Citations

ask() numbers the retrieved passages for the LLM and asks it to cite them as [1], [2], and so on. The answer’s citations list the passages it actually cited.

TypeScript

const answer = await doc.ask("What is the refund policy?");

answer.text;
// "Refunds are issued within 30 days [1]. Opened items are excluded [2]."

answer.citations;
// [
//   { n: 1, source: "policy.pdf", chunk: 4, page: 12, text: "..." },
//   { n: 2, source: "policy.pdf", chunk: 9, page: 14, pageEnd: 15, text: "..." },
// ]

Python

answer = doc.ask("What is the refund policy?")

answer.text
# "Refunds are issued within 30 days [1]. Opened items are excluded [2]."

answer.citations
# [Citation(n=1, source="policy.pdf", chunk=4, page=12, text="..."),
#  Citation(n=2, source="policy.pdf", chunk=9, page=14, pageEnd=15, text="...")]

Citation fields

Field	Meaning
n	The number used in the answer: 2 for [2]
source	File name, URL or fromText id
chunk	Chunk number in that source, from 1
page	First page of the passage (PDF only)
pageEnd	Last page, only when the passage crosses a page break
section	Heading path, such as "Policies > Refunds" (Markdown only)
text	The passage itself

Citations are listed in the order they first appear in the answer. Fields that do not apply are left out.

How citations are read from the answer

• [2], [1][3] and [1, 3] are all recognised.
• A number that matches no passage is ignored.
• A passage cited twice is listed once.

includeCitations: false in the ask options drops the instruction to cite and the source labels next to each passage. Most answers then have no numbers, so citations is usually empty.

Pages and sections in search results

Search results carry the same location in metadata: page, pageEnd and section. Use them to link to the right page of a PDF or the right heading of a Markdown file.

• PDF: pages are counted from 1.
• Markdown: section is the heading in effect where the chunk starts, with its parent headings, such as Setup > Linux > Troubleshooting. Headings inside fenced code blocks are ignored.

Older indexes

Indexes built before version 2.1 have no pages or sections. Answers still get citations, without those fields. Rebuild to add them:

TypeScript

await doc.build({ rebuild: true });

Python

doc.build(rebuild=True)

REST API and CLI

• The REST /ask response includes citations. See REST API.
• raglite ask prints the answer, then a Sources: list with each citation’s source, page and section.