GraphAide

Build knowledge graphs from any document with LLMs and Neo4j. GraphAide is a multi-agentic system that extracts entities and relationships from diverse sources, constructs knowledge graphs, and enables natural language reasoning over them—all with minimal setup.

30-Second Example

from graphaide import GraphAide, ModelConfig, GraphDBConfig

# Setup (once)
graphaide = GraphAide(
    model_config=ModelConfig(provider="openai", model_name="gpt-4o"),
    graphdb_config=GraphDBConfig(uri="neo4j://localhost:7687", username="neo4j", password="password"),
)

# Extract entities and relationships
result = graphaide.extract(file_path="document.pdf", ontology_path="schema.ttl")
print(f"Extracted {len(result.nodes)} entities, {len(result.edges)} relationships")

# Load to Neo4j
graphaide.ingest(file_path="document.pdf", ontology_path="schema.ttl")

# Query the graph
answer = graphaide.query("What are the main topics?")
print(answer.answer)

What It Does

• Extract entities & relationships from PDFs, text, images, or web data
• Ground entities to Wikidata for semantic enrichment
• Load to Neo4j for queryable knowledge graphs
• Reason over graphs with natural language questions
• Track chunk provenance for result attribution

Supported Providers

Documentation

📚 Quick links:

• QUICKSTART.md — 5-minute setup and CLI examples
• ARCHITECTURE.md — Technical deep dive (agents, workflows, patterns)
• REFERENCE.md — Complete API/CLI reference

Install

pip install graphaide

See QUICKSTART.md for detailed setup, Docker, and environment configuration.

Use Cases

• Knowledge Graph Generation — Extract structured data from documents into queryable graphs
• Semantic Search — Find related entities and concepts across datasets
• Domain-Specific Reasoning — Answer questions requiring cross-document reasoning
• Data Integration — Merge disparate sources with automated entity grounding
• Compliance & Audit — Track relationships and dependencies with provenance

Architecture

GraphAide uses a multi-agent architecture orchestrated by LangGraph:

Document → [FileLoader] → [Extractor] → [Nodes4EdgesLoader] 
         → [NodesLoader] → [EdgesLoader] → Neo4j
         ↓
      [VectorDB] ← [Embeddings] ← RAG Context

Factory pattern manages LLMs, vector stores, and graph databases:

• ModelFactory — Loads LLMs from OpenAI, Anthropic, Google, AWS Bedrock, or local LMStudio
• GraphDBFactory — Manages Neo4j/Neptune connections
• VectorStoreFactory — Manages ChromaDB or Qdrant vector indices
• AgentFactory — Builds extraction, loading, and query agents

Agents communicate via KGGenerationState (append-only shared state with reducers):

• raw_text, file_path — Input document
• nodes, edges — Extracted entities and relationships
• rag_context — Ontology constraints and Wikidata metadata
• messages — Status and debug info

Tested Configurations

Run Modes

Support

Please reach out to Sumit.Purohit@pnnl.gov for any questions.

Authors and acknowledgment

Please reach out to Sumit.Purohit@pnnl.gov for any questions.

Citation

If you use GraphAide in your research, please cite:

@inproceedings{purohit2024graphaide,
  title={GraphAide: Advanced Graph-Assisted Query and Reasoning System},
  author={Purohit, Sumit and Chin, George and Mackey, Patrick S and Cottam, Joseph A},
  booktitle={2024 IEEE International Conference on Big Data (BigData)},
  pages={3485--3493},
  year={2024},
  organization={IEEE}
}

The research described in this paper is partially supported by the Resilience Through Data Driven, Intelligently Designed Control (RD2C) Initiative at Pacific Northwest National Laboratory (PNNL) and the United States federal government. Pacific Northwest National Laboratory is a multiprogram national laboratory operated for the US Department of Energy (DOE) by Battelle Memorial Institute under Contract No. DEAC05-76RL01830. PNNL Information Release PNNL-SA205147.

License

Please refer LICENSE and DISCLAIMER files for details.

Project status

Active development