Reference

← Schema catalogue

knowledge.embedding

A vector embedding of a knowledge.chunk, keyed by (chunk_hash, model). Because chunks are content-addressed, an embedding is too — re-embedding the same text with the same model is idempotent, and a different model yields a distinct record (so vectors are never confused across models). Retrieval does cosine similarity over these; a vector DB (pgvector) is only needed at large scale — brute-force over the tenant's vectors suffices for a municipal corpus.