Retrieval · L1 · 15 min
Why retrieve
Learn in five minutes
- A model does not contain your private notes.
- Retrieval finds passages and puts them in the call.
- The 2020 RAG paper is the usual citation for that pattern.
- If retrieval misses, a fluent answer is still ungrounded.
- Say which passage you used.
Two systems can share a model and differ only in the notes they retrieve. The notes are the part you control.
Visual
Answer from one card
- Write three sentences on a card.
- Ask a question the card answers.
- Quote the sentence.
- Ask a question the card does not answer and refuse.
Quiz
Ecosystem
- Passage
- A chunk of source text the answer may use.
Sisters: Search indexes, Embeddings, Rerankers, Citations.
A voice agent and a chat agent can share the same note index.
The capstone answers only from notes you are allowed to use.
Config
- Write the goal in one sentence.
- Name the input and the allowed output.
- Save the trace next to the result.
Troubleshooting
- The answer looks fluent and is wrong. Check whether a source passage actually contains the claim.
- The loop never stops. Set a step cap and stop when the tool result repeats.
Sources and go deeper
- Lewis and co-authors, Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks, 2020 paper last checked 2026-10
- Retrieval-augmented generation — Wikipedia wikipedia last checked 2026-10