langgraph/how-tos/memory/add-summary-conversation-history/

[giscus[bot]]

langgraph/how-tos/memory/add-summary-conversation-history/

Build language agents as graphs

https://langchain-ai.github.io/langgraph/how-tos/memory/add-summary-conversation-history/

[chandanmaurya0]

After deleting the previous message in the State , I am still having Removed messages in the state like below. It is causing problem in next conversation.

 RemoveMessage(content='', id='run-05986a08-f4b8-493a-a620-f4daee8f00f2-0'),
 RemoveMessage(content='', id='732df2ec-b524-429f-9090-70ae5e19da33'),
 RemoveMessage(content='', id='run-6dbd8014-d1b4-4adb-8ec5-84a78259e888-0'),
 RemoveMessage(content='', id='07d29b9d-ad8c-4a40-8f90-5bf0fb68df49'),
 RemoveMessage(content='', id='run-f3c19cbe-2a23-493a-936b-7702545cd72b-0'),
 HumanMessage(content='i like Indian Cricket team and Virat Kohil as favourite player', id='61720457-25c8-443e-87b2-e1e94faeb7b4')] ```

[hwchase17]

I just ran the notebook and it worked for me. Are you running the example as is, or did you modify something?

[chandanmaurya0]

I tried creating a new python env and ran above notebook , then it worked as expected.
But when I am trying to run above notebook with a python env with below library installed, then I am getting RemoveMessage in the State even after deleting.

Here is my pyproject.toml file

name = "lang-memgpt"
version = "0.0.1"
description = "A simple memory-enabled agent for agents on LangGraph cloud."
authors = ["William Fu-Hinthorn <[email protected]>"]
readme = "README.md"

[tool.poetry.dependencies]
python = ">=3.9.0,<3.13"
langgraph = "^0.1.0"
langchain-fireworks = "^0.1.3"
# Feel free to swap out for postgres or your favorite database.
langchain-pinecone = "^0.1.1"
jsonpatch = "^1.33"
dydantic = "^0.0.6"
pytest-asyncio = "^0.23.7"
trustcall = "^0.0.4"
langchain = "^0.2.6"
langchain-openai = "^0.1.10"
langchain-anthropic = "^0.1.19"
pydantic-settings = "^2.3.4"
langgraph-sdk = "^0.1.23"
langchain-community = "^0.2.6"
tavily-python = "^0.3.3"
tiktoken = "^0.7.0"
python-dotenv = "^1.0.1"
pymongo = "^4.8.0"
langchain-groq = "^0.1.6"

[tool.poetry.group.dev.dependencies]
ruff = "^0.4.10"
mypy = "^1.10.0"
pytest = "^8.2.2"
langgraph-cli = "^0.1.43"

[tool.ruff]
lint.select = [
    "E",    # pycodestyle
    "F",    # pyflakes
    "I",    # isort
    "D",    # pydocstyle
    "D401", # First line should be in imperative mood
]

[tool.ruff.format]
docstring-code-format = true
docstring-code-line-length = 80

[tool.ruff.lint.per-file-ignores]
"tests/*" = ["D", "E501"]

[tool.ruff.lint.pydocstyle]
convention = "google"

[build-system]
requires = ["poetry-core"]
build-backend = "poetry.core.masonry.api"

[tool.pytest.ini_options]
asyncio_mode = "auto"

[tudoanh]

Won't work for graph that contains tool calls

[hinthornw]

You'd have to only truncate at demarcated lines (so keep ai message and accompanying tool messages together)

[MNovruzov]

You'd have to only truncate at demarcated lines (so keep ai message and accompanying tool messages together)

Hi, please provide a code example if possible

[Harshwardhanpjadhav]

This Summarization code gives error when integrated with tools with tools node

class State(MessagesState):
    summary: str

# Initialize the model to use for conversation and summarization
# model = ChatAnthropic(model_name="claude-3-haiku-20240307")

# Define the logic to call the model (replaces your assistant function)
def assistant(state: State):
    summary = state.get("summary", "")
    if summary:
        # Include the summary as a system message if it exists
        system_message = f"Summary of conversation earlier: {summary}"
        print(f"--------system summary-------- \n{system_message} \n ")
        
        print(f"--------state messages--------\n{state['messages']} \n")
        
        messages = messages = [SystemMessage(content=system_message)] + state["messages"]
        
        print(f"--------summary + state messages--------\n{state['messages']}\n")

    else:
        system_message = '''
        You are a helpful assistant.
        Answer the users question in short dont write long context 
        '''
        messages = state["messages"]

    response = llm_with_tools.invoke([SystemMessage(content=system_message)] + messages)
    return {"messages": [response]}

def determine_next_node(state: State) -> Literal["tools", "summarize_conversation", END]:
    messages = state["messages"]
    last_message = messages[-1]
    print(f"----------- Tools last_message ------------ \n {last_message}")
    print(f"----------- Messages ------------ \n {state['messages']}")

    if last_message.tool_calls:
        return "tools"
    # Summarize if there are more than six messages
    elif len(messages) > 6:
        return "summarize_conversation"
    return END

# Define the logic for summarizing the conversation
def summarize_conversation(state: State):
    summary = state.get("summary", "")
    if summary:
        # Extend the existing summary
        summary_message = (
            f"This is the summary of the conversation to date: {summary}\n\n"
            "Extend the summary by taking into account the new messages above:"
        )
    else:
        # Create a new summary
        summary_message = "Create a summary of the conversation above:"

    messages = state["messages"] + [HumanMessage(content=summary_message)]
    response = llm_with_tools.invoke(messages)

    # Remove all but the last two messages for context
    delete_messages = [RemoveMessage(id=m.id) for m in state["messages"][:-2]]

    return {"summary": response.content, "messages": delete_messages}


workflow = StateGraph(State)

workflow.add_node("assistant",assistant)
workflow.add_node("tools", ToolNode(tools)) 
workflow.add_node("summarize_conversation", summarize_conversation)
# Add tools node

workflow.add_edge(START, "assistant")  # from Start to assistant node

# Add conditional logic to decide whether to use tools, summarize, or end
workflow.add_conditional_edges(
    "assistant",  # Node after which the condition is checked
    determine_next_node  # Function to determine the next step
)

workflow.add_edge("tools", "assistant") 
# Add edges from tools and summarization to END
# workflow.add_edge("assistant", END)

graph = workflow. Compile(checkpointer=checkpointer)

from IPython.display import Image, display
display(Image(graph.get_graph().draw_mermaid_png()))

user_id = str(input("Enter user ID: "))
config = {"configurable": {"thread_id": user_id}}

res = graph.invoke({"messages": [("human", "get todays weather")]}, config)

for m in res['messages']:
    m.pretty_print()

Error

BadRequestError: Error code: 400 - {'type': 'error', 'error': {'type': 'invalid_request_error', 'message': 'messages.0: tool_result block(s) provided when previous message does not contain any tool_use blocks'}}