Skip to content
BENIBIX Learn

Retriever (RAG)

Also known as Retriever, Retrieval Step, Retrieval Component

The part of a RAG system that finds the best chunks for a question.

Think of it like

It is like a helper who gets the right books. Then the expert reads them.

Example

A user asks about refund rules. The retriever searches 8,000 chunks and passes the top 5 to the model, all from the refund policy.

Why it matters

Most wrong RAG answers start with a bad search. Better search often helps more than a bigger model.

Definition

The retriever is the part of a RAG system that finds the chunks to show the model. It takes the question and searches the knowledge base. It can use meaning search, keyword search or both. It hands back a short, ranked list of the best chunks. The model then writes its answer from those. If the retriever misses, the model never sees the fact.
The retriever sits in the middle: it takes the question and fetches chunks.

Common mistake

Blaming the model for a wrong answer without checking what the retriever found. Look at the chunks first.

Key takeaways

  1. 1
    Question in. Searches the knowledge base.
  2. 2
    Chunks out. A short, ranked list.
  3. 3
    Check it first. Misses here cause wrong answers.