Is LangGraph compatible with RunnableWithMessageHistory?

[danilopeixoto]

The RunnableWithMessageHistory appears to pass the proper input message and chat history to graph:

However, the entrypoint agent receives a null input message and proper chat history:

There are unhandled exceptions from streaming events (not sure if related):

langchain-ai/langchain#17199

NotImplementedError in LogStreamCallbackHandler.on_chain_end callback: NotImplementedError("Trying to load an object that doesn't implement serialization: {'lc': 1, 'type': 'not_implemented', 'id': ['builtins', 'object'], 'repr': '<object object at 0x7fafbdf43320>'}")

I think this is coming from __start__ runnables:

Versions

python                    3.11.0
langchain                 0.1.8
langchain-community       0.0.21
langchain-core            0.1.26
langchain-openai          0.0.6
langgraph                 0.0.26
langserve                 0.0.42
langsmith                 0.1.5

[BrownsugarZeer]

Hi @danilopeixoto

I'm interested in this question. Could you please tell me why you want to combine the two?

[hodgesz]

Seems like overlap, since you can also store message history in AgentState as part of LangGraph.

[danilopeixoto]

My goal was to attach a remote storage to RunnableWithMessageHistory.

[BrownsugarZeer]

Agree with @hodgesz . I thought RunnableWithMessageHistory and langgraph's ckeckpoint do a similar thing(all most), so used one of them would be fine. The only thing I found was checkpoint didn't handle the context length, so llm token would be ran out in some day(same as RunnableWithMessageHistory).

[hodgesz]

Maybe attach remote storage to LangGraph instead e.g., Redis, Postgres, etc.?

…

On Sun, Apr 28, 2024, 6:45 PM Danilo Peixoto ***@***.***> wrote: My goal was to attach a remote storage to RunnableWithMessageHistory. — Reply to this email directly, view it on GitHub <#141 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AANO4ZIJKP3UBFTW6YA3VTTY7WJ3VAVCNFSM6AAAAABDWEGRQKVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TENJWGU3TQ> . You are receiving this because you commented.Message ID: ***@***.***>

[danilopeixoto]

Thanks for suggestion. I found now the introduction of remote checkpoints:

• Graph State Persistence? #350 (comment)
• https://github.com/langchain-ai/langgraph/tree/main/langgraph/checkpoint

We may close the discussion.