← Back to Glossary
Chunking
Phase 03 · RAGtechnique

Splitting long documents into smaller text segments so retrieval can pull exact paragraphs rather than entire files.

Why it exists

Embedding an entire 50-page book into one vector dilutes specific details. Chunking keeps search precise.

How it works

Splits text by character counts, token limits, paragraph breaks, or semantic shifts (e.g. 500-token chunks with 50-token overlap).

Where you'll use this