chunk

One piece of a larger document, split up so each piece is small enough to embed and retrieve individually.

A chunk is one contiguous piece of a larger document, produced by splitting the document into pieces small enough that each one embeds into a single, meaningful vector and fits comfortably inside a prompt's context budget alongside several other retrieved chunks. Chunk size and overlap are chosen to keep each piece coherent — cutting arbitrarily mid-sentence or mid-idea makes chunks that embed and retrieve poorly.

Examples

A 40-page policy manual might be split into 200-word chunks with a short overlap between neighbors, so each chunk stays coherent and small enough to embed well.