Meta Pixel

Second brain for business

Stop explaining your business to AI.

A second brain orchestrates all your context: every client, process and decision in one knowledge graph that Claude, ChatGPT and your whole team read before they answer.

The problem

Your AI has no memory of your business.

Every chat starts from zero

You explain your clients, your prices and your tone of voice, get a good answer, and next week explain it all again. The AI never learns your business; it only borrows it for one conversation.

What it knows is stuck in one app

What you taught ChatGPT, Claude has never heard. What one person set up, the rest of the team cannot use. When someone leaves, their context leaves with them.

A longer prompt will not fix it

Pasting in more documents fills the context window, the AI's short-term memory, and the oldest detail quietly falls out. More prompting is not the fix. Memory is.

What it is

An LLM wiki for your business.

Andrej Karpathy, a founding member of OpenAI, described a better way to give AI memory: instead of re-reading raw documents every time, the model keeps its own wiki. Short, plain pages it writes and updates itself, each linked to the others.

We build that wiki for a whole business. It is the one place that orchestrates your context, so every AI and every person works from the same, current picture. If you already keep a personal one, here is where a second brain in Notion stops working for a team, and how a knowledge graph differs from a second brain.

Read: the LLM wiki for organisations
Second brain
The name most people know: one trusted place your knowledge lives, outside your head.
AI brain
A second brain your AI reads before it answers, so it knows what you know.
Company brain
The same thing for a whole team: shared by every person and every agent.
LLM wiki
How it is written: plain pages an AI model writes and keeps current itself.
Knowledge graph
How it is found: every link between pages is an edge the AI can follow.

How it works

From scattered context to one knowledge graph.

  1. 01

    Capture

    Meetings, emails, documents and your tools (CRM, accounts, drive) feed in as the work happens, not in a big project nobody finishes.

  2. 02

    Write

    The AI writes it up as short markdown pages: one per client, process, product, decision and person, in plain language a new hire could read.

  3. 03

    Link

    Every page links to the pages it touches. The links are the knowledge graph: a client links to its projects, a project to its people and decisions.

  4. 04

    Ask

    Any AI follows the links to the few pages that matter, reads them first and answers with the source page beside it, so you can check it.

  5. 05

    Keep current

    A daily review flags pages that have gone stale, pages nothing links to and pages that contradict each other, for a person to fix.

Open it in Obsidian

The brain is a folder of markdown files, so you can open it in Obsidian, see the whole graph and edit any page by hand. Ours lives in an Obsidian vault. Prefer Google Drive, SharePoint, GitHub or Notion? Those work too.

It gets smarter

It writes itself as the work happens.

Every job a person or an agent finishes writes back what it learned: a client's payment terms, a word the brand never uses, the day leads reply. Nobody sits down to update it.

Evals run alongside. Each run is scored against your own baseline and the next fix is proposed from what actually happened, so the brain and every workflow on it get better each week.

Recent writesExample
  • finance/invoicesNorthline pays on the 20th, not the 7th. Terms corrected.
  • marketing/brandNever say “leverage”. Added to the vocabulary.
  • finance/reconciliationStripe fees post a day late. Match on the day after.
  • marketing/cold-outreachWorkplace fleet leads reply most on Tuesdays.

Nothing here was typed by hand. Each line is a department recording what it learned while doing something else.

The difference

From a chat that forgets to a brain that knows.

A chat on its ownWith a second brain
MemoryForgets when the chat ends.Remembers everything it has been told, as pages.
A new client or hireYou explain it all from scratch.It reads the pages and is up to speed.
When a fact changesOld answers keep coming back.One page is updated and every answer follows.
Your AILocked inside one app.Claude, ChatGPT or Teams: same brain.
Who owns itThe app does.You do. Plain files in your own storage.
Can you check itNo source, take it or leave it.Every answer shows the page it came from.

Yours to keep

Files you own. Any AI you like.

Your brain is plain files in your own storage, under your own accounts. We connect it to Claude, ChatGPT, Teams and Slack over MCP, an open standard, so everyone asks it where they already work and gets the same answer.

Change models next year and nothing is lost: the brain stays, the AI on top of it is swappable. We never train a model on it.

Pricing the Kōwhai fit-out

Your data

Where your data lives, and who can act on it.

Files you own

The brain is plain markdown in your own storage, under your own accounts. Take it with you whenever you like; nothing is held hostage.

Read-only first

Connections to your tools start read-only. Writing back to a system comes later, and only where you have said yes.

A person signs off

Anything that would send, post or pay is left as a draft for a named person. Every run is recorded, so you can see what it did.

Ours, live

We run Sentry AI on our own second brain.

1,342linked pages
17top-level sections
Every 6 hoursevery agent re-reads the same current pages

Calls, meetings and decisions since January 2026, written up as linked pages our team and our agents ask every day.

See our second brain

AI Departments

Put departments on top of it.

Ready-made departments for marketing, finance and operations read your second brain before every job and write back when they finish. Install the base, then add a department.

pnpm dlx sentry-ai init
See AI Departments

Get started

Build your own, or have us build it for your team.

Build your own

For founders, sole traders and anyone who wants an AI brain of their own. Free, in your own Claude.

  • The second brain starter: Claude interviews you and writes your first pages
  • The starter playbook: the twelve pages to write first
  • A step-by-step guide to building it yourself

Want it running a whole department for you? See AI Departments.

We build it for your business

For a team that wants one brain shared by every person and every agent, built with you from the work you already do.

  • Connected to your tools and your team's Claude, ChatGPT or Teams
  • Private to your organisation, in your own accounts
  • Kept current every week, measured in your time saved

Questions

Second brains, AI brains and knowledge graphs, answered.

What is a second brain for a business?

A second brain is one place that holds how your business works: your clients, products, prices, processes, decisions and people, written as linked pages. Your AI reads it before it answers, so it already knows your business instead of you explaining it again in every chat.

Is an AI brain the same as a company brain?

Yes. Second brain, AI brain, company brain and business brain are four names for the same idea: a shared memory your team and your AI both read and write. We say second brain because it is the name most people already know.

What is an LLM wiki?

An LLM wiki is a set of plain-text pages that an AI model writes and keeps up to date itself, one page per client, process, decision or person, each linked to the others. Andrej Karpathy described the pattern for personal research; we build it for a whole business, so every agent reads the same pages.

Where does the knowledge graph come in?

Every link between two pages is an edge, so the pages together form a knowledge graph. That is what lets the AI find the five pages that matter for a question instead of re-reading everything, and it is why the brain keeps working as it grows to thousands of pages.

Do I need Obsidian?

No, but it is the nicest way to look at one. The brain is a folder of markdown files, so you can open it in Obsidian and see the graph, or keep it in Google Drive, SharePoint, GitHub or Notion. Nothing is locked inside a tool of ours.

Which AI does it work with?

Claude, ChatGPT, Microsoft Teams, Slack and anything else that speaks MCP, the open standard for connecting AI to your systems. The brain stays the same when you change models, so you can switch AI without starting again.

Who owns the brain, and is it private?

You do. It is plain files in your own storage, under your own accounts. We never train a model on it, and if you stop working with us you keep every page.

How does it stay up to date?

The work keeps it current. Meetings, emails and decisions are written up as pages as they happen, and a daily review flags pages that have gone stale, pages nothing links to and pages that contradict each other, for a person to fix.

Can I build my own second brain?

Yes. Start with our free second brain starter: Claude interviews you about how your business runs and writes the first pages. When a whole team and its agents need to share one brain, that is when we build it with you.