Get current state of the Langgraph graph outside of the Nodes (inside the main forloop)

[Sulomus]

We can access the state of the graph inside each of the Nodes such as like this:

async def thought(state: PlanExecute):
    # example node
    state["messages"] ...

But I want to access the state here:

app = workflow.compile()

async for event in app.astream_events(input={"message_history": conversation_buffer.messages}, version="v2"):
    app.state() 

Where is the state saved inside of the graph? I debugged the app and looked into the object itself and couldnt find it.

I know I can use persistence and in MemoryCheckpointer - but I am afraid it will take up too much space in memory (and thus my Azure Functions will be too expensive).

I wan to just access the state vai the object app 😄
How can I do that?

[vbarda]

@Sulomus you can use app.get_state(), see a how-to guide here https://langchain-ai.github.io/langgraph/how-tos/human_in_the_loop/time-travel/

[Sulomus]

I know that MemoryCheckpointer can be used - but as I said I am afraid it will take up too much space in memory.

But even ignoring that fact - I tried it and got Object of type AsyncCallbackManager is not JSON serializable langgraph.

The error is not in my code - it is raised in the app.stream() - this issue is discussed also here in this GitHub discussion.

Is there any other way? Or Memory Checkpointer is the only one?

[nfcampos]

Hi, I'd suggest using stream_mode="values". That way each of the chunks yielded in for loop will be the state at that time, and a new chunk will be emitted any time it changes.

[Harsh-Singh-Rajput]

Hi everyone,
I am also trying to view the agent state during the async graph stream as described in the guide shared by @vbarda . However, I am encountering an error in the event stream, even though the normal stream works fine when I remove the agent.get_state() method. Can someone help me determine if this is a bug or if I am missing something? I have tried both agent.aget_state() and agent.get_state() methods, but I am getting the same error with both.

Does this help clarify your issue?error for both.
Sample code

Error
Backend\utils\llm_response_handlers.py", line 75, in message_generator | snapshot = agent.aget_state(config=kwargs) | ^^^^^^^^^^^^^^^^ | AttributeError: 'Agent' object has no attribute 'aget_state'

[sitongye]

spelling error