Embeddings, explained
An embedding model reads a sentence or paragraph and outputs a vector, typically hundreds of numbers. 'How long is delivery?' and 'shipping times' produce vectors that are close, even though they share no words.
Chatbots use embeddings to find the passages in a knowledge base that answer a question, the retrieval step of RAG.
Because embeddings can miss exact codes, names or numbers, good systems combine them with keyword search.
In AgenBase
AgenBase splits your content into passages, stores their embeddings, and searches by meaning and by exact words together, so both 'delivery time' and an exact product code find the right passage.