Skip to main content

Core Concepts

OpenMetadata models an ontology with entities you already know. Nothing here requires a triple store: ontologies, concepts, properties, relationship types, and constraints are stored in the relational database as first-class OpenMetadata entities, with versions, owners, and change events.

The Mapping to Standards

The following table maps OpenMetadata entities to the standards they implement. OpenMetadata models your metadata against the canonical OpenMetadata OWL ontology, which uses the namespace:
It reuses established vocabularies rather than inventing everything: DCAT (datasets, distributions, catalogs), PROV-O (derivation and provenance), SKOS (concept schemes), Dublin Core Terms, DQV (data quality), FOAF, ODRL, and CSVW. om:DataAsset is declared owl:equivalentClass dcat:Dataset, so a DCAT-aware consumer understands your catalog without a custom adapter.

Ontologies

An ontology is an independently governed model. Its concepts, properties, and constraints are stored as separate entities that the ontology owns. Glossaries don’t own them, and existing glossaries aren’t converted into ontologies.

Concepts

A concept is a class, a SKOS concept, or both. Beyond name, description, and hierarchy, a concept carries:
  • A concept type: OWL_CLASS, SKOS_CONCEPT, or BOTH. It preserves the vocabulary a source ontology used, so imported ontologies keep their original typing.
  • An IRI: a stable, globally unique identifier. Renaming the concept keeps it.
  • Synonyms and concept mappings: alternative labels and alignments to external standards.
  • Superclasses and subclasses: derived from the SUBCLASS_OF constraints of the ontology, so you change the hierarchy by editing constraints or drawing Subclass Of relationships.
  • Abstract: marks a class with no direct instances. Data is classified by its subclasses.
  • Definition source: the IRI of the document or standard the definition comes from.
  • Deprecated and Replaced by: a deprecated concept stays resolvable for existing references, and Replaced by names the concept in the same ontology that replaces it.
  • Data assets: the tables and columns bound to the concept. This is the join between the semantic model and the physical catalog.

Properties

A property is a formal object, data, or annotation property owned by an ontology. Properties are inherited down the concept hierarchy. A property declared on Account shows as inherited on every subtype, with the ancestor that declares it, so it is obvious where it came from.

Relationship Types

A relationship type is an admin-governed definition of one kind of edge. It is what makes a relation typed rather than a generic “related” link, and it is what an agent or a reasoner consults to decide what can be inferred. Manage them at Settings > Governance > Glossary Term Relations (admin only). The relationship picker in Edit mode lists these types.

Properties of a Relationship Type

A relationship type has the following fields.

Types Shipped by Default

The following types ship with OpenMetadata.
Characteristics aren’t decoration. TRANSITIVE on broader is what lets the transitive-closure inference rule answer “every ancestor of this concept” in one hop. SYMMETRIC on relatedTo is what stops the graph from storing both directions. Set them deliberately.
For more information, see Relation Types.

Constraints

A constraint is a governed OWL statement about your concepts. The technical name is an axiom. Constraints are stored relationally, so they remain authoritative and editable when RDF is switched off. Class expressions are authored with a structured editor rather than free text, so what you build is always well-formed: Restriction operators: SOME, ONLY, VALUE, MIN, MAX, EXACT.
OWL profile guardrail. With RDF_STRICT_OWL_PROFILE=true (the default), OpenMetadata rejects authored axioms that fall outside the supported OWL 2 DL profile. That keeps the model decidable, which is exactly what bounds the work a reasoner has to do.
Your glossary is also projected to OWL, and that projection stays inside the OWL 2 DL profile so that a reasoner can classify it. Terms that use reserved vocabulary keep their standard meaning. A subClassOf relationship becomes a subclass axiom between declared classes, and seeAlso becomes an annotation. A chain or disjointness rule that names a reserved predicate is left out of the projection, and the server logs a warning.

Concept Mappings

A concept mapping aligns one of your concepts with a concept in an external vocabulary: FIBO, FHIR, schema.org, an internal legacy dictionary, anything with an IRI. Mappings are exported as the corresponding skos:*Match and owl:sameAs triples, which is what makes your ontology interoperable rather than merely internally consistent.

IRIs and Namespaces

Every concept, property, and constraint gets a stable IRI, minted from the ontology Namespace when you leave the IRI empty.
  • An ontology created in Collate’s Ontology Studio gets the IRI <base URI>/onto/<prefix> and the namespace <base URI>/onto/<prefix>#. The base URI is the server’s RDF_BASE_URI setting, which defaults to https://open-metadata.org/.
  • An ontology created by installing a library pack or importing a file into a new ontology gets a generated identifier of the form urn:openmetadata:ontology:<id>.
  • An IRI is preserved when you rename the definition, so references from other tools stay valid.
Changing an ontology’s namespace after concepts exist changes how new IRIs are minted. Decide the namespace before you start authoring at scale, the same way you would decide a package name.

Releases

Edits to an ontology collect as unpublished changes on its working draft. Publishing captures them as a numbered, immutable release. If the ontology has reviewers, a release is reviewed through the approval workflow before it goes live. Otherwise, an author or owner publishes it directly. Only one release is Active at a time. Users without edit permission, the knowledge graph, and Export of a published release read the active release. Users with edit permission can also read the working draft.

Ontology Layers

Every ontology declares a layer: L1, L2, or L3. Layers label the governance level of the model. The API creates ontologies on L3 by default. You can change the layer when you update the ontology.

Provenance

Relations, constraints, and imported concepts carry provenance, so “who asserted this and how” survives:
  • Relations record who created them and when, plus a provenance marker that distinguishes manually authored edges from imported, suggested, or inferred ones.
  • Library installs record the pack, its version, and the modules they came from on the ontology.
  • Inferred triples in the knowledge graph are written into separate named graphs and marked, for example om:inferred and om:inferredTagSource, so a derived tag is never mistaken for a curated one.