document_chunks

source 1 PII columns

Embedded chunks — the actual semantic-search surface.

Definition

One row per chunk. `chunk_text` is what the embedding worker consumes and what `search_documents` returns to AI traffic. `(document_id, chunk_seq)` is the natural key.

6 of 6

Column

Description

Tests · PII

document_id

FK → documents.document_id.

document +4 PII2 tests

chunk_seq

0-based position in the DoclingDocument hierarchy. Stable across re-extracts.

1 test

chunk_text

Paragraph-shaped chunk text. Soft cap ~512 tokens.

1 testPII

section

Heading path Docling resolved for this chunk, if any. Nullable.

page_start

First page this chunk's content appears on. Nullable when Docling can't resolve a page.

page_end

Last page this chunk's content appears on. Nullable when Docling can't resolve a page.