RAG (Retrieval-Augmented Generation)
RAG, retrieval-augmented generation, is a technique where a model retrieves relevant information and generates its answer from that, rather than from training alone.
The mechanism is straightforward: turn the question into a search, fetch the passages that look most relevant, put them in front of the model, and ask it to answer from them. It grounds answers in current company material and lets the system cite where each claim came from, which is why it became the default pattern for AI over private data.
Its weakness is that everything depends on the retrieval step. If the right passage is not returned, the model answers from an incomplete picture and sounds equally confident doing so. RAG also treats each request independently, so it has no view of which of two conflicting documents is current. Those are the gaps a context layer is meant to close.