Context Is Half
the Answer
Policies, glossaries and team notes tell an agent what exists. Rules tell it how to apply them. Prometheux keeps both as named, readable definitions in one governed place, so every answer traces back to the notes, data and rules behind it.
| order | net |
|---|---|
| SO-1182 | £4,200 |
| SO-1187 | £910 |
| SO-1190 | £12,040 |
Proven in production
Stuffed Prompts.
Or Governed Context.
Pasting documents into prompts makes answers slower, costlier and harder to trust. Prometheux treats knowledge like code: named, referenced explicitly, reviewable, wired into your business rules and traceable.
Knowledge as
Queryable Data
Upload a PDF, policy, glossary or regulation and it is chunked and embedded into the Context Layer. Context concepts expose those notes as a table with a fixed shape, so rules, LLM concepts and agents can use them like any other data.
One Standard Shape
Every context concept returns note_id, kind, title, text, source, created_at and folder_path.
Pinned or Searched
Pin specific notes by ID for stable references, or run a semantic query that grows with the Context Layer.
Typed Knowledge
Filter notes by kind: fact, document chunk, preference or observation.
From Notes to Concepts
The AI Assistant extracts concepts from ingested documents, so recurring knowledge becomes structured logic.
Give agents exactly the knowledge they need, not everything. Keep policies in one governed place. Turn tribal knowledge into reusable context.
Models Inside
Your Logic
LLM concepts put a language model in the ontology as a first-class step. Prompt templates reference other concepts, so every input is explicit, typed and part of the lineage.
Explicit Dependencies
Templates reference concepts with {{concept_name}}, which creates lineage from context and data to output.
Structured Output
Declare output_columns and the model's reply is parsed into a typed table, not free text.
Right Model per Task
Override provider and model per concept: a fast model for simple steps, a stronger one for hard ones.
Bounded and Safe
Row and token limits are enforced and reported, and injected data is framed as content, not instructions.
Control what each model sees and what it costs. Trace every AI output back to its context.
From Documents to
Governed Context
Ingest
Upload policies, glossaries, regulations and notes. They are chunked and embedded into the Context Layer.
Define Context Concepts
Pin the notes that must always apply, or define a semantic query for notes that keep arriving.
Wire Into Logic
Reference context from LLM concepts and rules, alongside your structured data.
Run With Lineage
Every output shows which context, data and rules it came from.
Knowledge That
Agents Can Use
Policy-Aware Credit Review
Combine lending policy notes with client data so every explanation cites the policy it applied.
Regulatory Documents
Bring submissions and guidelines into the Context Layer and extract the concepts they define.
Team Know-How
Capture observations and preferences as typed notes that future runs and agents reuse.
Context Plus Rules
Agents retrieve context and run deterministic rules over it, instead of improvising from a long prompt.
“Generative AI without ontologies is an engine without a road. Prometheux builds the road.”
Marco RomeiDeputy CIO · Banca Sella
“Prometheux enabled us to build a system that can access our databases, process all the client's historical variables, and provide an intelligent, understandable explanation autonomously.”
Massimiliano MalinHead of Credit Risk · Banca Sella
Questions,
Answered
What is context as code?
It means treating unstructured knowledge such as policies, glossaries and notes as named concepts in your ontology. They are queryable, referenced explicitly by rules and LLM steps, and show up in lineage.
How is this different from RAG?
RAG retrieves text and leaves the model to decide what to do with it. Here context is one input to your business logic: you choose which notes apply, and rules over your data determine the outcome.
How is this different from a catalog that gives agents business context?
A catalog describes data and certifies definitions for agents to read. Here context is wired into executable logic: rules over your data decide the outcome, LLM steps name exactly which notes they use, and lineage shows the notes, data and rules behind each output.
Can we manage context like the rest of our code?
Yes. Context and LLM concepts are named definitions that reference each other explicitly. Rules are plain text you can keep in your own version control, and ontologies export to a portable file and import back into a workspace.
Can I see which documents shaped an answer?
Yes. LLM concepts reference context concepts by name, so lineage shows which notes, data and rules fed each output.
What documents can I add?
PDFs, policies, glossaries and regulatory texts. They are chunked and embedded into the Context Layer, and the AI Assistant can extract concepts from them.
Can I control which model and how many tokens are used?
Yes. Each LLM concept can override the provider and model, and runs enforce row and token limits that are reported after each run.
Related Capabilities
Turn Knowledge
Into Context
Bring a policy and the data it governs. We'll show agents applying it with full lineage.
