langgraph/concepts/memory/

[giscus[bot]]

langgraph/concepts/memory/

Build language agents as graphs

https://langchain-ai.github.io/langgraph/concepts/memory/

[gautham-balraj]

Great !!

[daflood]

Are there any examples of how to use the PostgresStore class? BaseStore works great but I am having the hardest time with PostgresStore.

[Baptiste-ms]

Did you managed to get the Postgre memory components working ?

[dprat0821]

I believe it is worth reconsidering the current approach to managing the message state, as well as how nodes update this state. The conventional approach for state management (e.g., Redux) allows consumers to specify an action type (e.g., ADD_MESSAGE, REMOVE_MESSAGE) along with a corresponding payload. This design enables a single state to handle multiple action types, supporting not only basic CRUD operations but also more complex workflows.

In the current design, the state only supports a single reducer named add_message. Although this reducer allows handling deletion actions through the use of a special "RemoveMessage" symbol, it imposes limitations. This pattern is less scalable if we want to introduce new ways to manipulate messages in the future.

[hinthornw]

1. It's easy to write your own reducer. You can even reuse/extend if you want to still use the existing deduplication logic:

def my_reducer(existing, new):
     .... can do stuff here
     reduced = add_messages(existing, new)
    .... or here
    return reduced

2. add_messages supports updates, insertions, and deletions. You can do anything with these primitives.

[siddg97]

Where can one find all the different DB backed memory stores that are currently supported? I only see PostgresStore in the library reference, are there other options too?

[pranitsawant80]

I think We can use Redis as well as memory
https://langchain-ai.github.io/langgraph/how-tos/persistence_redis/?h=redis