Your Vocabulary.
Your Rules.
Domain-aware AI grounds LLMs in your organisation's definitions, policies and computed rule results. Answers use your terms, every input the model received is in the lineage, and your rules check the output before anything acts on it.
| trade | desk |
|---|---|
| T-88120 | Rates EU |
| T-88341 | FX Asia |
| T-88402 | Credit UK |
Proven in production
Knowing Terms.
Applying Rules.
Pasting a glossary into a prompt tells a model what exists. It doesn't tell it how your business applies it. Context is only half of the answer; domain-aware AI gives the model the rules too.
Documents
Become Data
Ingest PDFs, policies and notes into the Context Layer. Context concepts expose them as a queryable table, so unstructured knowledge sits in the ontology next to your structured data, with the same lineage.
Static Context
Pin specific notes by ID for stable references such as policies and glossaries.
Dynamic Context
Store a semantic query that returns the top matching notes and refreshes as the Context Layer changes.
Filter by Kind
Limit results to facts, document chunks, preferences or observations.
One Table Shape
Every context concept returns note id, kind, title, text, source, date and folder path.
Stop repeating your policies in every prompt. Keep business knowledge in one governed place.
Prompts Live
With Your Rules
An LLM concept is a prompt template that references other concepts by name. At run time Prometheux substitutes their results, calls the model and parses the response into the columns you declared. Your rules feed the model; the model's output can feed more rules.
Grounded by Reference
Write {{supplier_risk_notes}} and the model receives that concept's output, with the dependency tracked in lineage.
Typed Outputs
Declare output columns and types, so the response is structured data, not a paragraph to parse.
Model per Concept
Use a default provider or override the model, connection, temperature and token limits for each concept.
Rules Downstream
Vadalog rules can join against LLM results, so every model judgement is checked by your logic.
Get answers in your organisation's own terms. See exactly what the model was given. Check every model output against your rules.
From Documents to
Domain-Aware Answers
Ingest Knowledge
Add policies, PDFs and notes to the Context Layer alongside your connected sources.
Expose as Context
Pin the references that never change; set dynamic searches for the ones that do.
Write the Prompt
Create an LLM concept that references your context and rule outputs by name.
Check With Rules
Join the model's typed output back into Vadalog rules before anything acts on it.
AI That Knows
How You Work
Explained Credit Decisions
Banca Sella's system processes a client's full history with rules, then explains the outcome in plain language.
Read the case study →Supplier Risk Review
Combine supplier notes, dispute records and contract values, then let rules decide what gets escalated.
Policy-Aware Assistants
Pin internal policies as context so assistants answer with your definitions, not the internet's.
Peer-Reviewed Approach
Our research uses the provenance of rule-based reasoning to make LLMs domain-aware and explainable.
Browse the papers →“Generative AI without ontologies is an engine without a road. Prometheux builds the road.”
Marco RomeiDeputy CIO · Banca Sella
“The ability to query heterogeneous databases through a flexible ontology makes it possible to infer deep knowledge and automate complex tasks, delivering trustworthy insights that far exceed the capabilities of traditional systems.”
Anna FavaFormer Senior Manager, Data & AI · Dompé
Questions,
Answered
What is domain-aware AI?
An LLM that works from your organisation's definitions, policies and data rather than generic knowledge. In Prometheux, that knowledge sits in the ontology as context concepts and rule outputs, and LLM concepts reference it by name.
How is this different from RAG?
RAG retrieves text for the model to interpret. Here the model also receives the results of your business rules, its output is parsed into typed columns, and downstream rules can check it before anything acts on it.
How is this different from enterprise search or a context platform?
Those retrieve the most relevant documents for a model to read. Prometheux also hands the model the computed results of your business rules over live data, names every input in lineage, and checks the model's typed output with rules before anything acts on it.
How do I know what the model was given?
Every input is a named concept referenced in the prompt template, so lineage shows the exact context and rule outputs behind each answer.
Which LLMs can I use?
Each LLM concept uses a default provider or overrides the provider, model and named connection, with credentials stored separately in Settings.
Are LLM concept results deterministic?
No, and Prometheux says so: LLM concepts may give a different answer on a re-run. Keep decisions in deterministic Vadalog, SQL, Cypher or Python concepts and use the LLM for judgement and language.
Related Capabilities
Teach AI How
Your Business Works
Bring a policy, a glossary and a dataset. We'll show you an LLM grounded in your own rules.
