How can I create a custom chat history as I want?

[HoangNguyen689]

Our bot has multiple steps, and we want to test each step separately.
With AgentExecutor and persistent storage (e.g., DynamoDB), we can easily create and pass the chat history when invoking the agent executor.

history_store.add_user_message(user_msg)
history_store.add_a_message(ai_msg)

agent_executor.invoke({
    "input": msg,
    "today_date": now,
    "chat_history": history_store.messages,
})

How can I add the chat history when using a checkpointer?

[vbarda]

@HoangNguyen689 check out these guides

https://langchain-ai.github.io/langgraph/how-tos/persistence/
https://langchain-ai.github.io/langgraph/concepts/memory/#short-term-memory

[HoangNguyen689]

@vbarda
Thanks for your references.
But before I ask this question, I have already read all the docs about persistence memory :)
Unlike agent_executor's chat_history, which stores the messages sequentially and is readable in the database,
the checkpointer is a black box. So, it's hard to add the messages that follow a scenario or specify the current agent responding to the current task.
That's the reason I asked the above question. Without specifying the scenarios, it is impossible to test cases as we want. Also, the evaluations on Langsmith become useless.
For Langsmith, we want a better integration from Langgraph. At least, the evaluation should be worked with Langgraph.

[vbarda]

@HoangNguyen689 there is a few things here :)

Unlike agent_executor's chat_history, which stores the messages sequentially and is readable in the database,
the checkpointer is a black box. So, it's hard to add the messages that follow a scenario or specify the current agent responding to the current task.

i am not sure i understand what your exact use case is, but have you tried using graph.update_state()? you should be able to add messages to the existing history

graph.update_state(config, {"messages": [HumanMessage("my new message")]})

hope this helps!

Also, the evaluations on Langsmith become useless.
For Langsmith, we want a better integration from Langgraph. At least, the evaluation should be worked with Langgraph.

can you please elaborate on this? what are the issues you're running into langsmith/langgraph integration and evaluations? we'd love to improve!