Deductive AI

If every doctor is a person and Alice is a doctor, Alice is also a person. A query for people should include her without requiring someone to enter that extra fact manually.

Legra uses ontologies and rules to derive such conclusions. This is deductive AI: the result follows from stated premises and a rule, rather than a language model's prediction.

Reasoning belongs to the shared graph. SPARQL, Cypher, and GQL benefit from the same ontologies, rules, and inferred facts. The same applies to query languages Legra adds in the future.

In the example above, queries for people include Alice whether you write them in SPARQL, Cypher, or GQL. In Cypher and GQL, inferred types appear as labels and inferred relationships as edges.

Choose the query language that suits your work.

Use vocabulary relationships and supported RDFS, OWL, and Datalog reasoning to describe what follows from the knowledge in a workspace. Custom rules express relationships specific to a domain.

Rules are versioned knowledge too. Test a proposed change on a branch before applying it to shared work.

An explanation names the rule and facts behind an inferred result. For example, a person guaranteeing a loan may be classified as a guarantor by a business rule. Follow that conclusion back to the guarantee relationship.

Validation has a separate job: checking whether changes satisfy constraints. Neither mechanism establishes that an incorrect starting fact is true.