Ontology

Everything in AlpOS — analytics queries, AI retrieval, agentic workflows, the audit log — resolves against the same governed object model. We call that model the ontology. This page explains why, and what an ontology looks like in practice.

Why an ontology at all

The alternative is that every team builds its own view of the data: the analytics team writes joins, the AI team writes prompts that wrap those joins, the workflow team writes a third set of names for the same objects. Within six months, the platform is shipping three different answers to “who is this customer?” — and no auditor can reconstruct which definition produced any specific decision.

The ontology fixes this by making the object model first-class. Customer, Account, Transaction, Vessel, Incident — whichever objects matter for the institution — are defined once, with explicit fields, explicit relationships, explicit policy on who can read which attribute. Every component of the platform consumes those definitions. The AI cannot retrieve a field the analytics layer cannot see; the workflow engine cannot dispatch on a relationship the ontology does not declare.

What lives in a Davion ontology

For a typical engagement we model:

• Entities — the real-world things the institution cares about. People, organisations, accounts, devices, assets, cases.
• Relationships — how entities connect. Beneficial-ownership graphs, account-holder links, sensor-to-asset associations.
• Events — discrete things that happen. Transactions, alerts, sensor readings, status changes.
• Policies — read/write/use rules attached to specific fields, relationships, or whole entity classes.

The ontology is versioned in source control alongside the rest of the deployment configuration. Schema changes go through review and ship as part of the signed-update channel.

How it is built

In every engagement, the ontology is built collaboratively with the institution’s domain experts during the pilot phase. We do not ship a packaged ontology that the customer adopts wholesale — every regulated institution we work with has too much organisation-specific shape for that to be honest. We do ship starting templates for the sectors we serve (financial services, defense and intelligence, government), which compress the modelling phase from months to weeks.

What the ontology is not

It is not a knowledge graph in the academic sense, and it is not a vendor “data fabric.” It is an operational schema that ships with policy, observability, and an audit log. Its job is to make the institution’s data ready for AI without losing the structure that made the data trustworthy in the first place.