DataSourceStats

Type

How a corpus was built and what is in it, as returned by stats().

What stats() returns: document and chunk counts, documents by ingest state, storage, and the pipeline the corpus was built with. That pipeline is the configuration actually in effect, which is not necessarily the platform default and not necessarily the newest, since a corpus keeps its configuration until someone migrates it. Changing it is an explicit, owner-triggered rebuild.

DataSourceStats.ts

interface DataSourceStats {
  exists: boolean
  documentCount: number
  counts: { total, done, processing, error }
  chunkCount: number
  storageBytes: number
  lastIngestedAt: string | null
  pipeline: { version, embeddingModelId, dimensions, chunking, contextual, images } | null
}

Fields

exists

boolean

False for a corpus nothing has created yet; everything else then reads zero.

documentCount

number

Documents in the corpus.

counts

{ total, done, processing, error }

Documents by ingest state.

chunkCount

number

Total chunks across the corpus.

storageBytes

number

Original document bytes, not index size.

lastIngestedAt

string | null

When a document last finished processing.

pipeline

{ ... } | null

The configuration these documents were built with: embedding model, dimensions, chunking, contextual, and image settings.

Read the pipeline

const s = await Policies.stats();
s.pipeline?.chunking.maxChars; // chunk size in effect
s.counts.error;                // documents that failed