KXCO Sovereign AI

Rent the intelligence. Own the knowledge.

You run the model. The ontology holds the pattern. Your understanding of your own business compounds somewhere you can query it, cite it, correct it and take it with you.

Round Table over your own inference, on your own hardware. The same typed structure, the same provenance, the same post-quantum attestation, with the weights inside your estate and the record naming which model proposed every fact.

Request a briefing → Read the long form →
Customer-hosted Open weights ML-DSA-65 · NIST FIPS 204
What Sovereign AI Is

Your weights. Your hardware. The ontology over the top.

Every model an organisation uses is rented: somebody else's weights, on somebody else's hardware, on somebody else's release schedule. That is a fair arrangement for the reasoning and a poor one for the thing being reasoned over. Push your understanding of your own business into a model and you have handed over the asset and kept the invoice.

Sovereign AI draws the line exactly there. The inference runs where you decide it runs. Over it sits the layer that makes a model deployment defensible in a regulated setting: a typed structure for your domain, a provenance record on every fact, separation of duties between the operator and the person who accepts, and a cryptographic attestation chain that proves what happened.

KXCO delivers knowledge sovereignty. AI does not.

What It Delivers

Six things you hold at the end of it.

Not capabilities. Things the organisation owns and can prove.

Delivers
The model becomes replaceable
The ontology holds the pattern, so the model is a component rather than a foundation. Frontier this quarter, open weights next quarter, a model fine-tuned on your own accepted history after that. Not one rule inside the ontology changes when you move, because none of them were written in terms of a vendor.
Delivers
Your context stays in your estate
A prompt is not a throwaway string. It carries your pricing logic, your client context and your operating procedure. When the inference runs on hardware you control, all of that stays where your policies already apply, on the way out as well as on the way back.
Delivers
The record names which model
Every fact carries its origin, its source, the date it was true as of, the person who accepted it, and the specific model that proposed it. How much of this a machine wrote, and which machine, are answered from the record rather than from anyone's memory.
Delivers
Industry vocabulary on day one
Round Table ships with starting models for the domains institutions actually operate in: corporate oversight, trading equities, portfolio, AI sector, biotech research, object tokenization, customer behaviour, behavioural analysis and process. Your first day is spent on your own entities.
Delivers
Attestation anyone can verify
Material acceptances are signed with ML-DSA-65 and anchored on the open record. A counterparty, an auditor or a regulator checks the signature against a published key with open tooling, and satisfies themselves without involving you at all.
Delivers
Knowledge that compounds
The second dataset takes a fraction of the time the first did, because the vocabulary exists and the entities are already there. That curve is the asset, and it is the one thing in the estate that is worth more every year you keep it.
How It Works

From the install to the second dataset.

Six steps. You hold the controls at every one of them.

01
We install it with you
Five days on site, with the preparation before and the handover after. One real dataset, never a synthetic fixture, and one workflow taken all the way through review at volume. You leave owning the ontology.
02
You point it at your inference
The engine reaches a model through a single deliberate point in the system. Set the endpoint to your own serving stack and the whole engine follows: population, retrieval, assessment, extraction and the guide, all on your weights.
03
Your people fill the queue
The model reads your material and proposes typed structure. A named person with the right role accepts, corrects or rejects each proposal, and that acceptance is itself a recorded, attributable act. The model proposes. It never decides.
04
Every acceptance is attested
Material changes are signed with ML-DSA-65 and recorded, so the record is tamper-evident and independently checkable. This is the point at which an internal system becomes evidence.
05
The knowledge compounds
Each dataset lands into the same model and connects to what is already there. Questions people used to ask each other start being asked of the model, and the answers carry sources and dates.
06
You change models whenever you want to
A better model appears, and pointing at it is a change of address. The knowledge does not migrate, because it never lived in the model. You should never have to migrate the truth.
What To Have Ready

One thing, and it is worth getting right.

Sovereign AI runs on your inference, so the one prerequisite is the endpoint itself: GPU inference serving an instruct-tuned open-weight model materially larger than seven billion parameters, reachable over your own network, with usage reporting enabled on the streaming endpoint.

That figure is measured rather than assumed. A model that cannot hold a schema-constrained tool definition cannot populate an ontology, and the difference between a model that can and one that cannot is visible within a single run. Getting the endpoint right before day one is what turns the installation week into five days of your own data rather than five days of infrastructure.

Bring the endpoint and we bring everything above it.

What It Costs

An upgrade on the licence, not a fourth price list.

Sovereign AI sits on the institutional and enterprise licences, the way the Round Table upgrade does.

One time
Installation
Five days on site$25,000

Fixed price. One real dataset, one workflow through review at volume, and you leave owning the ontology.

Credits in full against the first year of the licence where you proceed within 60 days.

Per year
Platform licence with sovereign deployment
Institutional$150,000 to $190,000
Enterpriseby negotiation

An uplift on the standard institutional band, which runs $120,000 to $150,000.

Forward-deployed engineering beyond the installation is quoted per engagement.

Request a briefing → The KXCO ontology →

Take the business as seriously as the US Government takes its wars.

The knowledge your organisation runs on, and the cryptography protecting it, are the two things that decide whether it still owns itself in ten years.

Request a briefing → KXCO Sentinel →

KXCO is a technology provider. Sovereign AI is software and forward-deployed engineering. The inference runs on infrastructure the customer controls, and the ontology, the manifest and the attestation chain are the deliverables.