An embedding is a list of numbers that captures what a piece of text means. Two texts about the same thing get vectors that sit close together, even when they share no words. That is what makes “how long do refunds take?” find an article titled “Returning an order”.

​
Get an embedding

You need an embedding model added under AI → Models (see Providers and models).
Send many texts in one call when you have them. The answers come back in the order you sent them.

​
You often don’t need to do it yourself

  • To search your own records by meaning, add a Vector field to a table and let Pawabase fill it in each time a record is saved. See Vector search.
  • To let an agent answer from documents, add them to a knowledge base. Pawabase splits, embeds and searches them for you.
Calling embed yourself is for cases those don’t cover: comparing two texts, grouping, or storing vectors somewhere else.

​
Keep the size consistent

Every embedding model produces vectors of a fixed length, shown as Vector size on the model. A vector field in a table must be the same size as the model that fills it. If you switch to a model with a different size, make a new field and fill it again.

​
Cost

Embeddings are cheap but add up when you embed a lot. AI → Usage shows the tokens and cost of each model so you can see it before the bill does.