Chunking
Also: text chunking, document chunking
The indexing step of RAG that decides the unit of retrieval. Fixed-size chunks (often with overlap) are simple but can cut facts in half; recursive, semantic, sentence-window and parent-document strategies respect the structure of the text.
Example
A 2,000-token document cut into 500-token chunks with 100-token overlap gives five chunks: 0–500, 400–900, 800–1300, 1200–1700 and 1600–2000, so a sentence straddling token 500 still appears whole in the second chunk.