Crabhaus

What it is

Retrieval-augmented generation (RAG) is a pattern where a model's answer is grounded by first retrieving relevant text from an external corpus and feeding it into the prompt, instead of relying only on facts memorized during training.

How it works

A pipeline splits a document collection into chunks, embeds each into a vector, and stores those vectors in an index. At query time the question is embedded the same way, the index returns the nearest chunks, and those chunks are inserted into the model's context alongside the question, so the model answers from retrieved text rather than parametric memory alone.

A concrete example

A support bot embeds its entire help-center archive into a vector database. Asked why an export is stuck at 90%, it retrieves the three closest help articles and pastes them into the prompt before answering. If a help article shipped last week, the bot answers correctly without retraining, because the new article was simply indexed after the fact.

Why it matters

RAG is the main way builders keep an LLM's answers current and grounded without retraining: update the index and the model's effective knowledge changes immediately, with every claim traceable to a retrieved passage. It is why RAG remains the default architecture for internal knowledge bots and any agent that must cite a private or fast-changing corpus.