One record of what your business knows, that people and AI can trust.
Most of what your company needs to know, it already knows. It is spread across inboxes, drives, systems and people’s heads: the pricing rule in a spreadsheet, the reason you stopped doing it the old way in somebody’s memory, the commitment you made to a client in a contract nobody has opened since.
A company brain brings that together as one record. Every entry has a source and an owner, so an answer can be checked, and someone decided it belongs there, so nothing is in it by accident. It is kept in plain files your team can open and read, and it connects to the systems you already run rather than replacing them.
The AI tools your people use read from it, under your rules. That is what makes them useful about your business rather than about business in general: why that matters.
It is not a data migration, a new system of record or an eighteen-month programme. It reads alongside what you already have.
Three layers: the record, the knowledge graph and the context.
Underneath, the design has twelve parts: what a company brain is made of.
The record
What your company knows, kept as plain files your team can open and read: decisions and the reasons for them, policies, commitments, what your own terms mean. A named person approves every source before anything is read from it, each with a date to review it. The database built from these files is disposable; the record is not.
The knowledge graph
Every fact linked to what it touches: this invoice to that customer, the customer to the complaint, the complaint to the renewal at risk. The links are authored, by people or as each note is written, not guessed by having AI read everything.
The context
What a person or an AI tool is given when it asks: the facts that answer the question, each with its source, and only what that person is allowed to see. The permission check runs before every answer, in the system rather than in an instruction to the AI.
Honest about what it does not know.
Every answer carries its source. When the evidence is not there, it is built to say so, and why, rather than guess.
It won’t answer everything.
A count or a total gets a checked query or a refusal, never an estimate. Before you sign, we tell you how many of your real questions fall outside what it can answer.
It won’t know what nobody told it.
There is no magic import. Day one is thin, and the value builds as work flows through it. We would rather you knew that now than found out in month two.
It won’t guess its links.
It does not have AI read everything and decide what relates to what, and it does not turn your questions into database queries on the fly.
It won’t act without an owner.
No agent is switched on without a named owner, and what any AI may do is a permission you grant, which expires unless you renew it. It claims no accuracy figure until your own questions have measured one.
Yours, either way.
Plain files you can open, an index that rebuilds from them, on your own servers or in your own cloud account. Nothing is locked into any vendor, us included, and you can take it in-house at any time.
You decide who can see what, and the system enforces it, rather than an instruction the AI is asked to follow. Every question, answered or refused, is recorded with who asked it and which sources the answer drew on, and every permission you grant expires unless you renew it. For your most sensitive material, we can build it on a private or open-source model that runs in your own environment.
Before we build yours, we run ours: how we run ours.
Built for you, then run by your team or by us.
We build it and hand it over, and your team runs it: a fixed fee, then nothing keeps billing but your own cloud and AI usage. Or we build it and run it for you, keeping it current, checked and growing: a set-up fee, then a monthly fee set by the size of your company. That is the route for most companies, because the person who would look after it is hard to hire and hard to keep.
What else you might try, and where each one stops.
A copilot over your shared drive
It searches whatever is stored: drafts, duplicates, the version withdrawn in March. Good for finding a document; it cannot tell which one counts. A company brain holds what someone has decided is true, with an owner and a source on every entry.
A second brain or a notes app
It works because one person owns it and may read all of it. A company has hundreds of readers who must not, many writers, personal data and the law, and it has to outlive everyone who built it.
A data warehouse and dashboards
Good at the numbers your systems already record. It holds what happened, not why: the decision, the reasoning, the commitment, what your terms mean. A company brain holds those, and leaves the counting to a checked query.
An AI vendor’s platform
It sells the plumbing: memory, connectors, a model. Useful, and generic. What your company means by a good outcome, which exceptions matter and how decisions really get made arrive in nobody’s default schema. That part is the work, and here you own it rather than rent it.
Better prompts
A better instruction changes the question. It does not change what the model knows, so everyone prompting the same model lands on the same answer. Context only you have is what moves it.
Hiring someone to build it
Possible, if you can find them. The person who would build and look after a company brain is hard to hire and hard to keep. We build it with your team, then hand it over or run it for you.
Questions about company brains
“What is a company brain?”
One governed record of what your business knows, connected to the systems you already run and readable by the AI tools your people use. Every entry has a source and an owner, and who can see what is enforced by the system itself.
“Is a company brain the same as a knowledge graph?”
The knowledge graph is part of it: the links between the facts. A company brain is the whole layer: the record the graph is built from, the graph itself, and the rules that decide what each person and AI tool can see.
“How is it different from enterprise search or RAG?”
Search, and retrieval for AI (often called RAG), return whatever matches your question, whether or not it is current or approved. A company brain retrieves from a record someone has decided is true, with an owner and a source on every entry, and it is built to say so when the evidence is not there rather than guess.
“Do we need to replace our systems?”
No. It reads from what you already run. Nothing is replaced or migrated, and we build only the connections the work needs.
“Which AI tools can use it?”
Claude, ChatGPT, Copilot and Gemini are built to connect through the open standard AI assistants use, under the same rules, so the brain outlives any one choice of tool.
“Who owns it?”
You do, either way: plain files and an index that rebuilds from them, on your own servers or in your own cloud account. Nothing is locked into any vendor, us included, and you can take it in-house at any time.
“How long does it take to build?”
Bounded work, measured in weeks, with senior people working alongside your team. It starts with a discovery call and the question audit, so you know what it would and would not answer before anything is built.
Start with a conversation about your company.
A discovery call comes first: what your business needs to know, where that knowledge lives today, and whether a company brain is the right answer. If it is not, we will say so.
Then the question audit: about two hundred real questions from your business, answered in writing, so you know what a company brain would and would not do for you before anything is built. It is a small fixed fee, credited against the build or the service if you go ahead.
Not ready to talk? The ten-minute diagnostic (opens in a new tab) shows how much of how your company runs a model can reach today, and the skills run the same method in your own AI assistant.