[LangGraph] Memory

Memory

Memory is key for any agentic conversation which is Multi-Turn or Multi-Agent colloboration conversation and more so if it spans multiple days. The 3 main aspects of Agents are:
1. Tools
2. Memory
3. Planners

Explore External Store for memory

There are 2 types of memory for AI Agents, short term and long term memory which can be explained below.
Further reading can be at this link

Conversation memory can be explained by this diagram below which explains the turn by turn conversations which needs to be accessed by agents and then saved as a summary for long term memory

Create an external Memory persistence

We can leverage multi-thread, multi-session persistence to Chat Messages. Ideally you will leverage persistence like Redis Store etc to save messages per session.

Memory Management

• We can have several Patterns
  • We can have each Agents with it's own Session memory
  • Or we can have the whole Graph have a combined memory in which case each agent will get it's own memory

The MemorySaver or the Store have the concept of separating sections of memory by Namespaces or by Thread ID's and those can be leveraged to either 1/ Use the graph level message or memory 2/ Ecah agent can have it's own memory via space in saver or else having it's own saver like we do in the ReACT agent