Home / Platform / Knowledge and ontology
Knowledge and ontology
Agents that know what your words mean.
An ontology describes your business in your own terms. A knowledge graph fills it from your records, documents and code. Agents, reports and twins all reason over the same picture, under the same access rules.
Ontology Studio: describe your business to Kai, approve what it proposes.
Reads what you haveOffice files old and new, mailboxes, scans and CAD drawings.
Ontology Studio
A business user builds the ontology. By describing the business.
Open Ontology Studio and tell Kai, Camaan's builder agent, how your work hangs together. Kai proposes the types and relationships in plain words, asks before it writes, and never overwrites what is already there. The page follows along as it builds.
- "An IT service desk" became Customer, Ticket, Service and Support agent, with three relationships
- "Add service level agreements" extended it, live
- Every write approved by a person, and signed into the ledger
The knowledge graph
Filled from what your organisation already holds.
Records from your systems, text from your documents and structure from your code land in one graph, typed by your ontology, so an agent can follow a question from a ticket to the service, the contract and the code behind it.
Your systems' data
Tables in the lake engine map onto your ontology's types, with lineage from the source to every answer.
Text, read carefully
Entities and relationships drawn from documents, each linked to the passage it came from.
Parsed, not guessed
Python, JavaScript and TypeScript, SQL, COBOL, shell scripts and configuration are parsed into the same graph, and prose that mentions them is linked to them.
Rules first, models second
Deterministic extraction first. A specialised decision model can propose a relationship, accepted only above a confidence threshold, with a language model as the fallback.
Start again, safely
Rebuild a knowledge base's graph as a durable job that survives a restart. Remove a source, and what came from it goes too.
Large graphs, paged
Every entity and edge of a graph can be read page by page, so big estates stay usable.
Highlighted: formats added for an industrial group's document estate.
The knowledge base
Reads the files a plant, a bank or a hospital really keeps.
Not just this year's Office files. Twenty-year-old Word and PowerPoint files, spreadsheets with macros, multi-page scans, whole mailboxes and engineering drawings.
- PowerPoint slides in order, with speaker notes
- Spreadsheet values only; macros never run
- Mailboxes with every attachment read as its own file
- CAD title blocks, text, dimensions and layers
- Anything unreadable is listed as skipped, not lost
Built-in OCR
Twelve writing systems, read on your own servers.
Scans and images are read by OCR that runs inside your boundary, including on sites with no internet connection. Or use an OCR model of your choice.
One picture, many uses
Everything else reads the same knowledge.
The ontology isn't a side project. It is how agents, reports and twins agree on what a customer, an asset or an order is.
Agents
Answer and act in your terms, and follow relationships rather than guess them.
Scenarios
Reports that ask the same question the same way, against the same types.
Twins
Live models of assets and how they connect, typed by your ontology.
Marketplace offers
Industry knowledge packaged once, mapped onto your ontology at install.
Read: Building an ontology by talking it through See the use cases