Skip to main content

AI & embeddings

An AI store lets you work with raw text or images instead of vectors. Ahnlich AI runs an embedding model for you, turns each input into a vector, and stores it — so you get semantic search without running a model yourself.

StoreVector DB store · :1369[0.91, 0.12, …] vectorAI store · :1370text / imageembedding model
DB stores take a vector directly; AI stores take text or an image and embed it for you (index model = stored data, query model = searches).

DB store vs AI store​

  • Vector DB store (port 1369) — you compute vectors and send them.
  • AI store (port 1370) — you send raw_string or raw_binary inputs and the server embeds them.

Both expose the same operations; only the input type differs — the Vector DB / AI switch on each operation page shows both.

Index model & query model​

An AI store is created with two models:

  • index model — embeds the data you store.
  • query model — embeds incoming search queries.

They're usually the same model (e.g. all-minilm-l6-v2 for text, resnet-50 for images) so stored data and queries share one vector space. Ahnlich ships several models; pick one that matches your modality.

CREATESTORE my_store QUERYMODEL all-minilm-l6-v2 INDEXMODEL all-minilm-l6-v2 \  PREDICATES (author, category) STOREORIGINAL

Preprocessing & original data​

  • Preprocess action — controls how inputs are prepared before embedding (e.g. truncating or padding to the model's expected size vs. erroring on a mismatch).
  • Store original (STOREORIGINAL) — keep the raw text/image alongside its vector so results can return the source, not just the embedding.