Memory Layer Architecture

Vector storage and semantic search powered by PostgreSQL with pgvector.

Key Components

  • Vector Store: PostgreSQL + pgvector for 1536D embeddings
  • Embedding Generation: Automatic via OpenAI text-embedding-3
  • Semantic Search: Cosine similarity with IVFFlat index
  • Metadata: JSONB for flexible attributes
  • Caching: Redis for hot embeddings