Services · 01 - Flagship

AI that survives a security review.

Our flagship engagement: self-hostable, multi-tenant capable, governed, auditable from the first commit. We design and build the application - you get it running on your infrastructure, with the assurance evidence that answers the security questionnaire, and the source code in your repository.
01

Who this is for.

Regulated organisations that need AI in production, not another pilot that stalls at infosec. Banks. Insurers. NHS-facing teams. Public-sector buyers. Professional services firms whose clients won’t accept their data being processed in someone else’s tenant.

If a security questionnaire has ever ended a procurement, this is the engagement that prevents it next time.


02

What you get.

A working application on your infrastructure, with:

a.

Tenant data isolated at the database

PostgreSQL row-level security, with a cross-tenant test suite proving it table by table.

b.

Every AI interaction logged and governed

Which persona, which model version, which prompt, which output, which user, when.

c.

Documented architecture

Security model, data flow, model and persona governance, audit trail design.

d.

The Assurance Pack

A structured evidence bundle that answers the questions enterprise security and public-sector assessments actually ask.

e.

The source code, in your repository

No proprietary runtime, no captive SaaS.

We can operate it for you under Managed Run afterwards, or hand it over to your team cleanly with a runbook and an evaluation harness.

03

Why it’s different.

Isolation is enforced by PostgreSQL row-level security, one layer below the application, where a single bug in application code can’t quietly leak data across customers. We prove it with a cross-tenant test suite that runs against every table, on every change.

That single architectural choice changes the conversation with infosec, with procurement, and with the auditor. It’s why the engagement is called “Governed AI Delivery” rather than “AI build”. The governance is the deliverable, not an afterthought.

Why it passes a security review
04

How we deliver fast.

Three things, working together.

A hardened foundation underneath every project.

We start every engagement on top of Prototypical, our application foundation. It deploys into your own environment and supports multi-tenancy inside it where the application calls for it. Database-enforced isolation, audit backbone, durable job queue, provider-agnostic AI gateway: all there on day one. We build the parts that are actually about your problem, not the foundations beneath them.

Senior engineers with AI-assisted development.

Our engineers are senior, and they pair with frontier coding assistants every day. A working prototype in the first week. Production code in weeks, without the failure modes of foundations rushed under time pressure.

Governance designed in, not bolted on.

The audit log, the evaluation harness, the persona governance, the isolation tests: all scaffolded from the first commit. By the time you reach the security review, the answers already exist.


05

A typical engagement.

2-4 wksprototype sprint
6-14 wksto v1 in production
01 Discovery We sit with the people who’ll use the application and the people who’ll operate it. We agree the success metric. We surface the integration constraints and the data realities. 1 wk
02 Prototype Sprint A working version against real data and real users. Engineered well enough that if you say yes, we don’t throw it away. 2-4 wks
03 Design & Build Working software at the end of every iteration. Weekly demos. Nothing built that doesn’t connect to the success metric. 6-14 wks to v1
04 Hand-over or hand-off Take the system into your team with a runbook and an evaluation harness, or roll into Managed Run and we operate it for you. -
06

A typical stack.

The shape varies by problem. A common configuration:

Front-end

A modern web framework - the Prototypical default - or the front-end your team already runs, where that fits better.

Back-end

TypeScript or Python services, containerised, deployed on Azure, AWS, GCP or your own data centre.

Data plane

PostgreSQL with row-level security and pgvector for retrieval; object storage for unstructured content.

AI layer

Anthropic, Azure OpenAI, Google Vertex; open-weights options where cost, latency or sovereignty demand it.

Orchestration

OpenWeave for long-running, resumable agent work; lighter orchestration for simpler flows.

Persona serving

Persona Factory for governed, versioned personas, served over the Model Context Protocol.

Read about the platform underneath
07

Engagement shape.

Typical entry point

Prototype Sprint (2-4 weeks) or Design & Build (6-14 weeks).

Output

Working application in production on your infrastructure; cross-tenant test suite; Assurance Pack; source code yours.

Team

Product lead + designer + senior full-stack + AI engineer + DevOps.

Follow-on

Managed Run, capability transfer, or expansion to further applications.

What are you trying to put into production?

If you can describe the problem and the constraint, we can scope a prototype in two weeks.