GraphRAG performs multi-hop reasoning over entities and relations in a knowledge graph. It is well suited to questions where understanding complex links between entities matters.

RAG vs GraphRAG

	RAG (Vector)	GraphRAG (Graph)
Best for	"What is X?"	"How are A and B related?"
Retrieval	Semantic similarity	Entity-relation traversal
Strength	Single-hop fact lookup	Multi-hop reasoning, relational understanding
Storage	Vector DB (Chroma)	Property Graph (NetworkX)

Data Ingestion

Method A: Automatic extraction (default)

GraphRAG shares the same profile directory as RAG. An LLM extracts entities and relations from your documents.

Tips for better extraction:

• Keep entity names consistent in the text (e.g. always use the full name instead of mixing nicknames and titles)
• Write paragraphs around clear entities and relations
• Reduce noise (headers, footers, tables of contents, copyright notices, etc.)

Method B: Pre-built triplets (JSONL)

Place JSONL files under data/profiles/{profile}/triplets/:

Field	Type	Required	Description
subject	string	Yes	Subject entity
relation	string	Yes	Relation type
object	string	Yes	Object entity
subject_type	string	No	Subject type (e.g. "Person"); default "entity"
object_type	string	No	Object type (e.g. "Organization"); default "entity"
subject_properties	object	No	Extra subject attributes (reserved for multimodal, e.g. {"image": "img/a.png"})
object_properties	object	No	Extra object attributes (multimodal reserved)
confidence	float	No	Confidence score (default 1.0)
source	string	No	Source identifier

Examples:

{"subject": "Entity A", "relation": "works_at", "object": "Entity B"}
{"subject": "Entity A", "relation": "works_at", "object": "Entity B", "subject_type": "Person", "object_type": "Organization"}

Auto-detection priority

1. If a graph index already exists → load it directly
2. If profiles/{profile}/triplets/*.jsonl exists → Method B (import without LLM)
3. If there are documents under profiles/{profile}/ → Method A (LLM extraction)

Graph visualization

After building, the graph is saved as an interactive HTML file:

storage/{profile}/graph/knowledge_graph.html

Open it in a browser to explore entities and relations.

Tuning extraction parameters

When reading documents from the profile directory, the default is SimpleLLMPathExtractor for entity and relation extraction.

kg_extractor = SimpleLLMPathExtractor(max_paths_per_chunk=10)

For stricter extraction with predefined entity and relation types:

from llama_index.core.indices.property_graph import SchemaLLMPathExtractor

kg_extractor = SchemaLLMPathExtractor(
    possible_entities=["Person", "Company", "Product", "Technology"],
    possible_relations=["developed", "belongs_to", "uses", "located_in"],
)

Web UI

In the GraphRAG panel you can:

• Browse all entities and relations
• Rebuild the knowledge graph