Repeated human messages in MessageGraph

[markpeace]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

DEBUG=True

    from agent.memory import Memory
    memory = Memory(payload)

    from agent.jsonencoder import json_parse_chain
    from agent.agent_main import Chain_Main_Agent

    chain_main_agent = Chain_Main_Agent(memory)

    
    from agent.toolset import tool_executor, converted_tools

    from langgraph.prebuilt import ToolInvocation
    import json
    from langchain_core.messages import FunctionMessage

    def trim_messagelog(messages):
        if DEBUG: print("trim_messagelog called");   
        from pprint import pprint

        pprint(messages)                                    
        return messages[-1]

    def call_main_agent(messages):   
        response = chain_main_agent.invoke({})
        
        if DEBUG: print("call_main_agent called");                                        

        return response                                     

    def use_tool(messages):    
        last_message = messages[-1]                             
        action = ToolInvocation(
            tool=last_message.additional_kwargs["function_call"]["name"],
            tool_input=json.loads(last_message.additional_kwargs["function_call"]["arguments"]),
        )
        response = tool_executor.invoke(action)
        function_message = FunctionMessage(content=str(response), name=action.tool)

        if DEBUG: print("Suggesting Tool to use..."+action.tool);                                                                                

        return function_message

    def render_output(messages):

        import json

        response = json_parse_chain.invoke({"conversation":messages, "thread_id": memory.thread_id})

        if DEBUG: print("Rendering output");   

        from langchain_core.messages import AIMessage                                     

        response = json.dumps(response)
        return AIMessage(content=response)

    from langgraph.graph import MessageGraph, END
    workflow = MessageGraph()

    workflow.add_node("trim_messagelog", trim_messagelog)
    workflow.add_node("main_agent", call_main_agent)
    workflow.add_node("use_tool", use_tool)
    workflow.add_node("render_output", render_output)

    workflow.set_entry_point("trim_messagelog")

    def should_continue(messages):                                        
        last_message = messages[-1]
        if "function_call" not in last_message.additional_kwargs: return "render_output"
        else: return "continue"


    workflow.add_edge("trim_messagelog", "main_agent")
    workflow.add_conditional_edges(
        "main_agent", should_continue, 
        {
            "continue": "use_tool", 
            "render_output":"render_output",
            "end": END
        }
    )
    workflow.add_edge('use_tool', 'main_agent')
    workflow.add_edge('render_output', END)


    app = workflow.compile(checkpointer=memory.checkpoints)
    
    from langchain_core.messages import HumanMessage
    '''
    input = payload.get("input") or "What is Rise for?"
    inputs = [HumanMessage(content=input)]

    response = app.invoke(inputs, {"configurable": {"thread_id": memory.thread_id} } )
    
    print(response[-1].content)

    
    '''
    
    inputs = [HumanMessage(content="My name is Mark")]
    response = app.invoke(inputs, {"configurable": {"thread_id": memory.thread_id} } )
    print(response[-1].content)

    inputs = [HumanMessage(content="What is my name?")]
    response = app.invoke(inputs, {"configurable": {"thread_id": memory.thread_id} } )
    print(response[-1].content)
    
    inputs = [HumanMessage(content="What is your name?")]
    response = app.invoke(inputs, {"configurable": {"thread_id": memory.thread_id} } )
    print(response[-1].content)
    
    return response[-1].content

Description

When I inspect the cumulative content of MessageGraph, I find that the HumanMessage is always repeated twice? How do stop this happening?

System Info

langchain==0.1.9
langchain-community==0.0.24
langchain-core==0.1.26
langchain-fireworks==0.0.1
langchain-openai==0.0.7

[dosubot[bot]]

Hey @markpeace, great to see you back! Hope everything's been going well on your end.

The issue you're experiencing with the HumanMessage being repeated twice in the cumulative content of MessageGraph might be due to the way the add_message method is implemented in your Memory class or how the messages are managed in your workflow.

In your provided code, it seems like you're invoking the app with new inputs multiple times. Each time you invoke the app, it processes the inputs and adds them to the MessageGraph. If you're not clearing the MessageGraph or the memory between invocations, the same HumanMessage could be added multiple times, leading to the repetition you're seeing.

Here's a simplified example to illustrate this:

inputs = [HumanMessage(content="My name is Mark")]
response = app.invoke(inputs, {"configurable": {"thread_id": memory.thread_id} } )
print(response[-1].content)

inputs = [HumanMessage(content="What is my name?")]
response = app.invoke(inputs, {"configurable": {"thread_id": memory.thread_id} } )
print(response[-1].content)

In this example, the HumanMessage "My name is Mark" is added to the MessageGraph during the first invocation. During the second invocation, the HumanMessage "What is my name?" is added, but "My name is Mark" is still in the MessageGraph from the previous invocation, leading to it being repeated in the cumulative content.

To prevent this, you could clear the MessageGraph or the memory between invocations. However, this would also remove any context or state from previous invocations, which might not be what you want if you're trying to maintain a conversation with the user.

If you want to maintain the conversation context while preventing the repetition of HumanMessage, you might need to modify your workflow or the Memory class to handle this. For example, you could add a method to the Memory class to remove the last HumanMessage before adding a new one, or modify the invoke method of your MessageGraph to only add new HumanMessage if it's different from the last one.

Please review your Memory class and the workflow in your provided code to see if they're handling the messages in a way that could lead to this repetition. If you're still having trouble, please provide more details about your Memory class and how you're using the MessageGraph in your application, and I'll be glad to help further.

For more information on how the MessageGraph and Memory class work, you can refer to the LangChain documentation and the source code:

• LangChain Documentation
• MessageGraph Source Code
• Memory Class Source Code

Sources

• libs/langchain/langchain/schema/chat_history.py
• libs/core/tests/unit_tests/fake/memory.py

About Dosu

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.