bdi-mental-states

Model beliefs, desires, and intentions as BDI ontology classes linked by motivational properties: beliefs motivate desires, intentions fulfill desires and specify plans. Ground every belief in a WorldState via refersTo and attach Justification instances for explainability.

T2B2T (Triples-to-Beliefs-to-Triples) is a bidirectional pipeline: incoming RDF triples are translated into belief instances via BeliefProcess, BDI deliberation selects intentions and plans, then execution results are projected back as RDF world states using bringsAbout.

Yes, through Logic Augmented Generation (LAG). Serialize the BDI ontology into the LLM prompt, generate triples, then validate them against ontological restrictions such as intentions fulfilling desires and beliefs referencing world states, retrying with error feedback when inconsistent.

Use competency queries that check motivational chains, plan task ordering, temporal validity windows, and justifications. Validation queries verify constraints like every intention fulfilling exactly one desire and every belief referencing a world state.

Avoid BDI modeling when you only need persistent facts or entity memory across sessions, or when you only need agent topology and handoff coordination. BDI is justified when explainable agency matters: why an agent believed something and which plan it committed to.

Beliefs without temporal bounds cannot be garbage-collected or conflict-checked during diachronic reasoning. Assign hasValidity with TimeInterval start and end times so SPARQL filters can resolve overlapping or stale beliefs about the same world state.