Artificial intelligence
Analytical capability that never leaves your building.
Most commercial AI tools send your data to servers you do not control, in a jurisdiction you did not choose. We give institutions that need to keep sensitive material under their own control a working alternative: locally hosted, open-weight language models on infrastructure the client owns, so the capability is theirs outright and the data never leaves it. We can advise on this independently, or source, deploy and run the whole thing for you.
Requirement and data assessment
We start by establishing what the institution actually needs the tool to do, what data it would touch, and where it will save your analysts and officials the most time.
- Mapping of the specific analytic tasks a model would support, not a generic list of AI use cases
- Classification of the data involved, and how that shapes what can be built and how quickly
- Independent advice on the fastest route to real benefit, including a smaller first step where that gets you there sooner
Model and licence selection
Open-weight models carry different licences, and getting the terms right from the outset means the capability you build stays fully yours to use. We select and licence the right model for the client's own use and jurisdiction.
- Selection of the strongest available open-weight model for the task and the budget
- Licensing, export control and end-use checks cleared before deployment, not discovered after
- An agreed position on data residency and jurisdiction, settled up front
Deployment, hardware and supply
We size the hardware to the work and host it on infrastructure the institution owns or controls. Where the client wants it, we also source, procure and supply that hardware and software, and manage the whole system end to end.
- Infrastructure sized against realistic day-to-day use, not a vendor benchmark
- Deployment on client-owned or client-controlled infrastructure, so the capability answers to the client alone
- Sourcing, procurement and ongoing management of hardware and software, where the client wants the whole capability delivered and supported, not just advised on
Training, mentoring and handover
The same principle as every other service on this site. We train the people who will use it, hand over a system the institution can run and grow on its own, and stay on to support it for as long as it is wanted.
- Training for the analysts and officials who will use the tool day to day
- Ongoing support and review, for as long as the client wants it
- Handover documentation written for the institution that owns the system
Where we stand
- Advice first, full delivery if you want it.
- Our assessment of what AI can do for you is independent and not contingent on a sale. If you then want us to source, supply, deploy and manage the hardware and software as well, we do that too, under the same governance we apply to any other capability we build for a client.
- Nothing leaves the building.
- The model runs on infrastructure the institution owns or controls, answering to nobody but the client. No call back to a third party's servers, and no dependency on an internet connection staying up.
- Built to be checked, not just trusted.
- A locally hosted language model is a powerful drafting and pattern-finding tool. We design the checking step into the workflow from the start, so your people get the speed without losing the judgement that catches a wrong answer.
- Compliant by design.
- Model licences, export control and the end use of the institution concerned all shape what we deploy and where. We clear this before we propose anything, so the capability you get is one you can actually keep.
- A capability we are actively building out.
- We hold this to the same standard as everything else we do: named practitioners, a clear method, and straight answers about what stage a capability is at.
Built to be run by your own people, on your own terms, for as long as you need it.
Who this is for
Institutions handling material they need to keep under their own control, whether because of classification, data protection law, or a straightforward preference for owning their own data. This sits alongside the assessment and capability work described elsewhere on this site, and we can take it from a first conversation through to a system your own people run day to day.