What is embedding?
An embedding maps text into a high-dimensional vector where semantically similar inputs land near each other. Twin embeds two things: the intent of every compiled skill, so a re-phrased goal can find it, and every chunk of ingested content, so a natural-language query can retrieve it.
Why it matters
It is the mechanism behind both halves of the product that behave like magic. The dispatch cache matching a re-worded request and the content store answering a question it was never asked in those words are the same operation on two different corpora.