Skip to main content
The Graph RAG node builds knowledge graphs from your documents and enriches retrieval with entity and relationship information. It extracts entities (people, organizations, locations, etc.) and their relationships, then uses this structured knowledge to provide richer, more contextual retrieval results. Graph RAG node overview

​
Overview

The Graph RAG node enhances retrieval by:
  1. Building a knowledge graph — Extracts entities and relationships from documents
  2. Intelligent chunking — Splits documents using title-based chunking
  3. Entity extraction — Identifies people, organizations, locations, products, and more
  4. Relationship mapping — Discovers connections between entities
  5. Enriched retrieval — Augments retrieved chunks with graph context
Graph RAG is ideal when your documents contain rich entity relationships (e.g., organizational structures, product catalogs, research papers with citations). For simpler use cases, consider Smart RAG or standard Chunking + Retrieval.

​
When to Use Graph RAG

ScenarioRecommendation
Documents with many entities✅ Recommended — Extracts and links entities
Organizational/relational data✅ Recommended — Maps relationships
Research papers with citations✅ Recommended — Connects authors, papers, concepts
Product catalogs✅ Recommended — Links products, categories, features
Simple Q&A⚠️ Consider Smart RAG — Less overhead
Quick prototyping⚠️ Consider Smart RAG — Faster setup

​
Using the Graph RAG Node

​
Adding the Graph RAG Node

  1. Open your flow in the Flow Builder
  2. Drag the Graph RAG node from the sidebar onto the canvas
  3. Connect your Dataset node to Graph RAG
  4. Connect a Question or Testset node for queries
  5. Double-click the Graph RAG node to configure

​
Input Connections

The Graph RAG node accepts input from:
Source NodePurpose
DatasetRequired — Provides documents to process
QuestionProvides queries for retrieval (simulation/testing)
TestsetProvides multiple queries for comprehensive testing
To simulate Graph RAG, you must connect a Question or Testset node as input. Without queries, the node cannot demonstrate retrieval results.

​
Output Connections

The Graph RAG node can connect to:
Target NodeUse Case
RerankingFurther improve result quality with LLM-based scoring
LLMGenerate natural language responses
AnalysisEvaluate retrieval performance
ResponseOutput retrieved results directly

​
Configuring the Graph RAG Node

Double-click the Graph RAG node to open the configuration panel: Graph RAG configuration

​
Top K

The number of chunks to retrieve for each query:
ValueUse Case
1-3Precise, focused answers
4-6Balanced coverage (default: 5)
7-10Comprehensive context

​
Add Metadata

When enabled, includes document metadata in chunk content:
  • File name
  • Page number
  • Position information

​
Add Page Annotation

When enabled, includes page-level annotations in chunk content.
Only available for files processed with the Graphor parsing method.

​
Add Document Annotation

When enabled, includes document-level annotations in chunk content.
Only available for files processed with the Graphor parsing method.

​
How Graph RAG Works

​
Processing Pipeline

  1. Chunking — Documents are split using title-based chunking
  2. Entity Extraction — LLM identifies entities in each chunk
  3. Relationship Extraction — LLM discovers connections between entities
  4. Graph Storage — Entities and relationships are stored in a knowledge graph
  5. Retrieval — Queries find relevant chunks via similarity search
  6. Enrichment — Retrieved chunks are augmented with entity/relationship context

​
Entity Types

Graph RAG extracts these entity types by default:
Entity TypeExamples
PersonJohn Smith, Dr. Jane Doe
OrganizationAcme Corp, MIT, WHO
LocationNew York, Europe, Building A
ProductiPhone, Model X, Widget Pro
TechnologyPython, Machine Learning, REST API

​
Relationship Types

Default relationship types include:
RelationshipDescription
IS_AClassification (e.g., “Python IS_A Programming Language”)
PART_OFComposition (e.g., “Engine PART_OF Car”)
MEMBER_OFMembership (e.g., “John MEMBER_OF Team A”)
WORKS_FOREmployment (e.g., “Jane WORKS_FOR Acme Corp”)
KNOWSAssociation (e.g., “Alice KNOWS Bob”)

​
Viewing Results

Graph RAG provides two views for results:

​
Results Tab

Shows retrieved chunks enriched with entity and relationship information:
  • Question — The query being answered
  • Chunk Content — Original text
  • NLP Entities — Extracted entities with descriptions
  • NLP Relationships — Connections between entities
  • Metadata — File name, page number, score

​
Graph View Tab

Visual representation of the knowledge graph:
  • Nodes — Entities colored by type
  • Edges — Relationships between entities
  • Interactive — Zoom, pan, and explore the graph
Graph RAG graph view

​
Pipeline Examples

​
Simple Graph RAG Pipeline

Dataset → Graph RAG → LLM → Response ← Question Best for: Knowledge-intensive Q&A with entity-rich documents.

​
Graph RAG with Reranking

Dataset → Graph RAG → Reranking → LLM → Response ← Question Best for: Maximum precision with entity context and LLM reranking.

​
Graph RAG Evaluation Pipeline

Dataset → Graph RAG → Analysis ← Testset Best for: Evaluating retrieval quality with entity enrichment.

​
Graph RAG vs. Other RAG Nodes

FeatureGraph RAGSmart RAGAgentic RAG
Entity extraction✅ Yes❌ No❌ No
Relationship mapping✅ Yes❌ No❌ No
Graph visualization✅ Yes❌ No❌ No
ConfigurationTop K, metadata, annotationsTop K, metadata, annotationsNone
Processing timeLonger (entity extraction)MediumFast
Best forEntity-rich documentsGeneral documentsQuick setup

​
Best Practices

  1. Use for entity-rich content — Graph RAG shines with documents containing many named entities and relationships
  2. Allow processing time — Entity extraction takes longer than standard chunking
  3. Explore the Graph View — Visualize your knowledge graph to understand extracted relationships
  4. Combine with Reranking — Add Reranking for better precision on complex queries
  5. Test with Testsets — Evaluate performance before deployment

​
Troubleshooting

If Graph RAG returns no results:
  • Verify a Question or Testset node is connected
  • Check that Dataset contains processed documents
  • Ensure documents have status “Processed”
  • Wait for entity extraction to complete (may take longer)
If the Graph View shows few entities:
  • Check document content — entities may not be present
  • Verify documents are properly parsed
  • Try documents with clearer entity mentions
If Graph RAG is slow:
  • Entity extraction requires LLM calls per chunk
  • Reduce number of documents in Dataset
  • Consider Smart RAG for faster processing
If page/document annotations don’t show:
  • Verify documents were parsed with Graphor method
  • Enable the corresponding checkbox in settings
  • Reprocess documents if needed

​
Next Steps

Smart RAG

Simpler RAG without entity extraction

Dataset

Configure document selection for Graph RAG

Reranking

Add LLM-based reranking for better results

Evaluation

Measure retrieval quality with metrics