In most AI assistants, memory ends where the context window stops. Stateless “amnesiac” agents repeat questions, forget past promises, and fail to build trust over time. In this session, we will show how to turn Neo4j into a long‑term memory layer for AI agents, enabling persistent, evolving personas that learn from every interaction.

Using a customer success copilot as a concrete use case, the session will model conversations, events, and user traits as a graph, then demonstrate how an agent will query and update this memory to adapt its behavior over weeks and months. You will see how to design a graph schema for agent memory, orchestrate calls between the LLM and Neo4j, and implement patterns for recall, summarization, and “self‑reflection.” By the end of the session, you will understand how to move from short‑lived chat history to durable, graph‑native memory architectures.