AI Agents · Setup

The Brand Brain

One folder that every AI agent reads before it works. Teach your business once, then use Claude, Codex, Hermes or Grokbot without explaining it again.

The problem: every agent starts at zero

You open a new AI tool and it doesn't know what you sell, who buys it or how you talk. So you explain it again. Then you switch tools next month and explain it again.

Skills and prompts don't fix this. A skill tells the AI how to do a job. It doesn't know your business. That's why the output still sounds like everyone else's.

The fix is one place your business lives, outside any single tool. Every agent reads it first. Change a note once, and every agent knows. We run our whole company this way: 17 Grokbot agents and Claude, all reading the same Obsidian vault.

Step 01

Vault

Download Obsidian (free) and make one vault called your brand brain. It's just a folder of plain text notes, so any AI that can read files can read it. Nothing is locked to one tool. Start with four notes:

brand-brain/
  business.md      what you sell, to whom, and why
  voice.md         how you sound
  rules.md         where things go, what needs your OK
  AGENTS.md        tells every agent to read the three above first

Step 02

Business

One note with the facts every agent gets wrong when it guesses: your offer, your price, your buyer, and what makes you different. Write it like you'd brief a new hire on day one. The template is below.

Then add everything else

Your content, contracts and docs go in too

The four notes are the start. Anything your business already has in a doc can live in the brain, and every agent gets to use it. Drop files into a folder, and the next agent that opens the vault knows them.

content/

Past posts, scripts, newsletters, your best-performing videos

Agent: Copies what already worked instead of guessing

clients/

Contracts, proposals, onboarding docs, call notes

Agent: Drafts follow-ups and scopes on your real terms

sales/

Pricing sheet, offer docs, common objections and your answers

Agent: Answers a prospect the way you would

proof/

Testimonials, case studies, results with their source

Agent: Uses real proof, never an invented number

sops/

How you run recurring jobs, step by step

Agent: Follows your process, not a generic one

research/

Competitor notes, saved posts, market research

Agent: Pitches ideas grounded in your market

Contracts and client files are private. Keep the vault in a private repo, and only give cloud agents the folders they need.

Step 03

Voice

Paste in your best posts, emails and pages. Then ask Claude to turn them into rules: how you sound, words you use, words you'd never say. Rules beat examples alone, because the agent can check its own draft against them.

Step 04

Rules

One note that says where things go and what needs your approval before it ships. This is what stops an agent from publishing, sending or spending on its own.

Step 05

Connect

Point every agent at the same folder.

Agents on your computer (Claude Code, Codex): open them inside the vault folder. They read AGENTS.md and follow it.
Agents in the cloud (Grokbot, Hermes): give them a synced copy. We back the vault up to a private GitHub repo every 30 minutes, so cloud agents read the same notes.

Now you can swap or add agents anytime. The brain stays.

Templates

Copy, paste, fill in

business.md

The facts every agent gets wrong when it guesses. Write it like you would brief a new hire on day one.

# Business

## Offer
- What we sell:
- Price (or range):
- What the buyer gets:

## Buyer
- Who buys (role, situation, not demographics):
- The problem they have right now:
- What they tried before that didn't work:
- What makes them say yes:

## Why us
- What we do that others don't:
- Proof (real numbers, with where each one comes from):

## Never say
- Claims we can't prove:
- Promises we don't make:

Voice prompt (writes voice.md)

Paste this into Claude with your best posts, emails and pages attached.

Here are posts, emails and pages I wrote. Turn them into voice rules for voice.md:
1. How I sound, in 3 to 5 traits. For each: what it sounds like, and what to avoid.
2. Words and phrases I use often.
3. Words I'd never use.
4. How I open, explain, prove and close.
Use only what's in my writing. Don't invent traits. Quote a line of mine for each rule.

rules.md

Where things go, and what an agent must never do without your OK.

# Rules

## Where things go
| What | Folder |
|---|---|
| Drafts and content | content/ |
| Clients | clients/<name>/ |
| Reference and research | library/ |

## Needs my OK first
- Anything that gets published or posted
- Any message sent to a customer or prospect
- Any spend
- Any new note outside the folders above

## Always
- Every number has a source.
- Client results are labelled as client results.
- When I correct you, add the fix here so it doesn't happen again.

AGENTS.md

The note every agent opens first. Claude Code reads CLAUDE.md by default, so make that one line: "Read AGENTS.md first."

# Read this first

This folder is my brand brain. Before any task:
1. Read business.md, voice.md and rules.md.
2. Follow rules.md for where things go and what needs my OK.
3. Write in the voice from voice.md.
If something you need isn't in these notes, ask me. Don't guess.

Want this set up for your business?

We turn these workflows into working marketing and sales systems.

Get a quote Book a call

Unlock the Templates

Steps 3 to 5 and all 4 copy-paste templates: business.md, the voice prompt, rules.md and AGENTS.md. Free, opens instantly.