A reasoning engine that holds AI to your business rules
Language models produce the most probable answer. In procurement, compliance, safety and clinical work, a probable answer that breaks a rule is a liability. The Archisus platform encodes your rules as a formal ontology and checks every AI output against them before it leaves the system, with a record of which rule decided what.
The architecture
Three layers, one reasoning machine
The platform turns the rules, relationships and data of your organisation into something an AI system must respect, not just read.
The constitution
Ontology layer
An OWL knowledge graph that models your entities, rules, relationships and constraints as formal logic: a purchase order requires an approved supplier; an approved supplier requires a valid certificate; a certificate is valid only before its expiry date. It is not a document index or a vector store, it is a model of how your domain works.
The engine
Reasoning engine on your graph store
The Archisus engine runs OWL 2 inference and consistency checks, SHACL validation and SHACL rules, and query-based rules in SPARQL, or in Cypher through neosemantics on Neo4j. Outputs that break a rule are blocked, logged with the rule that failed, and escalated.
The reach
Federated data layer
Semantic adapters connect SAP, SCADA historians, ERPs, maintenance systems and document stores, and query each one in place. Data is not copied to a new platform: only query results move, inside your security perimeter.
Graph stores
Runs on the graph store you already use
The knowledge graph is persisted in your existing store. The reasoning runs in the Archisus engine, not in the database, so the same rules and the same validation apply whichever store holds the graph.
AWS Neptune
Your graph stays in your AWS account. The engine queries it with SPARQL and applies OWL 2 inference and SHACL validation on top.
Ontotext GraphDB
For organisations already standardised on GraphDB, the ontology and the validation shapes run against your existing repositories.
Neo4j
On Neo4j, the ontology is mapped through the neosemantics (n10s) plugin and query-based rules run in Cypher.
Digital Experts
How a Digital Expert works
A Digital Expert is an AI agent for one business process, such as procurement compliance or regulatory reporting. It runs continuously inside a limited set of permitted actions, in four steps:
Perceive
Takes in live data from connected systems: ERP transactions, sensor readings, documents, workflow changes and API events.
Reason
Queries the knowledge graph to find which rules apply to that data and what they require.
Act
Carries out the permitted action in your systems: raises a record, starts an approval, drafts a filing or issues a work order.
Validate
Every proposed output is checked against the SHACL constraints before release. Anything that breaks a rule is blocked, logged with the rule and escalated to a person.
What it looks like
An answer you can audit
Procurement Digital Expert
“Rule 47.3: Supplier ACME-12 blocked. ISO 9001 certificate expired 3 days ago. PO-2024-887 halted pending re-certification. Compliant alternative: Supplier ACME-07 (certificate valid until 2027-03).”
Compared
Ontology reasoning, RAG and a plain LLM
Retrieval (RAG) and language models are good at finding and writing text. Governed decisions need something they do not provide on their own.
| Plain LLM | RAG | Archisus | |
|---|---|---|---|
| Where the rules live | In the prompt, as suggestions | In retrieved documents, as context | In a version-controlled ontology, as formal logic |
| Can an output break a rule? | Yes, on edge cases | Yes: the model reads the rules but is not bound by them | It is checked before release; violations are blocked and escalated |
| Relationships between facts | Not modelled | Similar text, not relationships | Modelled and reasoned over (e.g. a clause superseded by an amendment) |
| Audit trail | None | Source citations | Rule, data and decision recorded at the moment of decision |
| When a policy changes | Rewrite prompts | Re-index documents | Update the ontology once; every Digital Expert follows it |
The approaches combine: Archisus can use document retrieval inside an ontology-governed pipeline, and it works with the language model you already use.
Questions
Frequently asked questions
Does Archisus replace our language model?
No. The ontology layer is model-agnostic. It governs whichever model you use, including one you already run, and the checks still apply when you change or upgrade that model.
Do we need to migrate our data?
No. The federated layer queries your systems in place. No operational data is copied to Archisus, and the models run on-premise or in your private cloud.
Which graph databases are supported?
AWS Neptune, Ontotext GraphDB and Neo4j. The graph is stored in the one you already use; the reasoning runs in the Archisus engine.
What kind of reasoning does the engine do?
OWL 2 inference including consistency checks, SHACL validation and SHACL rules, and query-based rules in SPARQL, or in Cypher through neosemantics on Neo4j.
What happens when a rule changes?
The ontology is updated and the change applies to every Digital Expert that uses it, with no retraining and no prompt rewriting.
Do we need clean data first?
No. Data is only "clean" relative to the rules it has to meet. The DNA Audit defines those rules and shows which data issues actually matter for the process you want to start with.
The DNA Audit Is Your First Step. And Your Lowest Risk.
In two weeks we map every critical reasoning dependency, broken data link, and governance gap in your current AI or data workflow. You leave with a draft ontology model, a prioritised use case roadmap, and a clear decision on whether - and how - to proceed. No commitment beyond the audit.
Knowledge architecture map
Every logic dependency, data relationship, and constraint your AI needs to reason correctly - visualised and documented.
Gap & risk report
Broken data links, contradictory rules, and governance gaps that are invisible today - surfaced and prioritised by business impact.
Draft ontology & pilot roadmap
A working ontology model for your highest-priority domain, plus a ranked shortlist of pilot use cases with estimated ROI and delivery timeline.
Book Your DNA Audit
2-week engagement · Fixed scope · No ongoing commitment required
Message Sent!
We will be in touch shortly.
