RAG nodes

Retrieval-Augmented Generation: embedders, vector stores, document loaders, retrievers.

5 nodes.

---

document_chunker — Document Chunker

Splits documents into smaller chunks for embeddings.

Config fields

Field	Type	Required	Values
chunkSize	number	no	—
chunkOverlap	number	no	—
separator	string	no	—
strategy	string	no	separator | character | recursive | token

Example config

{
  "chunkSize": 500,
  "chunkOverlap": 50,
  "separator": "\\n\\n",
  "strategy": "separator"
}

---

embeddings_azure_openai — Embeddings Azure OpenAI

Generates embedding vectors using Azure OpenAI embedding deployments.

Config fields

Field	Type	Required	Values
endpoint	string	yes	—
deployment	string	yes	—
apiVersion	string	no	—
secretRef	object	yes	—
inputKey	string	no	—
outputKey	string	no	—

Example config

{
  "endpoint": "https://my-azure-openai.openai.azure.com",
  "deployment": "text-embedding-3-large",
  "apiVersion": "2024-10-21",
  "inputKey": "user_prompt",
  "outputKey": "embedding"
}

---

extract_citations — Extract Citations

Phase 9.4 — scans an LLM answer for [N] markers, resolves each to the upstream retrieved document, and emits structured citations the chat UI renders as clickable footnotes. Wire after llm_call when the upstream chain produced documents.

Config fields

Field	Type	Required	Values
answerPath	string	no	—
documentsPath	string	no	—

Example config

{
  "answerPath": "answer",
  "documentsPath": "documents"
}

---

rag_retrieve — RAG Retrieve

Retrieves context chunks from provided documents or vector store.

Config fields

Field	Type	Required	Values
queryTemplate	string	no	—
topK	number	no	—
documents	array<string>	no	—
embedderId	string	no	—
vectorStoreId	string	no	in-memory-vector-store | knowledge-base | pinecone-vector-store | pgvector-store | azure-ai-search-vector-store | qdrant-vector-store | chroma-vector-store | weaviate-vector-store | redis-vector-store
knowledgeBaseId	string	no	—
searchMode	string	no	vector | bm25 | hybrid
bm25Weight	number	no	—
vectorWeight	number	no	—
rrfK	number	no	—
candidatesPerRanker	number	no	—
vectorStoreConfig	object	no	—
embeddingSecretRef	object	no	—

Example config

{
  "queryTemplate": "{{user_prompt}}",
  "topK": 3,
  "embedderId": "token-embedder",
  "vectorStoreId": "in-memory-vector-store"
}

---

rerank — Rerank

Reorders candidate documents by relevance via a cross-encoder reranker (Cohere / Jina / Voyage). Plug after rag_retrieve for the precision boost.

Config fields

Field	Type	Required	Values
queryTemplate	string	no	—
topN	number	no	—
providerId	string	no	cohere | jina | voyage
model	string	no	—
secretRef	object	no	—

Example config

{
  "queryTemplate": "{{user_prompt}}",
  "topN": 3,
  "providerId": "cohere",
  "model": "rerank-english-v3.0"
}