Neo4j GraphRAG enables developers to build graph retrieval augmented generation (GraphRAG) applications using the power of Neo4j and Python. As a first-party library, it offers a robust, feature-rich, and high-performance solution, with the added assurance of long-term support and maintenance directly from Neo4j.

Here’s how to use it with Langtrace:

​
Setup

  1. Install the Langtrace’s SDK and initialize the SDK in your code.
pip install langtrace-python-sdk
  1. Install the Neo4j driver and the Neo4j GraphRAG library.
pip install neo4j
pip install neo4j-graphrag
  1. Setup environment variables:
export LANGTRACE_API_KEY=YOUR_LANGTRACE_API_KEY
export OPENAI_API_KEY=YOUR_OPENAI_API_KEY # this is assuming you're using OPENAI for inference

​
Usage

Initialize Langtrace before creating your Phidata agent:

Python
from langtrace_python_sdk import langtrace  # Must precede other imports
from langtrace_python_sdk.utils.with_root_span import with_langtrace_root_span

from neo4j import GraphDatabase
from neo4j_graphrag.generation import GraphRAG
from neo4j_graphrag.indexes import create_vector_index
from neo4j_graphrag.llm import OpenAILLM as LLM
from neo4j_graphrag.embeddings import OpenAIEmbeddings as Embeddings
from neo4j_graphrag.retrievers import VectorRetriever
from neo4j_graphrag.experimental.pipeline.kg_builder import SimpleKGPipeline

langtrace.init()

AUTH = (os.getenv("NEO4J_USERNAME"), os.getenv("NEO4J_PASSWORD"))

neo4j_driver = GraphDatabase.driver(os.getenv("NEO4J_URI"), auth=AUTH)

Create and run the RAG pipeline:


ex_llm=LLM(
   model_name="gpt-4o-mini",
   model_params={
       "response_format": {"type": "json_object"},
       "temperature": 0
   })

embedder = Embeddings()

@with_langtrace_root_span("run_neo_rag")
async def run_rag():
    # 1. Build KG and Store in Neo4j Database
    kg_builder_pdf = SimpleKGPipeline(
        llm=ex_llm,
        driver=neo4j_driver,
        embedder=embedder,
        from_pdf=True
    )
    await kg_builder_pdf.run_async(file_path='data/contract.pdf')
    
    create_vector_index(neo4j_driver, name="text_embeddings", label="Chunk",
                    embedding_property="embedding", dimensions=1536, similarity_fn="cosine")

    # 2. KG Retriever
    vector_retriever = VectorRetriever(
        neo4j_driver,
        index_name="text_embeddings",
        embedder=embedder
    )

    # 3. GraphRAG Class
    llm = LLM(model_name="gpt-4o")
    rag = GraphRAG(llm=llm, retriever=vector_retriever)

    # 4. Run
    response = rag.search("can you summarize the termination clause from the contract.")
    print(response.answer)
    
asyncio.run(run_rag())

​
What’s being traced?

With Langtrace, the following operations are automatically traced:

Knowledge Graph Building: -Document ingestion and processing -Entity extraction and relationship creation -Vector embedding generation

Search and Retrieval:

  • Vector similarity search operations
  • Subgraph extraction for context
  • Retrieved document chunks

LLM Generation:

  • Prompt construction with retrieved context
  • Model completion generation
  • Response processing

View all these trace details in the Langtrace dashboard:

traces traces

​
Resources