Open foundations, sovereign control: how we build our brains
We take licensed open foundations and bring them under sovereign, audited, on premises control, so the intelligence lives entirely inside your walls.
Why we start from open foundations
Every serious intelligence system rests on a base. The interesting question is not whether you use one, but who controls it once it is running. Mickai is a Sovereign Intelligence Operating System, a SIOS, and the word sovereign is the whole point. We take licensed open foundations, the kind an organisation can hold under a clear licence rather than rent through a metered pipe, and we bring them fully inside your walls. From that moment the base stops being someone else's service and becomes your infrastructure, governed by your rules and answerable to your audit.
We are deliberate about not naming the specific bases we build on. That is not evasion, it is engineering discipline. A foundation is a starting point, not the finished thing, and the value we add sits in the layers above it: the governance, the memory, the signing, the fifty specialist brains that turn a general model into a system a regulated organisation can actually stand behind. What matters to you is not the pedigree of the seed. What matters is that once it is planted in your ground, it never has to leave.
Sovereign means it runs on your hardware, full stop
The clearest test of sovereignty is physical. Mickai runs on the customer's own hardware. It runs on premises. It runs air gapped when that is what the work demands. There is zero data egress, no public cloud round trip, no quiet call home to a service you do not control. When a lawyer, a clinician, or an operator asks where their most sensitive data goes when the model thinks about it, the honest answer is the answer that keeps everyone employed: it goes nowhere. It stays on the machine in the room.
That constraint shapes everything upstream of it. A system designed to phone home can afford to be careless about what it keeps and what it signs, because the operator can patch behaviour on their servers later. A system that must run in a building with the network unplugged cannot. So we build for the hardest case first, and the softer cases inherit the discipline for free.
Fifty brains under deterministic governance
A single general model asked to do everything is a liability in a serious setting. It is confident where it should be cautious, and it blurs the line between reasoning about the law and reasoning about a spreadsheet. So we do not ship one brain. We ship fifty, split into twenty five domain brains and twenty five operational brains, each with a defined remit and each held to account by the layer above it.
- Domain brains carry the subject expertise: the reasoning that belongs to a field and should not be improvised by a generalist.
- Operational brains run the machine itself: scheduling, memory, verification, and the checks that keep the domain brains honest.
- Deterministic governance sits over all of them, so the same input under the same policy produces the same, reviewable outcome rather than a fresh improvisation every time.
- No brain acts alone and unwatched. Governance decides what may run, in what order, under whose authority.
Determinism is the word that regulated buyers care about most, and it is the word most systems quietly cannot say. If you cannot reproduce a decision, you cannot defend it, and if you cannot defend it, it does not belong anywhere near a courtroom, a clinic, or a compliance file. We build governance to be repeatable on purpose, so that when someone asks why the system did what it did, the trail is there to walk.
Every action leaves a signed record
This is where sovereign control stops being a slogan and becomes evidence. Every action the system takes writes to a cryptographically signed audit record, the Open Audit Record. It is not a log you have to trust because the vendor says so. It is a record that can be verified independently, because it is signed, and the signatures use post-quantum cryptography (ML-DSA-65) so the proof does not decay as the threat landscape moves.
The practical effect is a change in posture. Instead of asking a supplier to promise good behaviour, an organisation can check it. Instead of hoping the model did not touch something it should not have, the operator can read what it did, confirm the record was not altered, and hand that record to an auditor, a regulator, or a court. Trust becomes something you demonstrate rather than something you request.
“Sovereignty is not a marketing word for us. It is the difference between promising your data stayed home and being able to prove it, on your own hardware, with a signature no one can quietly forge.”
Memory the customer owns
A brain without memory forgets your context every morning. A brain whose memory lives on someone else's servers hands your accumulated knowledge to a party you do not control. Neither is acceptable in a sovereign system, so we do the third thing: the memory belongs to the customer, sits on the customer's storage, and never leaves it. The longer the system works for an organisation, the more valuable that memory becomes, and every ounce of that value accrues to the organisation rather than to us.
That is the quiet compounding advantage of doing this properly. A rented intelligence gets more expensive as you rely on it. A sovereign one gets more useful, because the context it holds is yours to keep, yours to move, and yours to switch off if you ever choose to.
Built on a filed foundation of our own
The engineering does not stand alone. It stands on a body of intellectual property we have filed: 104 filed UK patent applications carrying approximately 2,340 claims, complete with full specifications, claims, and figures, now building toward examination and grant. These are filings, not grants, and we are precise about that. They describe how sovereign governance, the signed audit record, and the brain architecture actually work, in the detail that examination will test.
The market has started to notice the shape of this. On Crunchbase our founder now ranks number two, and the company Heat Score has reached 94 out of 100, having climbed from single digits. We read that signal for what it is: early, public, and encouraging. It tells us the thesis, sovereign intelligence that runs where the data lives, is landing with the people who watch this space closely.
Where this goes next
The direction of travel is set. As open foundations get stronger, our advantage does not erode, it deepens, because a better base makes the sovereign layers above it more capable without changing where the intelligence lives or who controls it. We will keep taking the best licensed foundations, keep bringing them fully on premises, and keep tightening the governance, the signing, and the memory so that the most sensitive work in the world can finally be done by an intelligence that never has to leave the room. That is how we build our brains, and it is why we build them the way we do.





