What Is a Sovereign AI Operating System?

A sovereign AI operating system is software that runs artificial intelligence entirely under the owner's control: on hardware they own, without sending data to anyone else, and with a record that proves what it did. Mickai is one. It is a British Sovereign Intelligence Operating System, or SIOS, built by Mickai LTD (Companies House 17166618). It behaves like an operating system because it hosts many applications on one shared engine, and it is sovereign because the intelligence is owned, offline and provable. In plain terms: the model runs locally, nothing leaves the machine, and every action leaves an audit record you can verify yourself.

• It runs AI on hardware you own and control, not on someone else's servers.
• It keeps data offline: no cloud round-trip and no external dependency when the model is working.
• It hosts many applications (called studios) on one local engine (called Poros).
• Every action produces an audit record (the Open Audit Record) you can verify offline.
• Sovereign here means three precise things: owned, offline and provable.

A plain definition

Strip away the marketing and a sovereign AI operating system is easy to define. It is an operating system whose central capability is artificial intelligence, and whose defining property is that the intelligence stays under the control of whoever runs it. The operating-system part means it manages resources and hosts applications. The sovereign part means those applications draw on AI that is local, private and accountable to the owner, not to a vendor's cloud.

The word operating system is not decoration. Like a conventional OS, it sits between the hardware and the applications, provides shared services they all use, and keeps them isolated from one another. The shared service here happens to be intelligence, delivered by one local engine that every application calls.

How it differs from a cloud AI service

A cloud AI service sends your prompt and your data to a provider's servers, runs a model there, and sends an answer back. You depend on their uptime, their policies and their goodwill with your information. A sovereign AI operating system inverts that. The model runs on your machine, your data never leaves it, and there is no third party in the loop at the moment of inference. If your network cable is unplugged, a cloud service stops; a sovereign system keeps working.

The practical consequences follow from that one difference. There is no per-request data egress to worry about, no vendor to trust with confidential material, and no silent change to a model you did not choose. What you run is what you audited, on hardware you can point to.

How it differs from a single local model

Running one local model is a good start, but it is not an operating system. A lone model is a single capability with no applications around it, no isolation between tasks, no shared services and, usually, no record of what it did. It answers questions; it does not run a business.

A sovereign AI operating system adds the layers that make local intelligence usable and trustworthy. It gives you applications for real work, an engine that serves them all consistently, boundaries so one task cannot see another's data, and an audit record so every action can be checked afterwards. The difference is roughly the difference between a single program and a whole operating system: one does a job, the other hosts everything that does jobs.

The architecture: studios, one engine, one audit record

Mickai's architecture has three parts worth naming. The applications are called studios. There are sixty-three of them on one engine, covering different kinds of work, and they behave like the apps on any operating system: you open the one you need and it draws on the same underlying intelligence. Fourteen are production-ready at launch and forty-nine are in active development.

The engine is Poros. It runs a large, capable model locally, bound to the loopback interface only, and it is fully offline. Every studio talks to Poros; none of them talks to the outside world to think. That single, shared, local engine is what makes the whole system sovereign rather than a collection of disconnected tools.

The record is the Open Audit Record, or OAR. Every meaningful action is written into it, post-quantum-signed using FIPS 204 (ML-DSA-65), so that tampering shows. It is tamper-evident rather than tamper-proof, an important distinction: the point is not to claim nothing can ever be altered, but that any alteration is detectable, and the record can be verified offline without calling home to anyone.

What sovereign precisely means: owned, offline, provable

Sovereign is a word that gets stretched, so here is the exact meaning in this context. Owned: the system, the engine and the data run on hardware the owner controls, with no vendor holding the keys. Offline: inference happens locally, with no network round-trip required to produce an answer, so the system works in an air-gapped room. Provable: the system produces its own evidence, a signed audit record you can check yourself, so trust does not depend on taking anyone's word.

Those three properties are the test. If the intelligence lives on someone else's servers, it is not owned. If it needs the internet to think, it is not offline. If you cannot verify what it did without asking the vendor, it is not provable. A system has to satisfy all three before the label sovereign is honest.

Where Mickai fits

Mickai is a working instance of this definition rather than a thought experiment. It is British, built by Mickai LTD, with sovereignty, auditing and licensing enforced by mechanisms filed as 104 UK patent applications carrying 2,340 claims, patent-pending at the UK Intellectual Property Office. The design philosophy is to patent the mechanism, not the model, so the engine can host a capable model without the model itself being the product.

The closed beta is open at mickai.co.uk/beta, where we are onboarding a first cohort of design partners. For readers who want the underlying ideas in more depth, I have published technical preprints on Zenodo, under ORCID 0009-0000-5511-5858, covering offline attestation, sovereign inference, private knowledge bases and turning compliance into evidence.

Frequently asked questions

What is a sovereign AI operating system in one sentence?

It is software that runs artificial intelligence on hardware you own, keeps your data offline, and produces a verifiable record of what it did, so the intelligence is owned, offline and provable.

Is a sovereign AI operating system the same as running a model on my laptop?

No. Running one model locally is a component, not a system. A sovereign AI operating system adds applications, a shared engine, isolation between tasks and an audit record, so it hosts real work rather than answering a single query at a time.

Does it need an internet connection?

Not to think. In Mickai, inference runs on the local engine, Poros, bound to loopback only and fully offline. You can run it in an air-gapped environment. Network access is not required for the AI to do its work.

How can I trust what it did?

Through the Open Audit Record. Every meaningful action is written to a record that is post-quantum-signed with FIPS 204 (ML-DSA-65) and can be verified offline. It is tamper-evident, meaning any alteration is detectable, so you can check the system's behaviour yourself.

How many applications does Mickai have?

Sixty-three studios run on one engine. Fourteen are production-ready at launch and forty-nine are in active development. Each studio is an application for a particular kind of work, and all of them draw on the same local engine.

How is this different from a private or on-premise cloud model?

A private cloud still centralises the model on servers you must reach over a network and trust to behave. A sovereign AI operating system puts the engine on the owner's own hardware, keeps inference offline, and proves its behaviour with a signed, offline-verifiable audit record rather than a vendor's assurance.