THE NATUSAI CAPABILITY CATALOGUE
Bring your information, knowledge and workflows together in tools built around the way your organisation works. NatusAI helps teams put AI to work while keeping the context, choices and ownership close.
01 / CONNECT INFORMATION
Bring files, messages, systems and live sources into a shared working context so teams can spend less time hunting for what they need.
Example: A service team can bring a customer’s request, documents and account history together before responding.
Bring together the sources that shape everyday work instead of creating another isolated store.
Ingestion patterns can cover files, web content, REST and public APIs, feeds, streams, email, repositories, MCP resources, ontologies, tables and dataframes.
Prepare documents and data so people and AI can work from cleaner, more consistent information.
Parsing can cover PDF, DOCX, HTML, text, JSON, CSV, spreadsheets, presentations, code, email, media, structured data and XML. Preparation can include layout-aware extraction, OCR where configured, and text, date, number and entity normalisation with plain, recursive, token, sentence, paragraph, semantic, entity, relation, graph, ontology-aware, hierarchical, community, centrality and model-assisted splitting.
Find the people, organisations, places, events and relationships that turn content into business facts.
Semantic extraction can include entities, relations, triplets, events and coreference, with schema validation, confidence filtering, consensus metadata and configurable fallback methods.
02 / BUILD BUSINESS KNOWLEDGE
Connect facts, definitions and identities so the important relationships are available when people need to make sense of a situation.
Example: A product manager can see the customers, policies, documents and decisions connected to a release.
Link people, assets, documents, activities, policies and decisions into a shared context that explains how things relate.
Knowledge, context and agent-context graphs can support entity linking, graph expansion, keyword retrieval, semantic retrieval, hybrid retrieval and GraphRAG-style question answering. Trusted reference data can be seeded, loaded and merged.
Align the names and definitions different teams use so information can be understood consistently.
Ontology workflows can cover class and property inference, namespaces, naming conventions, OWL, RDF and JSON-LD export, SHACL shape generation and validation, SKOS vocabularies, competency questions and quality gates.
Surface records that may describe the same thing and show where sources disagree, so people can resolve ambiguity deliberately.
Quality workflows can use blocking and similarity methods, clustering, entity merge strategies, source-aware deduplication, conflict detection, investigation and resolution. Policies and thresholds are scoped to the application.
03 / FIND ANSWERS AND SUPPORT WORK
Let people search by meaning, use the relevant context and connect an answer to authorised actions and repeatable work.
Example: Someone can ask which policy applies to a case, see the supporting sources and start the approved follow-up.
Find relevant information together with the relationships and history that make it useful.
Embedding paths and vector stores can support semantic, metadata, namespace and decision retrieval. Hybrid retrieval can combine vector, graph, metadata, decision and proximity signals.
Combine explicit business knowledge with model-assisted work while keeping sources, rules and context inspectable.
Optional model providers can support extraction, embeddings and language tasks. Deterministic graph construction, rules and provenance can operate without an external model, with data paths chosen for the application.
Help an assistant move work forward by connecting selected knowledge and authorised operations to the people who need them.
Integration patterns include MCP tools, REST access, agent-context stores and adapters for common agent frameworks. Exposed operations, permissions and approval steps are part of application design.
04 / APPLY RULES AND REVIEW DECISIONS
Record the evidence, rules and actions around a decision so teams can review the outcome and improve the next one.
Example: A reviewer can inspect the policy check, evidence and recorded rationale behind an approval.
Evaluate facts and proposed actions against the policies and conditions your business sets.
Reasoning patterns can include forward and backward chaining, Rete networks, Datalog, SPARQL reasoning, temporal, abductive, deductive and graph reasoners, truth maintenance and human-readable explanations.
Capture the situation, options, confidence and outcome around a decision, with links to precedent and impact.
Decision records can support category and scenario fields, precedent search, causal links, impact analysis, policy checks and audit reports. Recorded impact paths show documented relationships for review.
Follow a fact back to its source and see how it was transformed, checked or updated.
Provenance can follow W3C PROV-O-oriented patterns with source tracking, operation attribution, integrity checks, graph and ontology versions, diffs and changelog entries. Teams can review the sources, rules and recorded actions behind an outcome.
05 / UNDERSTAND RELATIONSHIPS AND CHANGE
Explore the shape of your business over time, from important dependencies to the context that was available when something happened.
Example: A team can compare the current account picture with an earlier point in time and inspect the connected decisions.
Highlight central entities, communities, paths and structural similarities so teams can focus their investigation.
Graph analysis can include centrality, connectivity, shortest paths, communities, hierarchies, link prediction, structural similarity, validation and graph summaries. Distance-aware retrieval can inspect hop bands, paths, neighbourhoods and confidence decay.
Distinguish when something was true from when it was learned, then compare versions as context changes.
Temporal features can include bi-temporal facts, interval operations, natural-language time normalisation, point-in-time reconstruction, versioned snapshots, temporal query rewriting and temporal reasoning.
Test extracted knowledge, retrieval and decisions against checks that make improvement visible.
Evaluation patterns can include exact, keyword, regex, ROUGE, edit-distance, numeric, temporal and length checks, plus decision scoring, repeated sampling and model-assisted judging.
06 / DEPLOY AND RETAIN CONTROL
Choose the architecture, deployment and data lifecycle that fit your requirements, while keeping useful knowledge portable and accountable.
Example: You can start with one focused workflow, export the knowledge it creates and expand the system as your needs become clearer.
Let people inspect relationships, timelines, decisions, search results, quality and evidence without asking an engineer for every answer.
Explorer-style interfaces can combine graph, timeline, semantic-distance and embedding views with decision chains, duplicate review, ontology validation, lineage and exports. Lyra provides the reusable interface language; Cygnus remains the backend and database foundation.
Run the stages that load, prepare, understand, validate and publish information in a consistent way.
Pipeline, configuration, lifecycle and plugin-registry patterns can support directed stages, dependencies, parallel work, scheduling, retries, reusable templates and coordinated ingestion, extraction, graph construction, reasoning, evaluation, provenance and export.
Adapt the connected business tools to your systems, deployment choices and ownership priorities, with a managed path for portability and deletion work.
Architecture patterns can include graph, RDF/triplestore and vector storage, REST and MCP servers, local or external models, container deployment and exports such as RDF, JSON-LD, N-Triples, OWL, CSV, JSON, YAML, GraphML, GEXF, DOT, Parquet and Arrow. Vector enumeration, migration, reindexing and coordinated erasure can report per-store status, receipts and partial failures; deletion outcomes depend on the stores and backups in scope.
QUESTIONS PEOPLE ASK
NatusAI is designed to bring together files, business applications, databases, messages, repositories, feeds and other configured sources. The exact connectors, permissions and data routes depend on the application and deployment we design with you.
We can shape the deployment around your organisation’s hosting and ownership priorities. Cygnus provides the backend and database foundation, while model providers and connected services are chosen as part of the application design.
Answers can be connected to relevant sources, relationships, applied rules and recorded actions, giving people a clear context to review and discuss.
Portable exports and coordinated lifecycle workflows can support migration, reindexing and selected-record deletion across configured stores. Reports can show completed steps and failures, while the outcome depends on the stores and backups in scope.
BUILD YOUR CONNECTED UNIVERSE
Tell us where your information, decisions or workflows feel disconnected. We’ll help you map a practical place to begin.
