AI Operating System · Custom intelligence
Intelligence the business owns.
The company brain is the last stage of the roadmap: everything your business knows, connected into one living, queryable intelligence. Twenty years of decisions, pricing, projects and judgement, available to every person and every agent you run.
Roadmap · Stage 06
What a company brain is made of
The operating system puts your knowledge to work; the brain is where it compounds. Every skill file, context file and workflow the harness runs feeds one asset that gets smarter with use, and it belongs to you, not to a vendor.
Living knowledge graph
Your tools, documents and data vectorised into one connected institutional memory. Not a folder of files: a graph of how your business actually fits together, that every skill and agent can query.
Custom and fine-tuned models
Proprietary intelligence trained on how your business actually works: your calls, your documents, your decisions. The reasoning is frontier, the knowledge is yours alone.
Feedback loops
Prompts versioned, retrieval scored, answers graded against outcomes. The brain improves the way software does: measured, deliberate and continuous.
Ask it what only your business knows
A frontier model knows the world. It does not know your margins, your near-misses, or why the last three quotes like this one lost money. The brain answers the questions no search box and no new hire can, because it remembers everything the business has already learned.
It is the difference between AI that helps people write faster and AI that makes the whole organisation harder to compete with.
- “What did we quote for jobs like this one, and what did they really cost us?”
- “Which clients are showing the same warning signs the ones we lost showed?”
- “What did we learn the last time we ran this campaign?”
- “Who solved this problem before, and what did they do?”
Built and running
For The Ecommerce Accelerator we built specialised agents for every role, then connected them through one central operating layer that holds each agent's instructions and knowledge: a company brain the whole team works from. It is a large part of why the same five people now get 40+ hours back every week.
Read the Ecommerce Accelerator case studyHow a brain actually gets built
Not as an import. A brain is a by-product of the operating system running: every skill file, context file and workflow is a written form of how the business works, and the graph is what connects them. Four stages, in this order, because each one depends on the last.
The operating system goes in first
Skills, context files and role design over the Claude, ChatGPT, Copilot or Gemini organisation you already pay for. This is not a prerequisite we invented to sell a second thing: those files are the written form of how your business works, and they are what the graph is built out of.
The graph connects what already exists
Documents, tickets, quotes, calls, policies and the systems they live in, vectorised and linked so the relationships survive. The hard part is never the embedding, it is deciding which of three conflicting versions of a policy is the current one. That decision is the work, and it needs someone internal who can make it.
Models tuned on how the business actually works
Frontier reasoning, your knowledge. Tuning on your calls, documents and decisions is what moves an assistant from generically competent to specifically useful, and it is the point at which the asset stops being replicable by a competitor with the same subscription.
Feedback loops keep it honest
Prompts versioned, retrieval scored, answers graded against what actually happened. Without this a brain degrades quietly: the answers stay fluent while they drift out of date, and nobody notices until someone acts on one.
What a company brain is not
The phrase gets attached to a lot of things. Four it is not, so you can tell what you are being sold elsewhere.
Not a search box
Search finds documents. A brain answers questions about the business, including ones whose answer is spread across four systems and one person's memory.
Not a data warehouse
A warehouse holds structured rows for analysis. A brain holds meaning: how the business speaks, what it decided, why, and what happened next. Most of that never had a schema.
Not a vendor's assistant with your files in it
The knowledge lives in files and a graph you hold, reachable by any model. If leaving the vendor means leaving the memory behind, it was never institutional memory.
Not a one-off migration
It compounds or it decays. A brain that is loaded once and never fed by the work is a snapshot that gets less true every week, which is why the operating system underneath it matters more than the import.
Frequently asked questions
What is a company brain?
A company brain is the layer that holds what an organisation knows, connected so that both people and AI agents can query it. In practice it is three things: a knowledge graph over the tools and documents the business already has, models tuned on how that business actually works, and feedback loops that score whether the answers were any good. It is different from a search box because it answers questions about the business rather than finding files, and different from a chatbot because the knowledge belongs to the company rather than to a vendor.
How is this different from just uploading our documents to ChatGPT?
Uploading documents gives an assistant a pile of files. A company brain gives it a structure: which document is current, who owns a process, what a term means here, how a quote relates to the job it came from and what that job actually cost. Retrieval over an unstructured pile degrades as the pile grows, which is why the first month feels impressive and the sixth feels unreliable. The structure is the part that keeps working, and it is also the part nobody builds by accident.
Do we own it, or does it live inside your platform?
You own it. The knowledge lives in plain files and a graph you hold, reachable by any model over an open interface rather than locked to one vendor's assistant. That is deliberate: an institutional memory you cannot leave with is not an asset, it is a lease. If you stopped working with us, the brain and everything in it stays with you.
What does it need from us before it works?
Two things, and neither is data volume. It needs the AI Operating System underneath it, because the skills and context files people use every day are what feed the graph. And it needs someone internally who can settle questions like which of three conflicting policies is current. The organisations that get value quickly are the ones that treat those disagreements as the work rather than as a blocker.
How long does it take to be useful?
The first useful answers come out of the operating system in weeks, because the skills and context files are already a written form of how the business works. The graph itself compounds from there: it gets materially better across the first few months as more workflows run through it and the feedback loops start scoring what came back. Anyone promising a complete institutional memory in a fortnight is describing an import, not a brain.
Where has this actually run?
The Ecommerce Accelerator is the clearest example. We built specialised agents for each role, then connected them through one central layer holding every agent's instructions and knowledge. It is a large part of why the same five people get more than forty hours back every week. The case study on this site sets out how it was built.
Is this the same as a RAG system?
Retrieval-augmented generation is one technique inside a company brain, not the thing itself. RAG answers the question of how a model reaches a document at query time. A brain also has to answer which documents are authoritative, how entities relate, who owns what, and whether the answers are getting better or worse over time. Teams that build RAG alone usually find the retrieval is fine and the knowledge underneath it is the problem.
The brain is built on the operating system. Start there.
The knowledge graph is only as good as the harness feeding it, which is why every engagement lands with the AI Operating System first. The free audit maps where your knowledge lives today and what a brain over it would answer.