跳到主要内容
联系
Installed architecture

Your AI architecture, installed in your cloud

We install the components you govern AI with — model control, tools, memory, flows, observability — inside your own subscription and under your own keys. We hand them over working, and you take them from there.

Talk to the team Configure mine
Components
9 · one mandatory
Clouds
Azure · AWS · GCP · on-prem
You start with
One team, one cloud
Your cloud 9
Model control Mandatory
API control
Tool hub
Corporate memory
Flow engine
Observability
Model selection
Cybersecurity
AI engineering

All inside your subscription. Your keys, your network, your control.

Where you start

You don't buy this whole

You start with one team and one cloud, and it grows when there is a reason. Every step closes with scope and timeframe before the next one opens.

  1. Assessment

    You know you want to do something but not whether it is feasible, what data it needs, or what it takes. This answers that.

    Credited in full if you continue.

  2. First installation

    Model control for one team, in your cloud and under your keys. Sized to land without opening a long procurement process.

  3. Expansion

    Memory, flows, observability, security. Each one is added when you need it, not before.

  4. What runs on top

    Your corpus, your flows and your cases in production. Sized by the assessment in the first step.

The bays

Components, not an open-ended project

Only one piece is mandatory: model control. Without it there is nothing to govern. The rest is added when it is needed — or you bring it yourself, if you already run it.

  • Model control

    Mandatory

    Keys · Quotas · Expiry · Rate limits

    The mandatory step in every call to a model: keys per team, a spending ceiling, and a record of who asked what.

  • API control

    Keys · Quotas · Expiry · Rate limits

    The same control over keys, quotas and expiry that you hold over models, applied to your own APIs.

  • Tool hub

    MCP · Registry · Versioning

    One place to register, version and expose the tools your agents consume, instead of scattering them across projects.

  • Corporate memory

    Hybrid retrieval · Embeddings · Vectors

    The tiered retrieval chain installed and verified, ready for you to load your documentation. Not embed-and-search: retrieval iterates before it answers.

  • Flow engine

    LangGraph · LangChain

    The engine your flows will run on: with state, retries and traces. The flows themselves are built afterwards, on top.

  • Observability

    Traces · Cost · Failures

    Knowing what was asked, what the model answered, what it cost and why it failed. Without this, any discussion about quality is an opinion.

  • Model selection

    Quality · Cost per case · Latency · Fallback

    Which model serves each case and at what cost, measured on your data rather than public benchmarks. And which one takes over if the primary goes down.

  • Cybersecurity

    Cloud posture · Attack paths · Exposure

    Knowing which of everything you have deployed is actually reachable by an attacker, and how they would get there.

  • AI engineering

    Spec-driven method · Agent and flow archetypes

    Two things installed in your repositories: the method applications are built and maintained with, and the archetypes agents and flows are built from. Coaching is separate.

Configure it

Tick what you need and see what comes out

Two decisions — which cloud and which bays — and you have the whole scope in front of you: what has to be ready, and how each piece will be known to be done. Without leaving an email.

Where do we install it?

One or more

Which bays do you light up?

Model control is always in

It installs in your subscription, and you can run it without us

There is no platform of ours in the middle, no usage fee, and no operating contract this depends on to keep working. Everything we install stays as code in your repositories and as infrastructure in your cloud. If tomorrow you want to carry on alone, you carry on alone.

  • Your cloud account and your encryption keys
  • The code, in your repositories
  • No usage fee, no platform in between
If it isn't clear yet

Not sure what you need?

Every case is particular, and that shouldn't turn into a blank cheque. It is assessed with scope and timeframe fixed before anything starts.

Case assessment

You know you want to do something with AI but not whether it is feasible, what data it needs, or what it takes to get it into production.

  • Technical feasibility of the case, and why
  • What data is needed and what state it has to be in
  • Effort band to take it to production
  • The metric you will know it works by

Credited in full against the installation if you continue. With control and traces already in place, assessing a new case is a bounded exercise, not a project.

Let's talk about your cloud, not about the platform

Tell us which team will use it and which cloud you are on. The first step comes out of that.

Talk to the team