The Mickai Beta: Early Access to Sovereign AI You Own and Run Offline

The Mickai beta is open for applications. It is early access to the Mickai Sovereign Intelligence Operating System, our British sovereign-AI operating system, and it puts ten production-ready studios on hardware you already own, running entirely offline, with nothing leaving the building. At the centre of it is Poros, our sovereign inference engine. Poros runs a large, capable AI model locally, on a machine you control, and seals what it does into an audit record you can verify for yourself. Own it, run it offline, prove it. That is the whole of the offer.

For years the deal with capable AI meant sending your data out to get an answer back. The Mickai beta is the other way round. The model comes to where your data already lives, and it stays there. This piece explains what the beta is, what Poros does, what is ready today, and how to apply.

• What it is: early access (a closed beta) to the Mickai SIOS, applied for and gated, not an open download.
• What runs it: Poros, our sovereign inference engine, running a large model locally on hardware you already own, fully offline.
• What is ready now: ten production-ready studios open in the beta, out of sixty-three on the roadmap, with the other fifty-three in active development.
• The guarantee: nothing leaves your building, and every action is sealed under post-quantum cryptography to a tamper-evident audit record you can check offline.
• How to join: probe your hardware in the browser to see what it can run, then apply for an access key bound to your machine. Free of charge for the first cohort of beta testers.

What the beta actually is

The Mickai beta is a closed beta, which means you apply and we let you in, rather than handing the whole system to anyone who clicks a link. That is deliberate. Sovereign software that runs on your own hardware and holds your own data deserves a careful first cohort, not a free-for-all.

The path in is simple. You visit the beta page and probe your hardware directly in the browser. The probe reads what your machine can do and tells you which Mickai models it can run, honestly, before you commit to anything. If your hardware is a fit, you apply for an access key bound to that machine, and you receive a download matched to it. From that point on the software runs locally, offline, on hardware you own.

It is free of charge for the first cohort of beta testers. There is no cost for the models or the assistant during the beta. We are not selling anything on this page. We are opening the door and asking a first group of organisations to walk through it with us.

Poros: the engine that makes it possible

None of this works without an engine that can run serious AI where the data already lives. That engine is Poros.

Poros runs a large sovereign model on a single workstation-class GPU, the kind of machine an organisation can put under a desk or in a locked cabinet in its own building. No datacentre. No cloud tenancy. No shared infrastructure you do not control. And because real hardware is never tidy, Poros adapts to whatever it finds, from the newest cards down to older ones and plain CPU. The person deploying it runs the installer, and Poros works out the rest.

Offline has to mean offline, or it means nothing. Poros is a loopback-only component: it refuses any outbound network connection by design, with no telemetry channel and no quiet check-in. That is a strong claim, so it does not rest on our word alone. An independent adversarial review went at the engine specifically to break out of that boundary, and the egress guard held. For a security leader, that is the difference between a marketing promise and something you can put in front of a board.

Own it, run it offline, prove it

Those three phrases are not a slogan. They are the three things the beta actually delivers, and each one is precise.

Own it. The software runs on hardware you already own, inside your own building, under your own control. There is no cloud tenancy to rent and no third party holding your data as the price of getting useful work done. The intelligence comes to you.

Run it offline. Nothing leaves. Poros is loopback-only and the studios sit on top of it, so sensitive material never has to become someone else's network traffic. The engine simply has no way to reach the outside world, and, as above, that boundary was tested by people paid to break it.

Prove it. Every session seals the engine's identity and the model's provenance into a post-quantum-signed audit record. Mickai calls it the OAR. It is tamper-evident: if anyone alters the record, that shows, and a verifier can confirm the live chain against a signed checkpoint held where an attacker cannot rewrite it, with no internet required. We are careful with the word. The record is tamper-evident and independently verifiable offline, not unbreakable. What it rules out is silent, unnoticed alteration, so when someone entitled to ask wants to know which model ran, on which machine, for a given action, you have a signed record rather than a shrug.

Ten studios, one operating system

A studio is a working surface for a real job: drafting documents and spreadsheets, analytics over your own governed data, compliance and audit work, meetings and email, help desk, security operations, and more. In the Mickai SIOS every studio runs on the same Poros engine and seals its work to the same audit record, so the guarantees are consistent across the whole system rather than bolted on per feature.

Ten of those studios are production-ready, and those ten are what open in the beta. They are part of a larger operating system of sixty-three studios in total, with the other fifty-three in active development. One engine underneath, one audit record throughout, a growing set of surfaces on top.

Why we are being precise about the stage

It would be easy to describe a sixty-three-studio operating system as if all of it shipped today. It does not, and we will not pretend otherwise. Here is the honest shape of it.

What is real now is the engine and the ten production-ready studios. Poros is built. It has been independently reviewed. It runs a large model on a workstation GPU right now, sealed to the audit record, offline. The remaining fifty-three studios are in development, and they are coming, but they are not finished. The reason to take the whole vision seriously is that the hardest part, the part everyone else hand-waves, is the part that already works: running locally, refusing to leak, and proving itself. This is early access precisely because a first cohort helps us shape what comes next on a foundation that already stands.

Where this sits in what we have built

We hold 104 filed patent applications carrying 2,340 claims in total, all patent-pending at the UK Intellectual Property Office under named inventor Micky Irons. Our strategy is deliberate: we patent the mechanism, not the model. We do not seek protection on any particular AI system, and we do not disclose model lineage. We seek protection on how sovereignty, auditability, licensing and orchestration are enforced, which is exactly the terrain the beta puts in your hands.

Mickai is a registered UK trademark, and the operating system is built and held by Mickai LTD (Companies House 17166618). The value is in the enforcement architecture: the way an audit record is made tamper-evident, the way a licence is bound to hardware, the way the engine is kept local. That is what you get to try in the beta, not a demo of it.

How to get in

Everything starts on the beta page at mickai.co.uk/beta. Probe your hardware in the browser, see which Mickai models it can run, and apply for an access key bound to your machine. You receive a download matched to your hardware, and from there it runs offline on kit you own, audited by design. It is free for the first cohort. If you operate in a regulated sector, this is the shortest path to running serious AI on your most sensitive work without giving up custody of that work to do it.

Frequently asked questions

What is the Mickai beta?

It is early access, by application, to the Mickai Sovereign Intelligence Operating System. The beta opens ten production-ready studios that run entirely on your own hardware, offline, on top of our Poros inference engine, with every action sealed to a tamper-evident, post-quantum-signed audit record. It is a closed beta: you probe your hardware, apply for a machine-bound access key, and install a download matched to your machine.

What is Poros?

Poros is our sovereign inference engine, the part of the system that runs a large, capable AI model locally on hardware you control. It adapts to whatever GPU it finds, down to plain CPU, and it is loopback-only, so it refuses any outbound network connection by design. An independent adversarial review tested that boundary and the egress guard held.

Does anything leave my building when I use it?

No. Poros is a loopback-only component with no telemetry channel and no outbound connections, and the studios run on top of it. Your prompts, documents and data stay on the machine you own. That is the point of the design, and it was verified by an adversarial review paid to make it leak.

How do you prove what the AI actually did?

Every session is sealed into the Open Audit Record, a post-quantum-signed, tamper-evident ledger. A verifier can confirm the live chain against a signed checkpoint held off-box, with no internet required, so any silent alteration of the record shows. It is tamper-evident and independently verifiable offline, which we state precisely rather than calling it unbreakable.

What hardware do I need?

A workstation-class machine is enough. Poros runs a large model on a single workstation-class GPU and adapts down to older cards and CPU. Rather than publish a shopping list, we let you probe your own hardware in the browser on the beta page, which tells you which Mickai models your machine can run before you apply.

How many studios are in the beta, and how many are there in total?

Ten production-ready studios are open in the beta. They are part of a sixty-three-studio operating system, with the remaining fifty-three in active development. All of them run on the same Poros engine and seal to the same audit record.

Is it really free?

For the first cohort of beta testers, yes: there is no cost for the models or the assistant during the beta. It is a closed beta with limited places, so access is granted by application rather than open to everyone at once.

How do I apply?

Go to mickai.co.uk/beta, probe your hardware in the browser, and apply for an access key bound to your machine. If your hardware is a fit, you receive a download matched to it, and you can run a sovereign AI assistant offline on hardware you own.

You do not have to choose between capability and control any more. That was always a false trade. The Mickai beta is where we retire it, on your hardware, offline, and provable. Keep the intelligence. Keep the control.