angee.knowledge_graph_pgvector.retrieval

The plugin's retrieval backend — a lexical stub standing in for pgvector.

Registered into ANGEE_KNOWLEDGE_RETRIEVAL_CLASSES under the pgvector key by :mod:~angee.knowledge_graph_pgvector.autoconfig, so a vault selects it with retrieval_class = "pgvector" and the knowledge search_pages resolver dispatches to it through the vault's public retrieval_for seam — no edit to the knowledge addon.

This is a skeleton: it subclasses the bundled lexical backend and inherits its title/body icontains search, so the seam is exercised end to end while no real embeddings exist yet. A production plugin overrides :meth:search with an approximate-nearest-neighbour query over an embedding column it adds to knowledge.Page (a model extends + its own migration and pgvector index).

PgvectorRetrievalBackend

class PgvectorRetrievalBackend(LexicalRetrievalBackend)

Stub semantic backend: inherits lexical search until embeddings land.

Concrete registry leaf (key="pgvector"). A real implementation keeps the registration shape and replaces :meth:search with a vector-similarity query; the embedding column, its backfill, and the ANN index are this plugin's responsibility (a migration), never the knowledge addon's.