The AI system we install for clients, written up so you can see exactly what you would get. The same setup we run in our own businesses every day.
Most AI strategy is a shopping list. Pick a chatbot. Pick a workflow tool. Plug them into your business. Hope it works.
We don't do it that way. We treat AI like any other part of your business. It has to do specific work, follow specific rules, and genuinely add value. So we built it as four parts that work together. Each one does a different job. None is optional. Take any of them away and the rest fall apart.
We needed something different to install for our clients, so we built it. We use it ourselves first. The version below is the version we run every day.
Most AI tools start cold. Every time you use them, they are guessing from general knowledge. They do not know what your company has actually decided, what has worked, what has not, who said what last week.
Our setup starts with what is actually true in your business. It pays attention to the things you have decided and committed to. It picks up everything else from your real systems (your email, your calendar, your documents, your meetings) and keeps track of where each piece of information came from. When it tells you something, you can ask "where did you get that?" and it can tell you. It never tries to be the master copy of facts that other systems already own.
Confidentiality is the part most AI tools quietly drop on the floor. A document gets marked confidential. Then someone summarises it. Then that summary gets used somewhere else. Then it gets quoted in a meeting note. The "confidential" label fell off at step one. The sensitive parts are now travelling around with no protection.
Our setup handles confidentiality differently. Anything that comes from a confidential source is treated as confidential too, no matter how many times it gets used or transformed. A summary of a confidential document is still confidential. A search index built from confidential content is still confidential. The cloud only sees what has been explicitly cleared.
This is the first thing we have not seen anyone else doing yet. The practical effect: you can use AI on the sensitive parts of your business without those parts ever leaving your control.
Most AI either does nothing without being asked, or it acts on its own and you find out later (sometimes too late). Both versions fail in real businesses.
Our setup has three modes you can dial in for different kinds of work. Just propose: the AI suggests, you approve. Act and tell you: the AI does the thing and lets you know. Just handle it quietly: the AI does routine work without checking in. You decide which mode for which kind of work.
Money, contracts, anything legal, contacting people on your behalf: those always need your approval. Forever. Those rules do not bend. There is also a weekly limit on how much of your attention the AI is allowed to spend. It cannot flood you with notifications, even if it thinks it has good reason.
And it learns from you. Every time you correct it, that correction becomes a rule it follows next time. It never decides for itself that it is doing well.
Most AI gives you one confident answer. Even when the question is hard. Even when the right answer depends on judgment calls that could go either way. Ours is different on the calls that matter.
For the important decisions, four AIs from four different companies (each trained differently) argue the question out. A final pass pulls together what they all agreed on, what they disagreed on, and which arguments held up. You do not just see the answer. You see the disagreements.
Most AI smooths over arguments. Ours preserves them. When the four AIs agree, you can trust the answer more. When they disagree, you know exactly where the uncertainty is. That is the part you can act on.
This costs two or three times more than asking one AI, so we only do it on decisions that earn it. For everyday work, one AI is fine. For the calls that matter, the four.
Here is the whole thing in one picture. Your business stuff comes in on the left. The brain in the middle remembers what matters and keeps the private parts private. The AI works in the background through whatever you prefer to use (chat, email, voice, mobile). The four-way debate gets convened only for the calls that earn it.
We have looked at what everyone else is building. Two things in our setup we have not seen anywhere else.
There are roughly three ways most setups handle this today. Send everything to the cloud and trust the cloud (or a private tenant inside it). Put a "confidential" label on the sensitive folders and let everything else go to cloud. Or lock the whole stack on your own infrastructure and only use models you can host yourself, which is what law firms, hospitals and defence-related work do when regulation forces it.
All three have a cost. Cloud, even a private tenant, means trusting another company's servers and contracts with your most sensitive material. Labelling folders leaks because the moment a sensitive document gets summarised, searched, or quoted into a memo, the summary, the search index and the memo all sit in the "fine" folders, and the cloud sees them. The locked-on-premise version solves the confidentiality problem, but it cuts you off from the best models: you run open-source AI that lags the latest from Anthropic, OpenAI and Google by months, and people quietly start pasting things into ChatGPT on the side because the secured environment is too slow.
Ours works differently. Every piece of data carries information about where it came from. Anything derived from a confidential source, whether a summary, a search index, a draft or a conclusion, inherits the same status and routes the same way. The boundary is not a folder. It is a property of the data itself that travels with it wherever it goes.
You get the best cloud AI for the 80 to 90% of your work that does not need protection, and a structural guarantee on the rest. The hybrid where the boundary travels with the data instead of the folder is what we have not seen built anywhere else.
Most companies building AI right now pick one of these four parts and build that one thing well. Some focus on memory. Some on AI agents that act on their own. Some on getting multiple AIs to check each other. Each is useful by itself.
The real value comes from all four working as one system. One brain that knows your business, takes in your world, acts in the background, keeps the private parts private, and gives you the answer that survived an argument. We have not found anyone else doing it as one integrated thing.
Two genuine claims. Both testable. If you find someone else doing either of these, tell us, because we would like to know.
This is not theory. The version we just described is the same one drafting our follow-up emails right now, watching for what is slipping in our calendar, and proposing what we should do next. It runs every day across our businesses. The boring foundation work, making sure nothing hangs, making sure failures always come back as a clear signal, making sure the system can recover from a bad day, all of that gets done before any new feature ships.
When we install this for a client, the version they get is the same shape, fitted to their business. The four parts are the same. The judgment calls inside each part get tuned to your specific situation: what counts as true, what counts as confidential, what kinds of work the AI is allowed to handle on its own, which calls are big enough to bring all four AIs in.
That is the difference between a strategy deck and a working system. The deck describes what could be done. The working system is already doing it.
We confirm whether this is the right shape of engagement for your business, and what it would take to start. Honest read at the end of the call, either way.