Knowledge graph and ontology platforms compared

Two different things share the word “ontology”: the W3C standards stack (Ontology, OWL, RDF) and product-level “business ontologies” that bind definitions to enterprise data. Knowledge graphs for retrieval are covered in GraphRAG.

PlatformWhat “ontology” meansData stays whereActions / writebackStatus (fetched 2026-10-07)
Palantir Ontology (palantir-foundry-and-aip)Object, link, action types, functionsIntegrated into Foundry datasetsYes: action types submit governed editsPer Palantir docs; production status not dated here; proprietary
Microsoft Fabric IQ ontology (microsoft-fabric)Entity types, properties, relationships, metrics, rules, namespaces bound to dataYes, virtual layer over OneLake, semantic models etc.Not documented as writeback; context for agents and dashboardsPreview per Microsoft Learn page dated 2026-09-15; imports and exports RDF/Turtle/OWL; old experience retires 2027-01-31
Databricks (Genie, databricks-genie and databricks-ai-bi)Genie Ontology (per the Databricks AI/BI release notes: “Genie Ontology enabled by default (Public Preview)” on 2026-08-13; ontology snippets Public Preview since 2026-07-02 and available to all customers on 2026-08-13; Pages, Beta since 2026-08-12, are the human-modeled layer): “a unified context layer that gives Genie a business-aware map of your organization”; plus curated semantics (metric views, instructions) in Genie Agents, governed by Unity CatalogLakehouse, plus 50+ connected systems per launch coverageRead-focused context and Q&A; agent mode is GA, actions run through tools and MCPPublic Preview (Ontology and snippets); checked against Databricks release notes 2026-10-07, connected-source list from a conference recap (secondary)
Neo4jProperty graph, Cypher; no built-in business ontology found in the docs openedInside the graph DBDatabase writesProperty graph database (docs)
StardogRDF knowledge graph, SPARQL and GraphQL, OWL reasoning, virtual graphsVirtual graphs query data in placeDatabase writesVoicebox NL interface (docs)
TigerGraphProperty graph, GSQL, OpenCypherIn TigerGraphDatabase writesGraphRAG tooling, vector search; Savanna managed and self-managed DB (docs)

Reading the differences (author’s synthesis, opinion)

  • Standards-based vs product-based: Stardog and Fabric (import/export) speak RDF/OWL; Palantir’s model is its own.
  • Virtual vs ingested: Fabric and Stardog emphasise querying data in place; Palantir’s ontology is described as integrated with Foundry datasets (docs).
  • Actions vs context: Palantir couples ontology with actions and functions that change state; Fabric ontology, per its docs, supplies context for agents and dashboards. Graph DBs give traversal; business governance on top is not documented in the pages opened.
  • GraphRAG needs a graph store (Neo4j, TigerGraph, Stardog) or a graph layer (Fabric Graph, optional) while business ontologies help agents choose correct tables and metrics.

Sources

Open items

  • Databricks Genie Ontology: the Public Preview label on 2026-08-13 applies to ontology snippets specifically (AI/BI release notes 2026), whole-product status not stated; Pages are Beta (2026-08-12); the list of 50+ connected systems comes from a conference recap, not from Databricks docs.
  • No licence, edition or maturity comparison for Stardog/TigerGraph beyond docs pages.
  • Re-checked 2026-10-07 (primary): https://docs.databricks.com/aws/en/ai-bi/release-notes/2026 ; https://docs.databricks.com/aws/en/release-notes/product/2026/august ; Microsoft Learn ontology page (ms.date 2026-09-15: preview, RDF/Turtle/OWL import, RDF/Turtle export, old experience retires 2027-01-31; entity types, properties, relationships, metrics, rules, namespaces all confirmed). Palantir, Stardog, TigerGraph, Neo4j doc sites load; Stardog Voicebox and TigerGraph OpenCypher/hybrid graph+vector search appear in their doc navigation.