Expand description
Optional pgvector (PostgreSQL) backend for dense (embedding) search.
This module is behind the pgvector feature flag. It mirrors the
qdrant_store API (namespaced per-project tables, md5-derived point ids,
delete-by-file + upsert incremental sync) but talks to a self-hosted
PostgreSQL instance with the vector extension installed.
Deliberately dependency-free: SQL is executed through the psql CLI —
the same pattern the postgres context provider uses — so no async runtime
or native driver enters the dependency tree. Row output is read as
one JSON object per line (json_build_object(...)::text), which is robust
against arbitrary characters in file paths and symbol names.