For each question, the system searches a knowledge base, usually a vector database that finds texts by meaning rather than exact words. It retrieves the most relevant texts and passes them to the model as a basis for the answer. RAG is most common when the answer needs to draw on a knowledge base that is specific to your business or kept up to date.
Our standpoint
We choose RAG when data changes often, when it matters to show the source of an answer or when fine-tuning would be overkill.