Chunking
Splitting a document into smaller passages so each can be embedded, indexed and retrieved independently in a RAG pipeline.
How you chunk determines what an agent retrieves; clean headings and self-contained sections (the same structure that makes content answer-first) chunk cleanly and lower Cost of Retrieval.
- term
- Chunking
- category
- knowledge-memory
- short_def
- Splitting a document into smaller passages so each can be embedded, indexed and retrieved independently in a RAG pipeline.
- long_def
- Chunking decides the unit of retrieval: documents are cut into passages (by size, by heading, or semantically) before embedding, so a query returns the few most relevant chunks rather than a whole document. Chunk size is a tradeoff — too large dilutes relevance, too small loses context.
- see_also
ragembeddingsreranking- etymology_origin
- — verify-against-primary-at-build ↗ https://en.wikipedia.org/wiki/Retrieval-augmented_generation — 'chunking' is standard RAG-pipeline terminology
- related_to
ragembeddingsrerankingvector-database- contrast_with
- Unlike indexing a whole document as one unit, chunking retrieves passage-level pieces, so an agent gets the relevant paragraph instead of the entire file.
- example
- A 40-page manual chunked by section lets a query return just the two paragraphs that answer it, keeping the agent's context window focused.
- source
- https://en.wikipedia.org/wiki/Retrieval-augmented_generation
- status
- active
- why_it_matters
- How you chunk determines what an agent retrieves; clean headings and self-contained sections (the same structure that makes content answer-first) chunk cleanly and lower Cost of Retrieval.
- sameAs
https://en.wikipedia.org/wiki/Retrieval-augmented_generation- bridge_entity
- rag
- last_verified
- 2026-07-06
- md_twin
- /glossary/chunking.md
last verified · by Özden Erdinc