Vector Store

Vector Store stores embedding vectors in Drupal and answers "what is most similar to this?" with one small API. It works with any embedding provider, model or entity type, and needs no external vector database.

Why use it

• Runs where your site runs. The default backend works on any Drupal database, including SQLite. On MariaDB 11.7+ you can switch to native VECTOR columns with an HNSW index for larger datasets.
• Better similarity out of the box. Optional mean-centering removes the shared direction that makes every embedding look alike. Scores spread out and the
  closest matches are actually relevant.
• Sensible cutoffs, not magic numbers. Each set of vectors reports how similar random pairs of its vectors usually are, so callers can set a "similar enough"
  threshold based on your data.
• Cache-safe. Every write invalidates a cache tag for its collection, so pages built from similarity results stay correct.
• Extensible. Backends are tagged services
• No surprises. You choose the backend explicitly. If the selected backend can't run on your server, the module tells you why instead of silently using
  another one.

Who it is for

Module developers building related content, recommendations, semantic search or deduplication on top of embeddings. It powers the similarity recommendations in Entity Collection.

Requirements

Who it is for

Module developers building related content, recommendations, semantic search or deduplication on top of embeddings. It powers the similarity recommendations in Entity Collection (coming soon).

Requirements

• Drupal 11
• Optional: MariaDB 11.7 or later for the native backend