Build an entity graph to answer cross-document questions

Extract typed entities and relation triples, resolve name variants, then answer multi-hop questions over the graph.

Goal

Connect dispersed facts through explicit relations that can be audited.

Define Pydantic entities, relations, and source fields; extract with client.messages.parse(...) as in the notebook. [graph]
Extract subject-predicate-object triples per document, resolve alias candidates, and retain provenance. [graph]
Build a relevant subgraph and answer a multi-hop question; include node/edge sources for review. [graph]

Aliases resolve while original mentions remain traceable; a two-edge question is answered via graph paths, not model recall.

If entities merge incorrectly, review canonical-ID evidence and allow manual splits; uncertain relations remain candidates.

Preserve document-level ACLs; graph only authorized material and filter inaccessible nodes/edges before output.

Extraction can miss or mislink entities/relations; the in-memory graph is not a transactional graph database.

These values describe the upstream repository, not popularity or likes for an individual skill or playbook.

No metrics snapshot is available yet.

License: MIT