Enterprise
For the firms that were never allowed to use the cloud.
Not privacy-conscious. Not cautious. Actually prohibited: privileged case files, patient records, unpublished results, work under NDA, anything inside a regulator's perimeter. If that is you, most of the AI market has been selling you something you cannot legally buy.
Why this works when cloud does not
You cannot assure what you do not control. You can assure a machine in your own rack.
The processor problem disappears
Cloud AI makes a third party a processor of your most sensitive material. That brings transfer assessments, sub-processor chains and contractual terms they can revise. Running the model on your own hardware removes the processor from the arrangement rather than papering over it.
The control is physical, not contractual
"They say they do not train on it" is a promise. A network cable you can pull out is a property of the system. For an auditor those are not remotely the same class of evidence, and only one of them survives a change of vendor policy.
This is also why we do not offer a cloud fallback for large models. The moment a machine can burst to someone else's servers, every assurance above becomes conditional, and conditional is what disqualified you in the first place.
What Enterprise adds
One machine is a tool. This is a deployment.
The hardware is the same class as Professional. What you are buying is everything a regulated organisation needs around it.
Access
Per-user accounts and roles
Named users rather than a shared machine. Roles decide who can change models, who can reach which document sets, and who can only ask questions.
Audit
A log you can hand to a regulator
Who asked what, when, against which model and which documents. Exportable, tamper-evident, and retained on your terms because it never leaves your building.
Isolation
The switch, and the policy behind it
The physical network switch is the artefact. The deployment around it enforces which subnets the machine may reach at all, so the control survives someone flipping it back.
Data
Your document sets, segmented
Matter-level or department-level separation, so the model answering a question in one team cannot retrieve from another. This is usually the first thing compliance asks about.
Scale
Grow by adding machines, not by migrating
You are never buying the ceiling on day one, and never throwing the first machine away to get past it.
One machine, one team
128 GB. A department running models to about 120B, with its own document set.
Stacked for capacity
256 GB pooled, which opens the largest open models available. Capacity, not extra speed.
Multiple nodes
Independent machines per department or per matter, each serving its own users at full speed.
Worth being precise, because others are not: pooling memory across machines lets you run models that fit on neither, at roughly the speed of one. It buys capacity and headroom, not a faster single answer. If you need both, you want more nodes rather than a bigger pool.
Procurement
The questions your legal team will actually ask
These are what stall deployments, not the specifications. Our answers are here so you can forward this page instead of scheduling a call to extract them.
Can you complete our security questionnaire and DPIA?
Yes. Because inference happens on hardware you own, most of a standard DPIA collapses: there is no third-party processor, no international transfer and no sub-processor list. We complete the document with you rather than sending a PDF.
Will you sign our DPA and confidentiality terms?
Yes, and we will point out that a data processing agreement for the AI component is largely redundant here. We sign it anyway when procurement needs it on file.
What happens when the vendor disappears?
The machine keeps working. Models are on the drive, the weights are open, and nothing phones home for a licence check. That is the honest answer to continuity risk and it is not one a cloud provider can give.
Who is accountable when it breaks?
A named SetupMyAI engineer, on a support agreement with a response time in writing. Not a community forum and not a ticket queue.
Can we pilot before committing?
Yes. A single machine in one department, on the same terms, before any wider rollout. We would rather you proved it internally than took our word for it.
Deployment
An engineer arrives and does not leave until it works.
SetupMyAI handles the whole installation: hardware, models, your document sets, access control, integration with the systems you already run, and training for the people who have to use it. On-site across the UK and UAE, remotely everywhere else.
Typically two to six weeks from first call to a working deployment, depending on how many systems it has to touch and how long your security review takes.
Pricing is quoted per deployment. It depends on node count, document volume and integration scope, and we will not pretend a list price would be meaningful.
Send us the constraint, not the requirement.
Tell us what your material is and what you are not permitted to do with it. That is usually enough for us to say whether this is the right answer, and we will tell you if it is not.