Why and when should I use memory saver

[Hollum]

Hello! I am new to Langgraph so these questions may be silly but ya'll manage 🙏 I have created a simple graph where the LLM just asks questions. After five questions provided to the user it should end the conversation.

Every time the user sends an input i store the message in my postgres db (using prisma) and pass the message history to my graph. I initialise the graph with the previous history like the following snippet.

const awesomeFunction = (chatHistory: BaseMessage[], input: string) => {

...code

const GraphAnnotation = Annotation.Root({
    input: Annotation<string>(),
    chat_history: Annotation<BaseMessage[]>({
      reducer: messagesStateReducer,
      default: () => chatHistory, 
    }),
    context: Annotation<string>(), 
    answer: Annotation<string>(),
    questionCount: Annotation<number>({
      reducer: (state: number, update: number) => state + update,
      default: () => chatHistory.length, 
    }),
  });
}

I am having a hard time seeing the difference between this and using the memory saver/checkpoint-postgres. What benefits would I be missing?

[hinthornw]

Using a checkpointer would replace your need for manually managing in prisma. It also manages checkpointing of hte entire state, so if you want more values in the state or want to add more complexity to the graph and include things like human-in-the-loop approvals, that would all be supported.