Hi Stefano. We don’t have to create embedding an every time. Typically we create it and store in a vectorDB. After that we can use that to retrieve augment queries to LLM.