KGraphMemory

An in-memory knowledge graph system designed for AI agents to store, manage, and query structured knowledge during interactions with people and task execution.

Overview

KGraphMemory provides a high-performance, in-memory knowledge graph implementation that combines RDF triple stores with vector databases to enable AI agents to:

• Store and retrieve structured knowledge using semantic web standards
• Perform vector-based similarity search for semantic understanding
• Maintain context and relationships across agent interactions
• Execute complex SPARQL queries on knowledge representations
• Support multi-vector embeddings for different semantic aspects

Key Features

Dual Storage Architecture

• RDF Store: Uses pyoxigraph for fast SPARQL queries and semantic reasoning
• Vector Store: Uses Qdrant for similarity search and semantic retrieval
• Synchronized Operations: Automatic synchronization between stores

Ontology-Driven Design

• Built on the Haley AI Knowledge Graph ontology (haley-ai-kg-0.1.0.owl)
• Uses VitalSigns library for OWL parsing and Python class generation
• Type-safe edge relationships with source/destination constraints
• Support for vectorization annotations (isVectorized, isVectorizedType)

Multi-Vector Support

• Multiple semantic vectors per object (type, description, content, general)
• Configurable vector mappings for different object types
• Precise semantic search across different vector types

Agent-Centric Features

• KGraphMemory: Manages multiple knowledge graphs with shared embedding models
• Cross-graph operations: Search and query across multiple knowledge graphs
• Real-time updates: Dynamic knowledge updates during agent interactions
• Efficient querying: Hybrid search combining SPARQL and vector similarity

Architecture

KGraphMemory (Manager)
├── KGraph (Individual Knowledge Graph)
│   ├── KGraphRDFDB (pyoxigraph)
│   ├── KGraphVectorDB (Qdrant)
│   └── Vector Mappings
├── Shared Embedding Model
└── Cross-Graph Operations

Use Cases

• Conversational AI: Maintain context and knowledge across chat interactions
• Task Execution: Store and retrieve procedural knowledge for agent tasks
• Knowledge Management: Organize and query structured information
• Semantic Search: Find relevant information using natural language queries
• Relationship Modeling: Represent complex entity relationships and dependencies

Getting Started

from kgraphmemory import KGraphMemory
from sentence_transformers import SentenceTransformer

# Initialize with embedding model
embedding_model = SentenceTransformer('paraphrase-MiniLM-L3-v2')
memory = KGraphMemory(embedding_model)

# Create a knowledge graph
graph = memory.create_kgraph("agent_kb", "http://example.org/agent")

# Add knowledge objects
graph.add_object(entity)

# Search across graphs
results = memory.search_across_kgraphs("find information about...")

Documentation

See the docs directory for detailed documentation on:

• API Reference
• Ontology Structure
• Vector Mapping Configuration
• SPARQL Query Examples
• Integration Patterns