technical work. The learner can stay in conversation while Copilot handles files, tests, and setup in view.
- You can leave cleanly. The useful learning stays in the skill file rather than being trapped inside the Brainstem.
Who it is for
Sellers, architects, and partners who need to understand an agent pattern well enough to explain it and adapt it, and workshop learners who want to build one before they publish one. If you already know what you want to ship, go straight to the library and take the Copilot Studio solution. If you want to understand it first, install the Brainstem.