dsh-ontology

Most agent memory is a bag of strings: the agent writes prose, later reads prose, and nothing ever tells it that what it just recorded contradicts what it recorded last week. dsh-ontology upgrades memory into a typed ontology. You declare which classes exist, how they relate, and what domain and range constraints each relation carries. Every subsequent assertion is then validated against that vocabulary; a violation is rejected with the exact constraint named, rather than silently stored as another undistinguishable blob of text.

The core power lies in inference and traceability. Transitive, symmetric, and inverseOf rules are applied to a fixpoint so composition works correctly. Entailed facts are derived on read and never persisted; retract a premise and every fact resting on it vanishes cleanly, leaving no stale residue. Queries support subclass-aware class matching, triple-pattern matching, neighborhood traversal, and shortest-path search, far more reliable than grepping prose. Functional cardinality ensures a subject has at most one valid object per relation, and retraction cascades safely through every fact mentioning an entity, keeping the entire graph self-consistent.

It is built for teams that need their agent to manage structured domain knowledge over the long term: maintaining knowledge bases, tracking product specifications, mapping dependency graphs, or any scenario where every fact the agent remembers must be verifiable, derivable, and safely retractable.