Abstract—While Large Language Models (LLM) excel at semantic reasoning across concepts and domains, existing database systems, including those with vector capabilities and knowledge graphs, only support similarity search and graph traversal. They cannot perform inter-domain discov-ery, analogical reasoning, or causal chain discovery. This paper proposes a semantic reasoning database (ReasonDB), a novel database paradigm with a functional prototype that validates a couple of core capabilities. ReasonDB is the first system that treats embeddings as primary data and implements multiple reasoning modes as first-class opera-tions. Through three core innovations: vector-native storage with probabilistic semantics, semantic reasoning primitives, and machine learning-driven adaptive indexing, ReasonDB transforms databases from passive storage systems into active discovery partners. Experimental validation demonstrates two breakthrough results: 7× performance improvement in inter-domain analogical reasoning over similarity search, and successful causal chain discovery where vector similarity fundamentally fails (0.708 causal chain strength despite 0.074 cosine similarity). This paper investigates four key areas. First, how vector-native storage transforms query capabilities to enable discovery-based reasoning that uncovers novel rela-tionships across domains. Second, what new query classes and reasoning capabilities become possible. Third, how adaptive indexing improves performance over fixed strategies. Fourth, how these innovations integrate into a cohesive architecture. ReasonDB demonstrates that semantic reasoning requires fundamentally different database primitives. This enables entirely new classes of discovery operations impossible with current database architecture.
 
Keywords—semantic reasoning databases, analogical rea-soning, causal chain discovery, inter-domain discovery, vector-native storage, reasoning primitives, adaptive indexing, prob-abilistic semantics, database architecture


Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).