Skip to content

Property Graph Database Configuration

Database Selection

Set PG_GRAPH_DB to select the property graph backend. Use {TYPE}_GRAPH_DB_CONFIG (JSON blob) for per-store connection settings — this takes precedence over the generic GRAPH_DB_CONFIG fallback.

PG_GRAPH_DB=neo4j              # See supported values below
NEO4J_GRAPH_DB_CONFIG={"uri": "bolt://localhost:7687", "username": "neo4j", "password": "password"}

Framework backend is selected separately:

GRAPH_BACKEND=llamaindex       # llamaindex (default) | langchain
KG_EXTRACTOR_BACKEND=llamaindex  # llamaindex | langchain

LangChain-only stores (arangodb, apache_age, hugegraph, surrealdb, tigergraph, cosmos_gremlin) automatically set GRAPH_BACKEND=langchain — no explicit override needed. spanner uses LlamaIndex only (llama-index-spanner; langchain-google-spanner requires langchain-core<1.0 and is incompatible).

Property Graph Databases Supported on Both LlamaIndex and LangChain

These stores work with GRAPH_BACKEND=llamaindex (default) or GRAPH_BACKEND=langchain.

Neo4j

Primary knowledge graph storage with Cypher querying.

  • Dashboard: Neo4j Browser (http://localhost:7474)
PG_GRAPH_DB=neo4j
NEO4J_GRAPH_DB_CONFIG={"uri": "bolt://localhost:7687", "username": "neo4j", "password": "password"}

ArcadeDB

Multi-model database supporting graph, document, key-value, and search.

  • Dashboard: ArcadeDB Studio (http://localhost:2480)
PG_GRAPH_DB=arcadedb
ARCADEDB_GRAPH_DB_CONFIG={"host": "localhost", "port": 2480, "username": "root", "password": "password", "database": "flexible_graphrag"}

Embedded mode (in-process): 

ARCADEDB_MODE=embedded
ARCADEDB_DB_PATH=./arcadedb_data
ARCADEDB_EMBEDDED_SERVER=false

FalkorDB

Super fast graph database using GraphBLAS.

  • Dashboard: FalkorDB Browser (http://localhost:3001)
PG_GRAPH_DB=falkordb
FALKORDB_GRAPH_DB_CONFIG={"url": "falkor://localhost:6379", "database": "falkor"}

Ladybug

Embedded property graph database (Cypher, single .lbug file) with optional HNSW vector index.

  • Dashboard: Ladybug Explorer (http://localhost:7003, optional Docker)
PG_GRAPH_DB=ladybug
LADYBUG_GRAPH_DB_CONFIG={"db_dir": "./ladybug", "db_file": "database.lbug", "use_vector_index": true}

Memgraph

Real-time graph database with native streaming and graph algorithm support.

  • Dashboard: Memgraph Lab (http://localhost:3002)
PG_GRAPH_DB=memgraph
MEMGRAPH_GRAPH_DB_CONFIG={"url": "bolt://localhost:7687", "username": "", "password": ""}

NebulaGraph

Distributed graph database for large-scale data.

  • Dashboard: NebulaGraph Studio (http://localhost:7001)
PG_GRAPH_DB=nebula
NEBULA_GRAPH_DB_CONFIG={"space": "flexible_graphrag", "host": "localhost", "port": 9669, "username": "root", "password": "nebula"}

Amazon Neptune (Property Graph)

Fully managed AWS graph service — property graph mode.

  • Dashboard: Graph-Explorer (http://localhost:3007) or Neptune Workbench (AWS Console)
PG_GRAPH_DB=neptune
NEPTUNE_GRAPH_DB_CONFIG={"host": "your-cluster.region.neptune.amazonaws.com", "port": 8182}

Amazon Neptune Analytics

Serverless graph analytics with openCypher.

PG_GRAPH_DB=neptune_analytics
NEPTUNE_ANALYTICS_GRAPH_DB_CONFIG={"graph_identifier": "g-xxxxx", "region": "us-east-1"}

Property Graph Databases Supported Only with LangChain

These 6 stores only work with GRAPH_BACKEND=langchain (auto-selected when PG_GRAPH_DB is set to one of these). They support ingestion via add_graph_documents() and NL→Cypher/AQL/GSQL/SurrealQL retrieval via a TextToGraphQueryRetriever.

ArangoDB

Multi-model database with AQL graph traversal.

  • Dashboard: ArangoDB Web UI (http://localhost:8529)
  • Docker: docker compose -f docker-compose.yaml -f includes/arangodb.yaml -p flexible-graphrag up -d arangodb
PG_GRAPH_DB=arangodb
ARANGODB_GRAPH_DB_CONFIG={"url": "http://localhost:8529", "database": "flexible_graphrag", "username": "root", "password": "password"}

Apache AGE

PostgreSQL extension adding Cypher graph querying to Postgres.

  • Port: 5434 (separate from the pgvector Postgres at 5433)
  • Docker: docker compose -f docker-compose.yaml -f includes/apache-age.yaml -p flexible-graphrag build apache-age && docker compose ... up -d apache-age
PG_GRAPH_DB=apache_age
APACHE_AGE_GRAPH_DB_CONFIG={"host": "localhost", "port": 5434, "database": "flexible_graphrag_age", "username": "flexible_graphrag", "password": "password", "graph_name": "knowledge_graph"}

Azure Cosmos DB for Gremlin / TinkerPop

Gremlin-compatible graph traversal — works with both local Apache TinkerPop Gremlin Server and cloud Azure Cosmos DB for Gremlin.

  • Local Docker: docker compose ... -f includes/gremlin-server.yaml up -d gremlin-server (port 8182)
  • Cloud: Azure Cosmos DB for Gremlin — use wss:// endpoint with primary key from Azure portal

Local (TinkerPop Gremlin Server): 

PG_GRAPH_DB=cosmos_gremlin
COSMOS_GREMLIN_GRAPH_DB_CONFIG={"url": "ws://localhost:8182/gremlin", "username": "/", "password": ""}

Cloud (Azure Cosmos DB for Gremlin): 

PG_GRAPH_DB=cosmos_gremlin
COSMOS_GREMLIN_GRAPH_DB_CONFIG={"url": "wss://my-cosmos.gremlin.cosmos.azure.com:443/", "username": "/dbs/graphdb/colls/knowledge_graph", "password": "your_primary_key==", "partition_key_property": "partitionKey", "partition_key_value": "graph"}

The username format for Cosmos DB is /dbs/<database-name>/colls/<graph-container-name>. The password is the primary key from Azure portal → your Cosmos DB account → Keys.

partition_key_property must match the partition key path set on your Cosmos DB graph container (e.g. container created with /partitionKey → use "partitionKey"). Default is "partitionKey", auto-applied for any cosmos.azure.com URL.

partition_key_value is a fixed string written on every vertex. Using a fixed value keeps all graph data in one logical partition so traversals never cross partition boundaries. Do NOT use the entity type — that scatters vertices across partitions, forcing expensive cross-partition queries for most graph traversals. Default is "graph".

Apache HugeGraph

Distributed graph database with openCypher and Gremlin support.

  • REST API: http://localhost:8082
  • Hubble UI: http://localhost:8085
  • Docker: docker compose ... -f includes/hugegraph.yaml up -d hugegraph
  • Requires: uv pip install --override extras-overrides.txt -e ".[langchain,langchain-extras]" (includes hugegraph-python>=1.5.0)
PG_GRAPH_DB=hugegraph
HUGEGRAPH_GRAPH_DB_CONFIG={"host": "localhost", "port": 8082, "database": "hugegraph", "username": "admin", "password": "password"}

SurrealDB

Multi-model database with SurrealQL graph querying.

  • REST/WebSocket: http://localhost:8010
  • Surrealist UI: http://localhost:8011
  • Docker: docker compose ... -f includes/surrealdb.yaml up -d surrealdb
  • Requires: uv pip install -e ".[surrealdb-extras]" (isolated group — do NOT mix with langchain-extras)
PG_GRAPH_DB=surrealdb
SURREALDB_GRAPH_DB_CONFIG={"url": "ws://localhost:8010/rpc", "namespace": "test", "database": "flexible_graphrag", "username": "root", "password": "root"}

TigerGraph

Enterprise-grade distributed graph database with GSQL.

  • RESTPP: http://localhost:9002
  • GraphStudio UI: http://localhost:14240
  • Docker: docker compose ... -f includes/tigergraph.yaml up -d tigergraph
  • Requires: uv pip install -e ".[langchain,langchain-extras]" (includes pyTigerGraph>=1.0.0)
PG_GRAPH_DB=tigergraph
TIGERGRAPH_GRAPH_DB_CONFIG={"host": "http://localhost", "port": 14240, "restpp_port": 9002, "database": "MyGraph", "username": "tigergraph", "password": "tigergraph"}

Property Graph Databases Supported Only with LlamaIndex

These stores only work with GRAPH_BACKEND=llamaindex (default). The LangChain equivalents are either incompatible with current dependencies or not yet implemented.

Google Cloud Spanner (Graph)

Fully managed relational + graph database with openCypher and Graph Query Language (GQL) support.

  • Cloud only — requires a GCP project, Spanner instance, and database with Spanner Graph enabled
  • No emulator support: the Google Cloud Spanner emulator supports SQL only, not Spanner Graph — use a real GCP Spanner instance
  • Requires: uv pip install -e ".[spanner-extras]" && uv pip uninstall llama-index — llama-index-spanner pulls in the llama-index meta-package as a dependency; uninstall it immediately after to avoid version conflicts (the meta-package pins versions of llama-index-* component packages that can clash with the versions already required by this project)
  • LC is not supported: langchain-google-spanner requires langchain-core<1.0 (incompatible with langchain>=1.0). When a compatible release of langchain-google-spanner is published, Spanner will be promoted to "both LI+LC".

Authentication: set credentials_file for a service-account JSON key file, or rely on Application Default Credentials (GOOGLE_APPLICATION_CREDENTIALS env var or gcloud auth application-default login).

PG_GRAPH_DB=spanner
SPANNER_GRAPH_DB_CONFIG={"project_id": "my-gcp-project", "instance_id": "my-spanner-instance", "database_id": "my-database", "graph_name": "knowledge_graph"}

# With service-account key file:
SPANNER_GRAPH_DB_CONFIG={"project_id": "my-gcp-project", "instance_id": "my-spanner-instance", "database_id": "my-database", "graph_name": "knowledge_graph", "credentials_file": "/path/to/sa-key.json"}

Framework Configuration (Graph-Specific)

Variable
 Default Description
GRAPH_BACKEND llamaindex llamaindex or langchain — ingestion + retrieval backend
KG_EXTRACTOR_BACKEND llamaindex llamaindex (SchemaLLMPathExtractor) or langchain (LLMGraphTransformer)
USE_LC_TEXT_TO_GRAPH false Neo4j + LANGCHAIN_PG_VECTOR_SEARCH=true only: add TextToGraphQueryRetriever back alongside vector+neighborhood. All other LC stores auto-enable text-to-graph (it is their only graph retrieval path). For Neo4j with LANGCHAIN_PG_VECTOR_SEARCH=false (default), text-to-graph is also auto-enabled.
USE_PG_NEIGHBORHOOD true Enable GraphNeighborhoodRetriever (k-hop graph walk from seed entities; Neo4j only; auto-enabled when LANGCHAIN_PG_VECTOR_SEARCH=true)
LANGCHAIN_PG_VECTOR_SEARCH false Enable GraphEntityVectorRetriever (Neo4j only; entity-level vector seeding; auto-enables USE_PG_NEIGHBORHOOD; suppresses text-to-graph unless USE_LC_TEXT_TO_GRAPH=true)
ENABLE_KNOWLEDGE_GRAPH true Set false to disable KG extraction globally (vector-only RAG)

Skipping or Disabling Graph Extraction

Both PG and RDF graph writes can be bypassed while leaving previously ingested graph data intact for search and Q&A.

Global — ENABLE_KNOWLEDGE_GRAPH=false: disables KG extraction and all graph writes on every ingest call. Previously ingested graph data is preserved and still used for hybrid search and AI Q&A.

ENABLE_KNOWLEDGE_GRAPH=false

Vector-only RAG (no graph store at all):

PG_GRAPH_DB=none
ENABLE_KNOWLEDGE_GRAPH=false

Per-ingest — skip_graph=true: skips KG extraction for one request only. Vector and full-text stores are still updated. Available from the UI ("Skip graph extraction" checkbox), REST API (POST /api/ingest, /api/ingest-text, /api/test-sample), MCP tools (ingest_documents, ingest_text, test_with_sample), and the Python API (backend.ingest_documents(skip_graph=True), backend.ingest_text(skip_graph=True)). Also persisted per-datasource in the incremental sync config.

Further Reading