Flowing Mind is a UK AI consultancy for defence, central government and regulated organisations. We design, build and govern AI systems on infrastructure you control, up to and including fully air-gapped, so your data, your models and your audit evidence stay inside your boundary.
Fixed fees, published openly. No lock-in, no proprietary runtime, no per-seat licence on work you paid us to build.
The blocker
The pilot works. Someone shows it to the board. Then it reaches information governance, the DPO, or a procurement panel, and the questions start: where does the data actually go, who trains on it, what happens when the vendor changes terms, and can you show us the evidence?
Those questions are not obstruction. They are the job. But they are almost never answerable about a system that was architected around someone else's API, and by then the money is spent.
We build the answer in from the first week, because for our clients it is the constraint that decides everything else.
What we do
Most clients start with an assessment and stop there until they are convinced. That is the intended design: it should be cheap to find out we are not the right fit.
A structured review of your data, systems, controls and people, ending in a ranked list of AI opportunities with costs, risks and a build order.
Read more 02Open-weight models on your own infrastructure or in a UK region you control, up to fully air-gapped on NVIDIA DGX or HGX B300, hardened to DISA STIG and CIS Level 2.
Read more 03Retrieval over your documents, agents that complete real work, and automation wired into the systems you already run, built to be handed over, not rented back to you.
Read more 04The policies, registers and evidence trail that let you say yes to AI, and prove to a regulator, auditor or procurement panel exactly how it is controlled.
Read moreStraight talk
The word gets used to mean anything from genuinely air-gapped to "our US parent company promises not to look". Here is the honest taxonomy we use with clients, and what each tier actually buys you.
Your hardware, your network, your control plane.
Open-weight models on machines you own, up to and including a network with no internet egress at all. Every dependency mirrored inside the boundary: images, packages, drivers, licensing, model weights. The strongest position, and the highest operational burden.
Right for: defence, segregated government networks, patient records, privileged material.
A UK region, on a contract you hold.
Dedicated model instances inside a UK region of a cloud you already have a data processing agreement with. Data stays in-country and out of shared inference pools, with far less operational overhead than running your own accelerators.
Right for: most regulated production workloads, where the barrier is residency rather than physical control.
A frontier model under enterprise terms.
Sometimes the best model is a hosted one and the data is not that sensitive. That is a legitimate answer, provided you have written zero-retention and no-training terms, a documented residency position, and you have classified the data honestly.
Right for: public information, marketing, internal drafting, low-classification workloads.
We will happily recommend Tier 3 where it fits. A consultancy that always concludes you need the most expensive option is not giving you advice, it is quoting.
Who we work with
Inference on hardware inside your own perimeter, with no egress path at all. Hosts hardened to STIG and CIS Level 2, and the Secure by Design artefacts your accreditor expects.
MOD Secure by Design, JSP 604 and Def Stan 05-138
Departmental AI that survives GovAssure, Secure by Design and ATRS scrutiny, running on infrastructure the department controls rather than in someone else’s tenancy.
GovAssure against the NCSC CAF, and Secure by Design
Model risk documentation, explainability and consumer outcomes scrutiny, with the client data that makes AI useful never leaving a controlled environment.
SS1/23 model risk expectations and Consumer Duty evidence
Patient data cannot be posted to a US API. We build inside your boundary and produce the DPIA and information governance evidence that gets a project approved.
DSPT, DTAC and information governance sign-off
Privilege and client confidentiality make hosted AI a hard sell. Private retrieval over your matter files keeps the material inside the firm.
Legal professional privilege and SRA supervision duties
Councils, NDPBs and wider public services. Transparency obligations and public scrutiny raise the evidence bar, so we build to the standards your assurance team already answers to.
ATRS entries, AIME self-assessment and procurement scrutiny
Delivered work
Central government
An on-premise AI platform on NVIDIA DGX B300, running with no internet egress at all, hardened to DISA STIG and CIS Benchmarks Level 2, and mapped onto the NCSC Cyber Assessment Framework while it was being built rather than afterwards.
Defence
An NVIDIA HGX B300 platform inside a defence perimeter with no egress path of any kind, delivered under MOD Secure by Design with a live security case, DISA STIG and CIS Level 2 hardening, and every artefact provenance-checked at the transfer boundary.
How we work
Every engagement starts by drawing the boundary your data must not cross, then designing only within it. Where a hosted model is genuinely the right call, we say so, and document the residency and retention terms in writing.
No proprietary runtime, no per-seat licence on our own middleware, no hostage architecture. Open weights and open standards wherever they do the job, and a repository with tests and documentation that your team can run without us.
Regulated buying decisions are not won with demos. Every build ships with an evaluation set, measured results, a risk register entry and the paperwork an auditor or procurement panel will ask for.
A good proportion of proposed AI projects should not be built. Where the answer is a database query, a fixed rule or better process design, that is what we recommend, and it costs you an assessment, not a year.
Insights
Regulation
There is still no UK AI Act, but that does not mean no obligations. What actually binds UK organisations using AI in 2026, and what to put in place.
ReadSovereignty
A working definition of sovereign AI, the four questions that separate real data sovereignty from a residency claim, and where each deployment tier fits.
ReadBuying AI
Real UK market rates for AI consulting in 2026: day rates, fixed-fee audits, project costs and retainers, plus where the money usually gets wasted.
ReadCommon questions
A consultancy that designs AI systems around a data boundary rather than around a vendor. In practice that means open-weight models running on infrastructure you control (your own servers, or a UK region of a cloud you hold the contract for) so that prompts, documents and outputs stay inside your legal jurisdiction and never become training data for someone else.
No. There are three defensible positions: fully self-hosted on your own hardware, private deployment into a UK cloud region you control, or a hosted model under enterprise terms with UK data residency and no training on your inputs. Each is appropriate for different data classes. We help you decide which of your workloads belongs in which tier, and document the reasoning so it survives audit.
Our AI readiness assessment starts at £4,500 for a fixed-fee two-to-three week engagement. Governance work starts at £7,500, private deployment at £18,000 and full implementation at £25,000. We publish these figures because vague pricing wastes everyone’s first meeting.
Both, deliberately. Advice that has never had to survive production tends to be wrong, and building without governance tends to get blocked at the assurance gate. We assess, build, and produce the compliance evidence, then hand the whole thing over to your team.
Yes, and we have. It means mirroring every dependency inside the boundary: container images pinned by digest, operating system and language packages, GPU drivers and firmware, licensing served locally, model weights imported as hashed artefacts, and vulnerability feeds brought in on a defined cadence. We have delivered air-gapped platforms on NVIDIA DGX B300 and HGX B300 for central government and defence. The hard part is never the first deployment, it is sustainment, so we design that first.
Hosts, container runtimes and orchestration are built from DISA STIG baselines reconciled against CIS Benchmarks Level 2, in code, and scanned before handover, with every deviation documented alongside its compensating control. Platforms are mapped onto the NCSC Cyber Assessment Framework and the NCSC Guidelines for Secure AI System Development, and traced to NIST SP 800-53 and the NIST AI Risk Management Framework where a partner or supply chain expects that vocabulary.
A 30-minute call, no pitch deck. Tell us what you are trying to do and we will tell you whether AI is the right tool, what it would take, and what it would cost, or that you should not bother.