Bring documents, databases and connected systems into one governed surface, then ask questions of it in plain language — with row filters, column masks and a full audit record applied on every read, whichever way it is queried.
Sovereign AI for defence, government and enterprise
The complete AI platform that you can own.
Your data, your models, your applications and your agents, running on infrastructure you control.
- Sovereign deployment boundary
- Applications Conversation, consoles, data explorer, your own apps
- Application Framework Declare an app and the platform builds it
- Agent² Software Factory Agents design, build, test and ship it
- Atom Lakehouse, search, analytics, governance
- Anima Model serving, inference, GPU fleet
- Platform Foundation Identity · runtime · realtime · workflows · database · observability
- Apex Hardware, from a 6-litre micro server to full racks
- Cloud Mesh Networking, DNS, edge routing
- Managed sovereign cloud · Private cloud · On-premises · Edge · Air-gapped
This film requires JavaScript. The captions describe what is shown.
What organisations do with it.
Serve open models on your own accelerators through a drop-in OpenAI-compatible API, so existing tools and SDKs work unchanged and no prompt or document ever leaves your boundary.
Put an autonomous agent workforce behind your backlog. Specifications, designs, implementation, review, testing and deployment run as one gated pipeline that reports what it shipped and what it proved.
Deploy inference next to the data — a remote site, a vessel, a backpack — and keep serving when the link to the core is down, reconciling automatically when it returns.
One platform, the whole stack.
Most organisations assemble AI capability from a dozen vendors and lose control of it in the process. Agentycs is a single, self-contained system — from the accelerators up to the applications — that you can own outright.
Atom
A per-tenant lakehouse that unifies documents, structured data and external systems into one queryable surface.
Hybrid keyword and semantic search, full SQL over open columnar and vector storage, federated queries into databases you already run, natural-language questions answered with citations, and on-demand forecasting and anomaly detection.
Data and intelligence →Anima
One interface to every model — self-hosted on your accelerators, run at remote sites, or routed to approved external providers.
Text, embeddings, speech and vision under one serving plane. Models are placed automatically on hardware that can run them, shared across tenants with hard isolation, and served from your own model store so inference keeps working with no internet at all.
Models and inference →Application Framework
Declare an application and the platform generates, packages, signs and runs it.
Compose production blocks — forms, dashboards, review queues, document libraries, data explorers, embedded agents — from a typed SDK, a visual builder, a browser IDE, or a plain-language brief handed to an agent.
Application framework →Agent² Software Factory
An autonomous software pipeline: intent in, working and observed software out.
Agents perform the design, architecture, planning, implementation and delivery. Every stage is gated by automated verification, so quality is a property of the pipeline rather than of any one agent. People set direction and approve.
AI software delivery →Apex
Purpose-built AI servers, tuned as one system with the platform that runs on them — from a six-litre micro server up to full racks.
Vertically integrated with the platform for maximum throughput per watt, with portable, edge and air-gapped configurations. Bring your own hardware or virtualised cloud infrastructure instead if that suits you better.
AI infrastructure →Cloud Mesh
Sovereign, dependency-free web infrastructure: ingress, load-balanced DNS and zero-trust networking.
Traffic ingress with observability, firewalling and DDoS protection; geographic failover and latency-based routing; and authenticated site-to-site, agent-to-site and agent-to-agent networking with policy enforcement on every flow.
Networking and edge →Platform Foundation
The always-on substrate every application runs on, shared by all tenants under strict isolation.
Federated identity and one centralised authorisation decision, real-time live queries driven by change data capture, durable workflows that survive restarts, a horizontally scalable database, and observability shipped alongside every feature.
Identity, runtime and operations →Built for the organisations that cannot hand their data to someone else.
Different mandates, the same underlying requirement: capability that works without the data leaving.
Defence and intelligence
Classification boundaries, and sites that lose their link.
Sector context →Government and public sector
Residency, accountability for citizen data, and a credible exit.
Sector context →Enterprise
Model risk, provenance, and answers you can reproduce later.
Sector context →Software that builds itself, under supervision.
The Agent² Software Factory is how we build Agentycs. It is the same machinery we give customers: people express outcomes and grant approval, and agents do the design, architecture, planning, implementation and delivery end to end.
Production signals feed straight back in: regressions, defects and drift become new work the pipeline plans and fixes without anyone filing a ticket.
How the factory works-
Intent
Product owners describe outcomes in plain language.
-
Design
Competing approaches proposed, then scored by an adversarial review panel.
-
Plan
A dependency-ordered plan, structurally validated on every write.
-
Build
A fleet of coding agents working in isolated workspaces in parallel.
-
Prove
Static checks, tests, security review and live tests against real deployments.
-
Ship
GitOps promotion with convergence waits, canary rollout and instant rollback.
Runs where your mandate says it has to.
The same platform, the same interfaces, the same operational model — whether that is a managed sovereign region, your own datacentre, a six-litre micro server at a remote site, or a network with no route to the internet at all.
Managed sovereign cloud
We operate it, in your jurisdiction.
Private cloud
Your cloud account, your control plane.
On-premises
Your datacentre, your hardware.
Air-gapped
Fully disconnected, fully functional.
Agentycs-in-a-Box
A six-litre micro server that runs the whole platform on 400W.
Edge and remote sites
Inference next to the sensors.
Make AI operable, governed, and deployable under real-world constraints.
Capability is not enough on its own. What matters is who can direct it, inspect it, change it, and remain responsible for what it does.
Our mission and principlesBecome the leading secure, sovereign agentic AI and data analytics company, delivering real operational capability.
-
Control must be real
Ownership means the ability to inspect, change and withdraw — not a contractual promise about someone else's system.
-
Security is built into the path
Authority, isolation and policy are decided in the request path and fail closed, rather than being added as a review at the end.
-
People own the consequences
Automation raises human capability; it never removes the direction, exceptions or accountability that people are responsible for.
-
Reality is the test
Software is judged against real deployments under real conditions, not against a demonstration built to pass.
-
Keep the path open
Open formats, standard protocols and documented interfaces, so the way out stays as clear as the way in.
-
Sovereignty is a choice you keep making
Where data rests, which models serve it and who can reach it stay yours to decide, at every layer and on any given day.
Built to be inspected.
Sovereignty is worth nothing if you cannot verify it. Every access decision resolves through one audited path, and everything that runs can be traced back to what produced it.
One authorisation decision
Role, capability and attribute rules consolidate into a single audited verdict for every privileged operation, so there is one place to reason about access.
Governance that cannot be bypassed
Row filters, column masks and deny decisions are evaluated identically whether a request arrives over REST, SQL, a Postgres client, natural language or a graph query.
Isolation you can point at
Tenant scope is derived server-side from validated credentials, never from client input, and extends to per-tenant databases, search and accelerators.
Provenance on everything that runs
Applications are content-addressed, signed and attested, with a bill of materials, and are refused at execution if their trust level does not permit the runtime they were dispatched to.
Own the intelligence your organisation runs on.
Tell us what you need to build, where it has to run, and what you cannot let leave your boundary.