How OIS is built — reasoning layers, engines and principles.
OIS is a layered, governed system: connectors observe, specialised engines reason, Evidentia proves, OIS Studio presents, and people decide. This is the machinery underneath the model.
The layered architecture
Each layer has one job. Nothing is automated end-to-end; a person approves the important steps.
The reasoning layer
Underneath OIS is a system of specialised intelligences — engines — each built for one job: understanding, memory, relationships, patterns, causes, foresight, proof, governance, recommendation and explanation. OIS routes each task to the engine built for it, so every step stays reviewable, instead of one model giving one undifferentiated answer.
The engines
Cognita
Understands documents, claims, entities and context.
Discovery
Finds and compares relevant sources.
Memoria
Keeps structured memory and history.
Nexus
Maps entities, identities and relationships.
Structura
Detects recurring patterns and anomalies.
Praedicta
Highlights possible developments — not deterministic prediction.
Drivers
Identifies the causes and forces behind an event.
Sentiment
Reads perception, stance and legitimacy from evidence.
Insight
Synthesises signals and evidence into a usable read.
Evidentia
Tracks proof, provenance, gaps and verification.
Consilium
Proposes next actions for human approval.
Ethica
Checks governance, sensitivity and boundaries.
Vox
Explains in the audience’s language.
Orchestria
Coordinates the workflow and approval gates.
The engines are running — not a diagram
This reasoning layer is live code, operated continuously by Merlin, the OIS visible brain. Seven of the engines run on a schedule against real information and persist their own inspectable output. Across the ecosystem they help maintain a knowledge graph of over 15,000 entities and 13,000 relationships drawn from more than 400,000 tracked events — every signal traceable to its evidence, every important step gated by a person. This is proven in real environments; it is evidence, not a universal guarantee for every organisation.
Implementation principles
- Human-in-the-loop — AI proposes, people decide
- Read-only by default — connectors do not change your systems
- Evidence for every signal — traceable to its source
- Honest-State — what is not measured is shown as not measured
- No training on your data by default
- One governed path for AI — not scattered, ungoverned calls
- Copy-and-derive — source documents are never modified
- Every write gated and audited
Built to be governed
Nothing is dispatched, published, invoiced or decided automatically. Every important action passes a human-approval gate and leaves an audit trail. Governance is part of the architecture, not an afterthought.
See your organisation clearly.
Start with a read-only OISA audit — what is known, what is missing and what to decide.