RAG

RAG (Retrieval-Augmented Generation) is an architecture connecting a language model to an external knowledge base: before the model answers, the system searches for matching document fragments and adds them to the prompt. The model responds based on provided facts, not just memory from training. This is the standard way to build company assistants and bots based on proprietary documents: fewer hallucinations, current knowledge without retraining, and the ability to cite sources. RAG quality depends mainly on search quality - the model won't fix being given the wrong fragments.