Memory & context

The One-Folder Setup That Makes Claude Remember Your Business

6 minute readUpdated October 2026Explore more

TL;DR

Make one main folder. Put a CLAUDE.md with your rules on top, five short context files under it (business, offer, customers, voice, goals), a memory folder for decisions and lessons, and one folder per project. Open Claude Code inside that folder and the rules plus all five context files load at the start of every session. Everything below is copy-paste: the folder map, a prompt that interviews you and builds it, and every template.

This is the kit from the reel. You get the folder map, one prompt that interviews you and builds the whole thing, and a template for every file. Copy, paste, fill in the blanks, and Claude stops asking who you are.

Copy the folder map

textmy-business/
├── CLAUDE.md              rules on top, loads every session
├── context/
│   ├── business.md        what the business is
│   ├── offer.md           what you sell and for how much
│   ├── customers.md       who buys and why
│   ├── voice.md           how you sound
│   └── goals.md           what you are pushing toward now
├── memory/
│   ├── decisions.md       what was decided and why
│   └── lessons.md         what worked, what flopped
└── projects/
    ├── spring-launch/
    │   └── CLAUDE.md      context for this project only
    └── newsletter/
        └── CLAUDE.md

Why this works: Claude Code reads the CLAUDE.md in the folder you launch it from at the start of every session. That CLAUDE.md pulls in the five context files with @ imports, which also load at launch. A CLAUDE.md inside a project folder loads when Claude starts working with files in that folder, so project details only show up when they matter.

Paste one prompt and let Claude build it

Make an empty folder called my-business, open Claude Code inside it, and paste this. It asks you questions one at a time, then writes every file for you.

textI want one main folder that holds my whole business so you remember it every session.

Interview me first. Ask one question at a time, in this order:
1. The business: what it is, who runs it, how it makes money.
2. The offer: every product or service, price, and what the buyer gets.
3. The customers: who buys, their biggest problem, the words they use.
4. The voice: how I write and talk, words I use, words I never use.
5. The goals: the 1 to 3 things I am pushing toward this quarter, with numbers.
6. My active projects: name, goal, deadline, status.

Then build this exact structure in the current folder:

CLAUDE.md
context/business.md
context/offer.md
context/customers.md
context/voice.md
context/goals.md
memory/decisions.md
memory/lessons.md
projects/<one folder per active project>/CLAUDE.md

Rules while you build:
- Use only what I tell you. If you do not know something, write TODO, never guess.
- Keep every context file under 60 lines. Bullets, not paragraphs.
- Keep the top CLAUDE.md under 200 lines.
- If any of these files already exist, show me the change and ask before you overwrite.
- When you are done, print the folder tree and list every TODO you left.

Put your rules on top: CLAUDE.md

Save this as CLAUDE.md in the main folder. The lines that start with @ pull the five context files in automatically. Paths are relative to the CLAUDE.md file, so keep the context folder right next to it.

markdown# [Business name] - how to work here

Read before every task. These files describe the business:

- Business: @context/business.md
- Offer: @context/offer.md
- Customers: @context/customers.md
- Voice: @context/voice.md
- Goals: @context/goals.md

## Rules
- Write in the voice from context/voice.md. Never invent prices, results, or testimonials.
- If a fact is missing from these files, ask me. Do not guess.
- Every recommendation should move a goal in context/goals.md. Say which one.
- Project work lives in projects/<name>/. Read that folder's CLAUDE.md first.

## Memory
- Before you recommend anything, check memory/decisions.md and memory/lessons.md
  so you do not suggest something we already rejected.
- When we make a decision, append one line to memory/decisions.md:
  YYYY-MM-DD | decision | why
- When something works or flops, append one line to memory/lessons.md:
  YYYY-MM-DD | what happened | what we do next time
- Never delete memory lines. Mark old ones (superseded) instead.

## Where things go
- Final deliverables: projects/<name>/
- Do not create new top-level folders without asking.

Fill in the five context files

Short beats complete. Bullets only, real facts only, and write TODO where you do not know yet.

context/business.md

markdown# Business

- Name:
- What we do (one sentence):
- Who runs it (names and roles):
- How we make money:
- Where customers find us:
- Tools we use daily:
- Things that are true here that an outsider would get wrong:

context/offer.md

markdown# Offer

## [Product or service name]
- Price:
- What the buyer gets:
- Who it is for:
- Who it is NOT for:
- Main promise (use these exact words):
- Guarantee or refund terms:

(Copy the block above for each offer.)

context/customers.md

markdown# Customers

- Who buys (role, situation, not demographics only):
- The problem they wake up with:
- What they tried before us:
- Why they finally buy:
- Why they hesitate:
- Exact phrases they use (copy from emails, DMs, reviews):
  - "
  - "
- Best customer we ever had, and why:

context/voice.md

markdown# Voice

- Three words for how we sound:
- We say:
- We never say:
- Sentence length: short / mixed / long
- Emoji: yes / no / only in:
- Formatting rules (for example: no em dashes, no hype words):
- One paragraph we wrote that sounds exactly right:

> paste it here

context/goals.md

markdown# Goals (update every quarter)

Quarter:

1. Goal:            Number:         By:
2. Goal:            Number:         By:
3. Goal:            Number:         By:

## Not doing this quarter
- (Things we decided to ignore, so Claude stops suggesting them.)

Add the memory folder

Two append-only logs. The rules in CLAUDE.md tell Claude to check them before it recommends anything and to add a line whenever you decide something or learn something.

memory/decisions.md

text# Decisions log
# Format: YYYY-MM-DD | decision | why
# Newest at the bottom. Never delete lines; add (superseded) instead.

memory/lessons.md

text# Lessons log
# Format: YYYY-MM-DD | what happened | what we do next time

Add one folder per project

Inside projects/, make a folder for each thing you are actively working on and give it this CLAUDE.md. Reference folders (old campaigns, assets, exports) do not need one. Claude can still open those files when you ask.

markdown# Project: [name]

- Goal (ties to which goal in context/goals.md):
- Deadline:
- Status: not started / in progress / blocked / done
- Owner:
- Key decisions so far (also logged in memory/decisions.md):
- Files in this folder:
- Next step:

Open the folder and check it loaded

bashcd ~/my-business
claude

Type /context and look under Memory files: your CLAUDE.md should be listed there. Then run this test in a fresh session to confirm the five context files came in with it:

textWithout reading any new files, tell me: what we sell, who buys it,
how we sound, and our top goal this quarter. Then list anything you
feel unsure about.

If it answers correctly without you explaining anything, it works. If it gets something wrong, fix the context file, not the chat.

Keep it fresh with one end-of-session prompt

textBefore we stop: review what we did this session.
1. Append any decisions to memory/decisions.md (date | decision | why).
2. Append anything that worked or flopped to memory/lessons.md.
3. Update the Status and Next step in this project's CLAUDE.md.
4. If any context file is now wrong (price, goal, offer), show me the
   edit and wait for my yes before changing it.

Common questions

  • Do I need Claude Code for this, or does the regular Claude chat work?

    The automatic loading described here is how Claude Code works: it reads the CLAUDE.md in the folder you open it in, plus anything that file imports with @. In a regular chat you would have to paste or upload the files each time.

  • Will this use up my context window?

    A little. The top CLAUDE.md and every file it imports load at the start of every session. That is why the kit keeps CLAUDE.md under 200 lines and each context file short. Project CLAUDE.md files only load when Claude works with files in that project folder.

  • How is the memory folder different from Claude Code's auto memory?

    Auto memory is Claude Code's own notes, stored in ~/.claude/projects/<project>/memory/ on your machine. The memory folder in this kit is two plain log files inside your business folder that you control, so they travel with the folder.

  • How do I check that Claude actually loaded my files?

    Run /context in a session and look under Memory files. Then ask the test prompt on this page in a fresh session. If it can name your offer, customer, voice, and top goal without help, the setup is working.

  • What if Claude still ignores one of my rules?

    Instructions in CLAUDE.md are context, not hard locks. Make the rule concrete and checkable, remove anything that contradicts it, and trim the file. A long file is the most common reason a rule gets lost.

Want help filling in your five context files?

Get the other 3 in the memory stack, plus 8,000+ members - $9/mo, cancel anytime.

Join the Club