Retrieval-Augmented Generation (RAG) is a technique to improve LLMs answer precision by giving it external sources of data. This data often is documentation, allowing for the User to browse easily among the specific and updated data, and therefore, instead of having to search dozens of pages of documentation, the User just asks the LLM, for example. As a downside, if the source of data is incomplete, outdated, or simply wrong, it will produce bad answers. (but that's quite obvious)