LangGraph - Plan-and-Execute Error - Need Help

[mraguth]

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

Reference:
https://langchain-ai.github.io/langgraph/tutorials/plan-and-execute/plan-and-execute/#define-our-execution-agent

Description

At initial steps,
agent_executor.invoke({"messages": [("user", "who is the winnner of the us open")]})

I am getting this error. Please help:
KeyError: "Input to ChatPromptTemplate is missing variables {'{messages}'}. Expected: ['{messages}'] Received: ['messages', 'is_last_step']\nNote: if you intended {{messages}} to be part of the string and not a variable, please escape it with double curly braces like: '{{{messages}}}'."

System Info

Name: langgraph
Version: 0.2.23

[dosubot[bot]]

To resolve the KeyError in LangGraph when invoking agent_executor.invoke({"messages": [("user", "who is the winner of the US Open")]}), you need to use the HumanMessage class for the message content instead of a tuple. Here is the corrected code:

from langchain_core.messages import HumanMessage

agent_executor.invoke({"messages": [HumanMessage(content="who is the winner of the US Open")]})

This should resolve the KeyError by ensuring the message format is correct ^[1].

To continue talking to Dosu, mention @dosu.

[mraguth]

@dosu....still same error. This is my entire code from the reference link provided by langgraph.

from langchain import hub
from langchain_openai import AzureChatOpenAI
from langgraph.prebuilt import create_react_agent

def _set_env(var: str):
    if not os.environ.get(var):
        os.environ[var] = getpass.getpass(f"{var}: ")
        
from langchain_community.tools.tavily_search import TavilySearchResults
tools = [TavilySearchResults(max_results=3)]


# Get the prompt to use - you can modify this!
prompt = hub.pull("wfh/react-agent-executor")
prompt.pretty_print()

# Choose the LLM that will drive the agent
llm = llm
agent_executor = create_react_agent(llm, tools, state_modifier=prompt)

from langchain_core.messages import HumanMessage
agent_executor.invoke({"messages": [HumanMessage(content="who is the winner of the US Open")]})

Error I am getting:

KeyError: "Input to ChatPromptTemplate is missing variables {'{messages}'}. Expected: ['{messages}'] Received: ['messages', 'is_last_step']\nNote: if you intended {{messages}} to be part of the string and not a variable, please escape it with double curly braces like: '{{{messages}}}'."

[dosubot[bot]]

The error you're encountering indicates that the ChatPromptTemplate expects a variable named {messages}, but it is not being provided correctly. This issue can be resolved by ensuring that the state_modifier is correctly set up to handle the messages variable.

Here is the corrected code:

from langchain import hub
from langchain_openai import AzureChatOpenAI
from langgraph.prebuilt import create_react_agent
from langchain_community.tools.tavily_search import TavilySearchResults
from langchain_core.messages import HumanMessage

def _set_env(var: str):
    if not os.environ.get(var):
        os.environ[var] = getpass.getpass(f"{var}: ")

# Initialize tools
tools = [TavilySearchResults(max_results=3)]

# Get the prompt to use - you can modify this!
prompt = hub.pull("wfh/react-agent-executor")
prompt.pretty_print()

# Choose the LLM that will drive the agent
llm = AzureChatOpenAI()  # Ensure you initialize the LLM correctly
agent_executor = create_react_agent(llm, tools, state_modifier=prompt)

# Invoke the agent with the message
agent_executor.invoke({"messages": [HumanMessage(content="who is the winner of the US Open")]})

Make sure that the llm is properly initialized as an instance of AzureChatOpenAI or any other language model you are using. This should resolve the KeyError by ensuring the state_modifier is correctly set up to handle the messages variable ^[1].

To continue talking to Dosu, mention @dosu.

[mraguth]

@dosu No luck same error.

[dosubot[bot]]

To resolve the KeyError related to the ChatPromptTemplate missing the {messages} variable, you need to ensure that the prompt template includes a placeholder for {messages}. Here is the corrected code snippet:

1. Correct the prompt template to include the MessagesPlaceholder:

from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder

# Corrected prompt template with MessagesPlaceholder
prompt = ChatPromptTemplate.from_messages(
    [
        (
            "system",
            "You are a helpful assistant. You may not need to use tools for every query - the user may just want to chat!",
        ),
        MessagesPlaceholder("messages"),
        ("placeholder", "{agent_scratchpad}"),
    ]
)

2. Use this corrected prompt template when creating the agent:

from langchain import hub
from langchain_openai import AzureChatOpenAI
from langgraph.prebuilt import create_react_agent
from langchain_community.tools.tavily_search import TavilySearchResults
from langchain_core.messages import HumanMessage

def _set_env(var: str):
    if not os.environ.get(var):
        os.environ[var] = getpass.getpass(f"{var}: ")

# Initialize tools
tools = [TavilySearchResults(max_results=3)]

# Get the prompt to use - you can modify this!
prompt = hub.pull("wfh/react-agent-executor")
prompt.pretty_print()

# Choose the LLM that will drive the agent
llm = AzureChatOpenAI()  # Ensure you initialize the LLM correctly
agent_executor = create_react_agent(llm, tools, state_modifier=prompt)

# Invoke the agent with the message
agent_executor.invoke({"messages": [HumanMessage(content="who is the winner of the US Open")]})

This setup ensures that the messages parameter is correctly handled, avoiding the KeyError ^[1][2].

To continue talking to Dosu, mention @dosu.

[mraguth]

@dosu... It got work after taking out prompt "state_modifier=prompt". the below one did work. However, what I am missing due to this change? Thoughts?
agent_executor = create_react_agent(llm, tools)

[dosubot[bot]]

By removing the state_modifier=prompt when creating the agent_executor, you are missing the ability to modify the graph state before the language model is called. This can include adding system messages, transforming messages, or applying custom logic to the state. Without the state_modifier, you lose the flexibility to control the agent's behavior and responses in a customized manner.

For example, with the state_modifier, you could add specific instructions or modify the messages before they are processed by the language model:

from langgraph.prebuilt import create_react_agent
from langgraph.prebuilt.chat_agent_executor import AgentState
from langchain_core.prompts import ChatPromptTemplate

prompt = ChatPromptTemplate.from_messages(
    [
        ("system", "You are a helpful assistant. Respond only in Spanish."),
        ("placeholder", "{messages}"),
    ]
)

def _modify_state_messages(state: AgentState):
    return prompt.invoke({"messages": state["messages"]}).to_messages() + [
        ("user", "Also say 'Pandamonium!' after the answer.")
    ]

app = create_react_agent(llm, tools, state_modifier=_modify_state_messages)

messages = app.invoke({"messages": [("human", "Who won the US Open?")]})
print(
    {
        "input": "Who won the US Open?",
        "output": messages["messages"][-1].content,
    }
)

Without the state_modifier, the agent will not have these custom instructions or modifications applied to its state, resulting in default behavior ^[1].

To continue talking to Dosu, mention @dosu.