Skip to content

Overview

pgraf provides a high-performance property graph database implementation built on top of PostgreSQL, with specific optimizations for AI agent use cases.

What is a Property Graph?

A property graph is a graph data structure that consists of:

  • Nodes: Entities with properties and labels
  • Edges: Relationships between nodes with labels and properties
  • Properties: Key-value pairs attached to both nodes and edges
  • Labels: Categorization tags for nodes and edges

pgraf enhances this model with:

  • Strong typing via Pydantic models
  • Vector embeddings for semantic search
  • Full-text content storage and retrieval
  • Asynchronous API for high concurrency

Use Cases

pgraf is particularly well-suited for:

  • AI agent knowledge graphs
  • Semantic search applications
  • Document networks with relationships
  • Complex data modeling with typed relationships
  • Applications requiring both graph and vector search capabilities

Architecture

pgraf uses:

  • PostgreSQL as the storage backend
  • pgvector for vector operations
  • Pydantic for model validation
  • Async Python for high-performance operations

The architecture consists of several key components:

  • Graph Engine: Core graph operations (nodes, edges, traversals)
  • Vector Search: Embedding management and similarity search
  • Type System: Strongly-typed models for graph elements
  • Query Layer: Composable query building system