langgraph/how-tos/cross-thread-persistence/

[giscus[bot]]

langgraph/how-tos/cross-thread-persistence/

Build language agents as graphs

https://langchain-ai.github.io/langgraph/how-tos/cross-thread-persistence/?ref=blog.langchain.dev

[ITSAIDI]

It's better if we can store this Memory in such as local file. I use a .py file and with this code, every time i run the .py the memory gets initialized so I lost the information stored in previous runs !

[Denis-root]

Te entiendo, solo para agregar al parecer estos ejemplos son pruebas de concepto y creo que su uso son como para entornos de jupyter Notebook o Google collabs

[cividanieltorres]

Hello,

I think you are getting this behaviour because the memory store used here is InMemoryStore. This means that every time you rerun, the memory would initiate from scratch, losing all previous information. If you want to have real persistence, I believe the appropriate way of implementing this right now would be with PostgresStore. This would allow you to setup a Postgres where you could connect every time you rerun your graph.

Remember these are just examples for testing purposes, not for production environments.

Hope this helps.

[hinthornw]

Yupp! As the name suggests, this is "in memory" - desigend to be run ephemerally in a single session. For prod use cases we recommend one of hte other stores (such as the postgres one)

If you want to add some light persistence to it:

from collections import defaultdict
from typing import Any, Optional, Dict, Tuple
from langgraph.checkpoint.memory import PersistentDict
from langgraph.store.memory import InMemoryStore

class PersistentMemoryStore:
    def __init__(
        self, 
        data_file: str = "store_data.pkl",
        vector_file: str = "store_vectors.pkl",
        index: Optional[Dict[str, Any]] = None
    ) -> None:
        # Create the underlying InMemoryStore for operations
        self.store = InMemoryStore(index=index)
        
        # Create persistent dictionaries for data and vectors
        self.data_dict = PersistentDict(filename=data_file)
        self.vector_dict = PersistentDict(filename=vector_file)
        
        # Load existing data if available
        self._load_persistent_data()
        
        # Replace store's in-memory dicts with our persistent ones
        self.store._data = self.data_dict
        self.store._vectors = self.vector_dict

    def _load_persistent_data(self) -> None:
        """Load data from persistent storage files"""
        for container, name in [(self.data_dict, "data"), (self.vector_dict, "vectors")]:
            try:
                container.load()
            except FileNotFoundError:
                # It's okay if file doesn't exist yet
                pass
            except (EOFError, ValueError) as e:
                raise RuntimeError(
                    f"Failed to load store {name} from {container.filename}. "
                    "This may be due to changes in the stored data structure."
                ) from e
            except Exception as e:
                raise RuntimeError(
                    f"Unexpected error loading store {name} from {container.filename}: {str(e)}"
                ) from e

    def __enter__(self):
        return self.store

    def __exit__(self, exc_type, exc_val, exc_tb):
        self.close()

    def close(self) -> None:
        """Close and sync the persistent dictionaries"""
        self.data_dict.close()
        self.vector_dict.close()

# Create a persistent store that automatically handles file I/O
with PersistentMemoryStore(
    data_file="my_store_data.pkl",
    vector_file="my_store_vectors.pkl"
) as store:
    # Use the store as a regular InMemoryStore
    store.put(("users",), "user1", {"name": "Alice"})
    store.put(("users",), "user2", {"name": "Bob"})
    
    # Data will be automatically saved when the context exits

[grega913]

Getting the error when running code above. Can someone explain what is happening here.

KeyError Traceback (most recent call last)
Cell In[15], line 7
1 # Create a persistent store that automatically handles file I/O
2 with PersistentMemoryStore(
3 data_file="my_store_data.pkl",
4 vector_file="my_store_vectors.pkl"
5 ) as store:
6 # Use the store as a regular InMemoryStore
----> 7 store.put(("users"), "user1", {"name": "Alice"})
8 store.put(("users"), "user2", {"name": "Bob"})
10 # Data will be automatically saved when the context exits

File c:\Users\gs\Documents\AI24\LangChain_202409\lg_my_examples.venv\Lib\site-packages\langgraph\store\base_init_.py:880, in BaseStore.put(self, namespace, key, value, index, ttl)
875 if ttl not in (NOT_PROVIDED, None) and not self.supports_ttl:
876 raise NotImplementedError(
877 f"TTL is not supported by {self.class.name}. "
878 f"Use a store implementation that supports TTL or set ttl=None."
879 )
--> 880 self.batch(
881 [
882 PutOp(
883 namespace,
884 str(key),
885 value,
...
409 created_at=datetime.now(timezone.utc),
410 updated_at=datetime.now(timezone.utc),
411 )

KeyError: 'users

[ITSAIDI]

Thanks @cividanieltorres , is There Any Tuto to use PostgresStore with LangGraph ?

[RollsChris]

https://langchain-ai.github.io/langgraph/how-tos/persistence_postgres/

[lakshaytalkstomachines]

Why Can't this be just a node in graph that queries the db and loads the user data into State?

[hinthornw]

Are you asking why this is in the same node as the llm vs. in two nodes?

Your suggestion also works, we just didn't see the need to create a separate node for a how-to

[KartikeyShaurya36]

hi everyone , do anybody know how to clear the history of a thread

[edisonwd]

https://langchain-ai.github.io/langgraph/how-tos/memory/delete-messages/

[himaniMaxHive]

If there is any example, using PostgresStore, it will be very helpful.

[siddg97]

Is there a store with some other DB than Postgres? E.g. MongoDB, DynamoDB etc.