ArchitectureInfrastructure

Core Infrastructure Components

  1. Vector Stores: Cognee supports multiple vector store backends, which handle vector embeddings for semantic search and context retrieval. Supported vector stores include:

    • LanceDB (default local vector database)
    • Qdrant
    • Weaviate
    • PGVector (PostgreSQL extension)
    • Milvus
  2. Graph Databases: Cognee builds a knowledge graph from extracted entities and relationships. Supported graph stores include:

    • NetworkX (default in-memory graph)
    • Neo4j (for production-scale graph queries and persistence)
  3. Relational Databases: Cognee supports:

    • SQLite (default local relational database)
    • PostgreSQL
  4. LLM Providers and Cognitive Services: Cognee leverages LLMs to process, classify, and summarize text. By default, it expects an OpenAI-compatible API key but can also integrate with other providers like Anyscale or Ollama. Configure these in environment variables or .env files.

  5. Visualization: Optionally, integrate Graphistry for advanced graph visualization for NetworkX.

System Requirements

  • Python Environment: Cognee supports Python 3.9+.
  • Container Runtime: For Docker-based deployments, you’ll need Docker and Docker Compose.
  • Node.js & npm: Required if you intend to run the frontend UI locally.
  • Database Services: Depending on your chosen backend, ensure you have access to a running PostgreSQL, Neo4j, Qdrant, or Weaviate instance if not using defaults.