loading
loading
The component that takes a query and pulls back the most relevant chunks from your index to feed the model.
In a RAG pipeline the retriever is the step between the user's question and the model: it embeds (or keyword-matches) the query, hits the index, and returns the top-k chunks. Its quality caps the whole system — if the right passage isn't in what it returns, no amount of clever prompting saves you, the model just can't see the answer. Tune it with chunking, the number k of results, and a reranking pass. Example: query in, ten candidate snippets out, which then get stuffed into the prompt as grounding.
Plainly
Think of Retriever as a named building block in a big LEGO app. The component that takes a query and pulls back the most relevant chunks from your index to feed the model.
In practice
Use it when you are mapping how the app is structured or explaining a feature to a teammate or agent. In practice, define the owner, input, output, and failure mode before you rely on it.