Palace map — rooms of the co-intelligence system

Build your palace.

01

Create one room

A room is a named context space. Create a folder. Give it a purpose. Put a CLAUDE.md file inside it — this is the room's memory file, the thing your AI reads at the start of every session in that room.

The file can be two sentences. It just needs to answer: what is this room for, and what should never be forgotten here?

# Dev Room
This room holds the main product context.
Stack: Next.js 14 / PostgreSQL / Railway.

## Crystals
- yarn dev  ← the start command
- Never push to main without a PR
02

Plant one crystal

A crystal is a fact established once, never re-derived. The thing you keep having to explain to your AI at the start of every session. Write it down in your room's CLAUDE.md. That's it — it's now a crystal.

Good crystals are atomic, specific, and stable. "We use pnpm, not npm" is a crystal. "The project is complex" is not.

## Crystals
- Auth token lives in .env.local, not committed
- Component library: shadcn/ui — do not use MUI
- The staging env is at staging.acme.dev (separate API)
03

Seed one garden plant

The garden is where ideas grow across sessions. A garden plant is a question you're growing toward — not a task, not a feature, but a genuine open inquiry that deepens over time.

Create a garden/ folder alongside your rooms. Add a file for the plant. Name it.

# Plant: The Right Abstraction
**Planted:** 2026-04-28
**Question:** At what point does an abstraction
become the wrong one? Is it when it starts leaking,
or when you stop being able to see through it?

## Notes
- Session 1: leaky abstractions are honest —
  they tell you where the model breaks
- Session 3: the real cost is when the leak
  becomes load-bearing

The soul layer

The onboarding script asks you ten questions — your work, your values, a garden seed, how you like to think. Partway through, the AI pauses and offers you five or six names for itself. Not generic ones. Names shaped by what it just heard: the texture of your work, a value you named, the question you're growing toward. You pick one. It immediately becomes that.

soul/SOUL.md is written in that same session: character, working principles, what the AI has already learned about you. The palace accumulates facts. The soul accumulates perspective. Both start on day one.

04

Run your first session

Open a new conversation with your AI. Load the context. The trigger phrase can be anything — the point is that your AI reads the palace files before it speaks.

In Claude: start a session in the folder that contains your palace. The CLAUDE.md is read automatically. In other systems: paste the room context at the top of the conversation.

At the end of every session: update the room. Add crystals. Water the garden. The palace only works if it grows.

## Session close ritual
1. Did anything become ground truth today? → Crystal
2. Did any question deepen? → Water the plant
3. Did the room context change? → Update CLAUDE.md
4. What does the next session need to know? → Handover
Get the scaffold → See it in practice About Loci

The GitHub repo has room templates, protocol files, and the full methodology. This page is the minimum — the repo is the depth.