What are the 3 different types of memory on an agent?

In this video, you’ll see how Neo4j can be utilized as an agentic memory store for AI applications, focusing on three distinct layers of memory:

Short-Term Memory: This represents the immediate environment and current context of a conversation. In an agentic space, it is captured through conversational history—the exchange of messages between the user and the agent within a single session.

Long-Term Memory: This layer is created by extracting entities (people, places, things) and their relationships from short-term memory and storing them in a separate space. This allows information to be shared across different sessions and across multiple agents, enabling them to learn from one another over time.

Reasoning Memory: This layer preserves “reasoning traces,” which include the tasks an agent was asked to perform, the tools it used, and the sequence of its actions. It provides an auditable and traceable history of how an agent arrived at a particular result, which is crucial for self-learning and improving response quality.

The video features a demo called the Lenny’s Memory app, which uses these memory types to interrogate data from episodes of Lenny’s podcast. Users can learn how to implement these systems through the GraphAcademy’s Agentic Memory in #Neo4j online course or by exploring the open-source Neo4j agent memory package on GitHub.

00:07 – The importance of the memory layer for AI applications
01:40 – Defining Short-Term Memory
02:22 – Defining Long-Term Memory
03:06 – Defining Reasoning Memory as the preservation of “reasoning traces,”
04:23 – A demonstration of the Lenny’s Memory app,
06:08 – Visualizing how these memories are stored in Neo4j
07:14 -Resources