RDF Studio's built-in reasoner is a native Rust OWL 2 RL engine, built directly into its Oxigraph-based store. Reasoning happens inside the storage layer itself — no data leaves the store to be reasoned over.
The engine materializes inferences: it applies OWL 2 RL rules to your ontology and data, and writes the derived triples into the database.
someValuesFrom
allValuesFrom
hasValue
hasKey
intersectionOf
unionOf
A reasoner that is wrong is worse than no reasoner, because it is wrong quietly — it writes conclusions into your database and nothing looks broken. So the engine is checked against the standards body's own test cases, each of which carries its expected answer.
Every inferred triple carries a per-quad inferred flag — there is no separate "inference graph". This is why the UI can:
Separately, the ontology editor offers a non-persisted inference preview — dashed edges computed from the ontology files while you model. The preview disappears when toggled off; materialized inferences in the database do not.
Reasoning at scale runs here. The full OWL 2 RL closure of 133.6 million triples — the LUBM benchmark at 1,000 universities — takes about thirty-five minutes on a 10-core Apple M1 Max laptop with 64 GB of memory and derives 76.4 million facts, written into the store once so every query afterwards reads them directly rather than re-deriving anything. Loading that dataset takes about eleven and a half minutes, and all 28 benchmark queries return the answer counts LUBM publishes.
Next on the roadmap — LUBM-8000, the same benchmark at 8,000 universities: over a billion triples.
Three quarters of the closure's time is currently spent in two steps that use a single core — building the in-memory index and writing the derived facts back — while the rules themselves account for under a tenth of it across every core. Making those two use the whole machine is on the roadmap; it is an engineering item, not a limit of the approach.
Connecting to another store is about working with what you already run, not about escaping a ceiling here: GraphDB and any SPARQL 1.1 endpoint are supported today, with more connectors on the way. Each store applies its own reasoning strategy — see Truth Maintenance & Retraction for how those differ.