Tools & retrieval
What is Retrieval?
Retrieval is fetching candidate documents or records for a query.
Search APIs, keyword indexes, and vector stores are all retrieval tools.
Why it matters
Retrieval is the foundation of RAG and knowledge-grounded agents. Without good retrieval, the model operates on stale training data or hallucinations.
Retrieval is a tool, not a model feature. It uses search APIs, keyword indexes, vector stores, or hybrid systems to find candidate documents. The model then reasons over those candidates.
Key takeaways
- 1Retrieval quality determines RAG quality.
- 2Search APIs, keyword indexes, and vector stores are all retrieval tools.
- 3Retrieval is a tool call, not a model capability.
Common mistakes
- ✕Confusing retrieval with the model "knowing" something.
- ✕Using only keyword search when semantic search would find better results.
Related terms
Concept neighborhood
Terms linked from Retrieval in the glossary graph.
- Retrieval
- RAG
- Tool Calling
- Semantic Search