grafeo-llamaindex¶

LlamaIndex integration that implements the PropertyGraphStore interface backed by GrafeoDB. Build knowledge graphs from documents and query them with structured and vector search.

GitHub PyPI

Overview¶

grafeo-llamaindex provides GrafeoPropertyGraphStore, a full implementation of LlamaIndex's PropertyGraphStore interface. It supports:

Structured queries (GQL, Cypher, Gremlin, GraphQL, SPARQL, SQL/PGQ)
Vector similarity search via native HNSW indexes
Knowledge graph construction from documents
30+ built-in graph algorithms

All backed by GrafeoDB's embedded database - no servers or Docker required.

Installation¶

uv add grafeo-llamaindex
# or
pip install grafeo-llamaindex

Requires Python 3.12+, grafeo >= 0.4.4 and llama-index-core >= 0.14.

Quick Start¶

Basic Usage¶

from grafeo_llamaindex import GrafeoPropertyGraphStore

store = GrafeoPropertyGraphStore(db_path="./my-graph.db")

# Insert nodes and relationships
from llama_index.core.graph_stores.types import EntityNode, Relation

nodes = [
    EntityNode(name="Alix", label="Person", properties={"age": 30}),
    EntityNode(name="Gus", label="Person", properties={"age": 25}),
]
store.upsert_nodes(nodes)

relations = [
    Relation(source_id="Alix", target_id="Gus", label="KNOWS"),
]
store.upsert_relations(relations)

# Query
result = store.structured_query(
    "MATCH (p:Person)-[:KNOWS]->(f) RETURN p.name, f.name"
)

Vector Search¶

store = GrafeoPropertyGraphStore(
    db_path="./my-graph.db",
    embedding_dimensions=384,
    distance_metric="cosine",
)

# Nodes with embeddings are automatically indexed
results = store.vector_query(
    query_embedding=[0.1, 0.2, ...],
    similarity_top_k=10,
)

With PropertyGraphIndex¶

from llama_index.core import PropertyGraphIndex, SimpleDirectoryReader

documents = SimpleDirectoryReader("./data").load_data()

index = PropertyGraphIndex.from_documents(
    documents,
    property_graph_store=store,
)

query_engine = index.as_query_engine()
response = query_engine.query("What does Alix do?")

PropertyGraphStore Interface¶

grafeo-llamaindex implements all 8 abstract methods:

Method	Description
`upsert_nodes()`	Insert or update graph nodes
`upsert_relations()`	Insert or update relationships
`get()`	Retrieve nodes by ID or properties
`get_triplets()`	Get relationship triples
`get_rel_map()`	Get relationship maps for entities
`delete()`	Remove nodes or relationships
`structured_query()`	Execute GQL/Cypher/Gremlin queries
`vector_query()`	HNSW similarity search over embeddings

Vector Search Configuration¶

Parameter	Default	Description
`embedding_dimensions`	`None`	Vector dimensions (required for vector search)
`distance_metric`	`"cosine"`	`cosine`, `euclidean`, `dot_product`, `manhattan`

Graph Algorithms¶

Access 30+ built-in algorithms via store.client.algorithms:

PageRank, betweenness centrality, closeness centrality
Louvain community detection, connected components
Shortest path (Dijkstra, BFS)

Requirements¶

Python 3.12+
grafeo >= 0.4.4
llama-index-core >= 0.14, < 1

License¶

Apache-2.0