AI and intelligence
Trust and deployment
Technology
The technology under our products is designed around one test: could a person check it? Every answer carries its evidence, every forecast is scored, every action has an owner, and all of it can run inside your walls.
Most document AI cuts files into fragments and answers from whichever one matches first. Entrada reads documents with their structure intact: pages, sections, clauses, tables, dates, versions and references.
Because it keeps that structure, it can cite the exact paragraph, see when a newer document changes an older one, and say “not stated” when the evidence is not there.
Kybern’s agentic layer is a digital team that works from the World Model: agents for evidence, reconciliation, assessment, scenarios, briefings and action preparation, coordinated by an orchestrator that checks every step.
Agents earn trust in stages, from shadow to limited to production. Any agent can be paused, every run can be replayed, and agents never receive approval authority. People sign off what matters.
Entrada’s intelligent-entry family brings AI to gates, perimeters and remote sites, running on edge hardware so video never has to leave the site.
Guardian AI, Entry Intelligence and RemoteGuard are in development. We will publish details as each one is ready.
Kybern keeps one living picture of people, assets, places and plans, and how they connect. Every other part of the system works from it.
It is a digital twin of the organisation rather than of a machine. It runs today across 19 demonstration worlds, from fleet readiness and grid operations to hospital capacity and space.
Underneath the World Model is a graph in which every fact keeps its source and is labelled by how it is known: known, observed, inferred, uncertain, missing or conflicting.
Conflicting sources are kept side by side until one is confirmed, and missing reports are treated as unknown, never as fine.
Real failures travel along dependencies nobody wrote down. Kybern maps how parts of an organisation depend on each other, finds hidden single points of failure, and shows what breaks if one part fails.
Its reasoning draws on many sciences at once, from queueing and network science to feedback, causal and resilience analysis, and shows which methods it used and why.
Your data and your models stay under your control. Our products run local models on your hardware, with no public AI service in the loop and no per-token bills.
Entrada AI-in-a-Box packages it on a Meru M128 appliance: NVIDIA Grace Blackwell GB10, 128 GB of unified memory and 4 TB of storage. Unpack it, connect it to your network, start asking.
Every product we sell into defence and government can run fully isolated. Entrada’s Sovereign mode uses local models only and approved sources only. RecruitOS ships as a secure, licence-controlled package that runs with no external network at all.
RecruitOS has already run a full recruitment for the Indian Army offline and on premises.
We build for the review after the decision. Every answer carries its evidence. Every forecast is written down before the outcome and scored afterwards, misses included. Every decision record shows the evidence, the options, the doubts and who decided, on what authority.
Attempts to act outside authority are refused by the platform itself, not just hidden in the interface, and we test for it in the open.
Deployment options
The same software at every point, from fully isolated to fully managed.
One decision, a few sources, a measured result.
