Skip to content
16.8.13 Full Text and RAG Document Index

16.8.13 Full Text and RAG Document Index

The current DBMS manual provides full Markdown text and a page-level JSON index.

Full Text

LanguageURL
Englishllms-full.txt
Koreanllms-full.txt

Full text concatenates published DBMS pages in navigation weight order. Each page boundary includes the title, language, and canonical URL. The body preserves source Markdown.

JSON Index

LanguageURL
Englishllms-chunks.json
Koreanllms-chunks.json

Schema version 1 has the following top-level fields.

FieldDescription
schema_versionJSON contract version; currently 1
productMachbase DBMS
manual_versionProduct version covered by the manual
languageen or kr
document_countSize of the documents array
documentsDocument metadata array

Each document provides id, title, url, markdown_url, kind, parent_url, weight, and last_modified. JSON does not duplicate the body. Fetch each page’s Markdown from markdown_url or use llms-full.txt as the corpus.

Chunk Boundaries

The current index treats one published page as one document chunk. SQL syntax, SDK tasks, and operating procedures remain on separate pages where possible, so URLs and titles can serve as stable chunk identifiers. Preserve the page id when subdividing large dictionary pages.

Nondeterministic values such as build time are omitted. Neo, DBMS 8.5, drafts, and alias pages are excluded from both the index and full text.

Last updated on