Abstract:
Description logic (DL) reasoning retrieves classes related to a submitted class expression under an ontology’s axioms. AberOWL has provided this capability as a public service since 2015. We describe AberOWL 2.0, which retains the reasoning core and adds a distributed deployment, an agent interface, a machine-readable catalogue and container-based private deployment. A registry coordinates workers that hold classified ontologies and add, remove or update them while running. A Model Context Protocol server exposes ten tools for identifier grounding, ontology discovery, DL querying and SPARQL rewriting and execution at external endpoints. The catalogue uses MOD, DCAT and Hydra metadata. The measured public deployment contains 971 biomedical ontologies and 12.1 million classes, with 958 ontologies classified by ELK. We evaluated five language models on 162 scoped grounding terms, including 40 constructed negatives, and 120 DL queries over three ontologies. With grounding tools, accuracy on positive terms selected for resolver success was 92.6–96.7% under an instruction to answer. With the reasoning tool, exact answer-set match was 86.7–100% across model/task combinations without a syntax hint, compared with zero without tools. These results support agent use of the service on the evaluated tasks. Failure analysis identifies incorrect expression construction, altered answer sets and false assignment of suggested identifiers. A worked syntax example reduced accuracy for some models. The service, source code and saved evaluation responses are publicly available; historical service and ontology provenance limit exact repetition of the experiments.