llama-index-graph-stores-coordinode

LlamaIndex PropertyGraphStore backed by CoordiNode — graph + vector + full-text in a single transactional engine.

Installation

pip install llama-index-graph-stores-coordinode

uv add llama-index-graph-stores-coordinode

Requirements

• Python 3.11+
• Running CoordiNode instance

Quick Start

PropertyGraphIndex — Build from Documents

from llama_index.core import PropertyGraphIndex, SimpleDirectoryReader
from llama_index.graph_stores.coordinode import CoordinodePropertyGraphStore

# Connect to CoordiNode
store = CoordinodePropertyGraphStore("localhost:7080")

# Build graph index from documents (auto-extracts entities + relationships)
docs = SimpleDirectoryReader("./data").load_data()
index = PropertyGraphIndex.from_documents(
    docs,
    property_graph_store=store,
    show_progress=True,
)

# Query
engine = index.as_query_engine(include_text=True)
response = engine.query("Explain the attention mechanism in transformers")
print(response)

Load Existing Graph

from llama_index.core import PropertyGraphIndex
from llama_index.graph_stores.coordinode import CoordinodePropertyGraphStore

store = CoordinodePropertyGraphStore("localhost:7080")

# Load from existing data in CoordiNode (no re-extraction)
index = PropertyGraphIndex.from_existing(property_graph_store=store)

engine = index.as_query_engine()
response = engine.query("Who are the key researchers in graph neural networks?")

Vector + Keyword Hybrid Retrieval

from llama_index.core import PropertyGraphIndex
from llama_index.core.indices.property_graph import (
    VectorContextRetriever,
    LLMSynonymRetriever,
)
from llama_index.graph_stores.coordinode import CoordinodePropertyGraphStore
from llama_index.embeddings.openai import OpenAIEmbedding

store = CoordinodePropertyGraphStore("localhost:7080")
embed_model = OpenAIEmbedding(model="text-embedding-3-small")

index = PropertyGraphIndex.from_existing(
    property_graph_store=store,
    embed_model=embed_model,
)

# Combine vector similarity + keyword/synonym retrieval
retriever = index.as_retriever(
    sub_retrievers=[
        VectorContextRetriever(index.property_graph_store, embed_model=embed_model),
        LLMSynonymRetriever(index.property_graph_store),
    ]
)
nodes = retriever.retrieve("graph attention networks")

Manual Node and Relation Upsert

from llama_index.core.graph_stores.types import EntityNode, Relation
from llama_index.graph_stores.coordinode import CoordinodePropertyGraphStore

store = CoordinodePropertyGraphStore("localhost:7080")

# Upsert entities
nodes = [
    EntityNode(label="Person", name="Alice", properties={"role": "researcher"}),
    EntityNode(label="Concept", name="GraphRAG", properties={"field": "AI"}),
]
store.upsert_nodes(nodes)

# Upsert relationships
relations = [
    Relation(
        label="RESEARCHES",
        source_id="Alice",
        target_id="GraphRAG",
        properties={"since": 2023},
    )
]
store.upsert_relations(relations)

# Query triplets
triplets = store.get_triplets(entity_names=["Alice"])
for subj, rel, obj in triplets:
    print(f"{subj.name} -[{rel.label}]-> {obj.name}")

Direct Cypher

from llama_index.graph_stores.coordinode import CoordinodePropertyGraphStore

store = CoordinodePropertyGraphStore("localhost:7080")

result = store.structured_query(
    "MATCH (n:Person)-[:KNOWS]->(m) WHERE n.name = $name RETURN m.name AS name",
    param_map={"name": "Alice"},
)

Connection Options

# host:port string
store = CoordinodePropertyGraphStore("localhost:7080")

# TLS
store = CoordinodePropertyGraphStore("db.example.com:7443", tls=True)

# Custom timeout
store = CoordinodePropertyGraphStore("localhost:7080", timeout=60.0)

Capabilities

Related Packages

Links

• Source
• CoordiNode server
• LlamaIndex PropertyGraph docs
• Issues

Support

License

Apache-2.0