Remy Reference/Agent SDK/Vector databases/DataSourceChunk

DataSourceChunk

Type

One chunk exactly as it was indexed, as returned by chunks().

One chunk of a document exactly as it was indexed, as chunks() returns it. It carries the chunk's index, its text, the page and heading path it came from, and either character offsets into the extracted markdown or a boundingBox for PDFs. Reading these shows why a document does or does not match a query.

DataSourceChunk.ts

interface DataSourceChunk {
  index: number
  text: string
  pageNumber: number
  headingPath: string[]
  charStart: number | null
  charEnd: number | null
  boundingBox?: { topLeftX, topLeftY, bottomRightX, bottomRightY }
  vector?: string
}

Fields

index

number

0-based position in the document.

text

string

The chunk's indexed text.

pageNumber

number

Page the chunk came from.

headingPath

string[]

Enclosing headings, outermost first.

charStart

number | null

Offset into the page's extracted markdown. Null for PDFs, which carry a boundingBox instead.

charEnd

number | null

End offset, paired with charStart.

boundingBox?

{ ...corners }

Region of the page, for PDFs.

vector?

string

On requestbase64 of a Float32Array. Only when vectors: true was passed.

Inspect the split

const chunks = await Policies.chunks(documentId);
chunks[0].headingPath; // [0, 1]