Retrieval-augmented generation (RAG), explained
A language model alone answers from what it memorized in training. RAG adds a lookup step: the question is used to search your content, the best passages are given to the model, and the model is told to answer from them.
The search usually combines meaning-based search (embeddings, which find passages about the same idea even with different words) with keyword search for exact names, numbers and codes.
RAG is what makes business chatbots accurate and current: update the documents and the next answer changes, with no retraining of the model.
In AgenBase
AgenBase uses RAG with hybrid search (meaning and exact words) across your pages, files and text, and links each answer to the passages it used. Connected Google Sheets are read live at answer time.