How it works
Your AI foundation, layer by layer.
Claude is brilliant and starts every conversation knowing nothing about you. Crads-AI is the foundation underneath: four layers that onboard Claude to you and your business, built through the interview and daily use, and a method for solving real problems on top of them.
What it is made of
A chatbot forgets you. An assistant accumulates you.
The difference is a folder of plain files it reads every time it wakes up. That folder is what the app builds with you: four layers, each one making the assistant more yours. This page is about how yours gets built; the docs walk every part of a finished one.
C1
Context
What it knows about you.
Plain files, organised in layers. Mine has eight:
Written down once, never explained twice.
C2
Connections
What it can reach.
Context tells it who you are. Connections let it work where the work already lives:
Each one is a doorway you open once.
C3
Capabilities
Blueprints for your workflows.
A capability is a written recipe: the steps you'd give a new hire, saved so it runs the same way every time:
Describe the workflow once. Now it's a button.
C4
Cadence
The heartbeat.
Jobs that run on a schedule without being asked: the morning brief is waiting before you sit down.
You've hired an executive assistant. Give it feedback, and let go of the reins a little more each week.
The app climbs this ladder with you: Context first (the interview), then the doorways, then the blueprints, then the heartbeat.
Where it lives
Three places, all yours. Nothing sits with Crads-AI, because there is nothing central to sit with.
- The app, on your laptop. Windows or Mac. It builds the server, then it is how you open and steer your assistant every day.
- The mineral, on your server. A small machine in your own hosting account, built by the app with your own API token, holding only your computer's key. This is where the assistant runs and where its scheduled jobs happen while you sleep.
- The brain, on your GitHub. Everything it knows about you, as plain files in a private repository you own, pushed there every night. Lose the server and you lose nothing.
What you have once it is running
The app on your laptop, your assistant (we call it a mineral) on a server you own, and a brain repo on your GitHub that holds everything it knows about you. Your tools connected, a couple of workflows you actually use, and CRIT, the method for solving the next problem yourself. It is not "finished" on day one; roughly a third of it is built, and the rest compounds through daily use over the months that follow. The docs walk each part.
Where I come in
Free to run. A hand laying the foundation, if you want one.
Most people get there with the first hour in the docs. If you'd rather lay the foundation with someone who has set these up before, a walkthrough is an hour then 30 minutes, and a guided setup is an hour then two hours.