Graph RAG Configuration

Graph RAG builds a knowledge graph from your documents, enabling relationship-based queries and multi-hop reasoning that vector search alone cannot achieve.

Configuration Flow

Recommended Settings

The following settings optimize for relationship-based retrieval. Graph RAG requires LLM integration for entity extraction and graph traversal, making it the most powerful but also the most resource-intensive architecture.

Step 1: Project Setup

Project setup defines the identity and purpose of your RAG system. Graph RAG is designed for complex domains with rich entity relationships — choose domains where understanding connections between entities is critical.

Step 2: Knowledge Sources

Add your documents through direct upload or from your Document Library. Graph RAG benefits from well-structured documents with clear entity mentions and relationships. Higher quality input documents produce better knowledge graphs.

Step 3: Document Processing

Document processing prepares your raw text for both embedding and graph extraction. Graph RAG uses semantic chunking to preserve entity relationships across chunk boundaries, with larger overlap to ensure no connections are lost.

Step 4: Graph Editor

Graph Editor is unique to Graph RAG and provides tools to visualize, explore, and refine the knowledge graph extracted from your documents. Review the graph after extraction to verify entity accuracy and correct any extraction errors.

Step 5: Pipeline Configuration

Pipeline configuration controls how queries are processed and results are returned. Graph RAG combines vector search with knowledge graph traversal, using LLM-powered context assembly for rich, relationship-aware responses.

Step 6: API Setup

Once configured, generate your API key and integrate using REST API, Python SDK, or MCP Protocol. Test your endpoints before deploying to production.

How Graph RAG Works

1. Graph Extraction: LLM extracts entities and relationships from documents
2. Triple Storage: Entities and relationships stored in a knowledge graph database
3. Vector Indexing: Document chunks still embedded for semantic search
4. Query Planning: Incoming queries analyzed for entity references
5. Graph Traversal: Related entities found through relationship paths
6. Result Fusion: Vector and graph results combined for final answer

Entity and Relationship Types

Entity Types

• Person: Named individuals
• Organization: Companies, institutions, teams
• Product: Software, services, physical products
• Concept: Abstract ideas, methodologies
• Location: Cities, countries, regions

Relationship Types

• Hierarchical: Parent-child, manager-reporter
• Functional: Part-of, component-of
• Causal: Causes, leads-to, results-in
• Associative: Related-to, mentioned-with

When to Use Graph RAG

• Complex organizational structures or hierarchies
• Multi-hop questions ("Who manages the team that built X?")
• Domain with rich entity relationships
• Need for global summarization across documents
• Legal, medical, or research domains with interconnected concepts

Cost Considerations

Graph Quality Tips

• Use high-quality, well-structured documents for better entity extraction
• Review the graph after extraction to verify accuracy
• Delete incorrect edges manually through the Graph Editor
• Graph quality improves with more documents in the same domain
• Use BGE-Large-EN-v1.5 or Stella-EN-1.5B-v5 for better entity disambiguation