An ontology for what AI knows. A harness for what AI may do.
Your business logic and your software logic, in technical terms: an ontology that describes both, and a harness that governs every step AI takes on them.
Two concepts, each across both halves.
One description of your business and your systems, read by people and AI alike.
Your business logic
- Your terms.What a customer, a deal or a stage means in your business.
- Your rules.What counts, what is allowed, what needs a sign-off.
- Your processes.The steps work goes through, and who decides.
- Your decisions.What the business decided and why, kept so it can be found again.
Your software logic
- Your systems.Which software holds what.
- Your data.How records are structured and how they relate.
- Your interfaces.How data moves between your systems.
- Your documents.Files, mail and spreadsheets, where much of the knowledge sits.
AI proposes. Tested code decides what is written and what reaches a person.
Reading changes nothing
AI reads the records of your own installation to research, compare and answer, through access that cannot write. A question in plain words becomes an exact query, checked before it runs.
Proposals first
AI has no direct way into your records. Every change it suggests arrives as a proposal with its source, and nothing reaches a record any other way.
Review Spine
Two questions decide each proposal: is it correct, and does it matter? A reviewing agent runs the checks behind them. Your rules accept the clear cases, people decide the rest.
Audit trail
Every decision is kept with its reason and the version of the rules that applied. Any single change can be undone.
Where your data lives, and under which law.
- GDPR
- Data processed in the EU, in your own installation, when you use EU-hosted models.
- US CLOUD Act
- In the sovereign setup, no US-incorporated provider is in the path.
- EU AI Act
- Audit trail and human oversight are part of the harness, built with the Act’s record-keeping rules in mind.
- Open-weight models
- Everyday work runs on models you could host yourself.
Trust by architecture, not by promise.
Get in touch