Implementation:Infiniflow Ragflow Dealer Insert Citations

Knowledge Sources	RAGFlow
Domains	RAG, Information_Retrieval
Last Updated	2026-02-12 06:00 GMT

Overview

Concrete tool for inserting citation markers into LLM answers by matching to retrieved chunks provided by RAGFlow's Dealer class.

Description

Dealer.insert_citations matches segments of the LLM-generated answer to retrieved chunks using hybrid keyword+vector similarity. It inserts [ID:N] markers in the answer text and returns the set of matched chunk indices. Called from decorate_answer in DialogService when the quote option is enabled.

Usage

Called automatically during chat response post-processing.

Code Reference

Source Location

Repository: ragflow
File: api/db/services/dialog_service.py (L630-638 call site), rag/nlp/search.py (Dealer.insert_citations)

Signature

class Dealer:
    def insert_citations(
        self,
        answer: str,
        content_ltks: list[str],
        vectors: list,
        embd_mdl,
        tkweight: float,
        vtweight: float
    ) -> tuple[str, set]:
        """Insert citation markers into answer text.

        Args:
            answer: str - LLM-generated answer.
            content_ltks: list[str] - Tokenized content of retrieved chunks.
            vectors: list - Embedding vectors of retrieved chunks.
            embd_mdl: EmbeddingModel - For encoding answer segments.
            tkweight: float - Keyword similarity weight (1 - vector_similarity_weight).
            vtweight: float - Vector similarity weight.

        Returns:
            tuple[str, set] - (answer with [ID:N] markers, set of matched indices).
        """

Import

from rag.nlp.search import Dealer
# Or via settings.retriever

I/O Contract

Inputs

Name	Type	Required	Description
answer	str	Yes	LLM-generated answer text
content_ltks	list[str]	Yes	Tokenized chunk contents
vectors	list	Yes	Chunk embedding vectors
embd_mdl	EmbeddingModel	Yes	For encoding answer segments
tkweight	float	Yes	Keyword weight
vtweight	float	Yes	Vector weight

Outputs

Name	Type	Description
answer	str	Answer with [ID:0], [ID:1], etc. markers
indices	set	Set of matched chunk indices

Usage Examples

# Internal usage in decorate_answer
answer, idx = retriever.insert_citations(
    answer,
    [ck["content_ltks"] for ck in kbinfos["chunks"]],
    [ck["vector"] for ck in kbinfos["chunks"]],
    embd_mdl,
    tkweight=1 - dialog.vector_similarity_weight,
    vtweight=dialog.vector_similarity_weight,
)
# answer now contains [ID:0], [ID:1] markers

Related Pages

Implements Principle

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment