Chunk Retrieval

When vault search returns a truncated text_preview, use chunk retrieval to fetch the full chunk text and nearby chunks from the document manifest stored in Vault.

Endpoint

GET /vault/:id/objects/:objectId/chunks

When to use it

Search is optimized for relevance and compact responses. That means results often contain only a preview of the matched chunk. If the match lands inside a table, exhibit list, or a passage that spans several chunks, call read-chunks to recover the surrounding text. Example workflow:

Search returns chunk 13 with a partial exhibit table.
Request start=12&end=14.
Read the full text of chunks 12, 13, and 14 together.

Ranges are inclusive and capped at 10 chunks per request.

Parameters

Parameter	Type	Required	Description
`id`	string	Yes	Vault ID
`objectId`	string	Yes	Processed object ID
`start`	integer	No	First chunk index to return. Defaults to `0`.
`end`	integer	No	Last chunk index to return, inclusive. If omitted, only the `start` chunk is returned.

Example

curl "https://api.case.dev/vault/$VAULT_ID/objects/$OBJECT_ID/chunks?start=12&end=14" \
  -H "Authorization: Bearer $CASEDEV_API_KEY"

const response = await client.api.get(
  `/vault/${vaultId}/objects/${objectId}/chunks`,
  {
    params: { start: 12, end: 14 },
  }
)

for (const chunk of response.chunks) {
  console.log(chunk.index, chunk.text)
}

Response

{
  "object_id": "obj_abc123",
  "vault_id": "vault_def456",
  "chunks": [
    {
      "index": 12,
      "text": "full text of chunk 12...",
      "page_start": 5,
      "page_end": 5,
      "word_start_index": 5544,
      "word_end_index": 6055
    },
    {
      "index": 13,
      "text": "full text of chunk 13...",
      "page_start": 5,
      "page_end": 6,
      "word_start_index": 6056,
      "word_end_index": 6532
    },
    {
      "index": 14,
      "text": "full text of chunk 14...",
      "page_start": 6,
      "page_end": 6,
      "word_start_index": 6533,
      "word_end_index": 6881
    }
  ],
  "total_chunks": 42
}

For media-backed transcripts, chunks can include source audio/video timing when real word timing is available:

Media transcript chunk

{
  "index": 0,
  "text": "Good morning. Could you please state your full name for the record?...",
  "page_start": null,
  "page_end": null,
  "word_start_index": 0,
  "word_end_index": 181,
  "start_ms": 100,
  "end_ms": 75291
}

Response fields

Field	Description
`object_id`	Object that owns the returned chunks
`vault_id`	Vault containing the object
`chunks`	Full chunk entries for the requested range
`total_chunks`	Total stored chunk count for the object

Each chunk includes:

Field	Description
`index`	Chunk index within the document
`text`	Full text of the chunk
`page_start`, `page_end`	PDF page span when available
`word_start_index`, `word_end_index`	Word range when available
`start_ms`, `end_ms`	Source media timestamps for the first and last word in the chunk, in milliseconds, when the object is a media-backed transcript with real timing

Notes

end is clamped to the document’s final chunk if you request past the end.
Requests where start is out of range return an error.
Objects without stored chunks return total_chunks: 0 and an empty chunks array.
start_ms and end_ms are omitted for normal documents and text-only transcripts. They are not inferred when media word timing is unavailable.

Get Started

Platform

Resources

Endpoint

When to use it

Parameters

Example

Response fields

Notes

​Endpoint

​When to use it

​Parameters

​Example

​Response fields

​Notes

Endpoint

When to use it

Parameters

Example

Response fields

Notes