How to embedded the parameter when migrating from AgentExecutor to langgraph

[HoangNguyen689]

I already read the migration tutorial here: https://python.langchain.com/v0.2/docs/how_to/migrate_agent/#basic-usage
But I still can't find a way to migrate from AgentExecutor to langgraph in my own definition.
The AgentExecutor is defined below

prompt = ChatPromptTemplate.from_messages(
    [
        ("system", SYSTEM_INIT_PROMPT),
        MessagesPlaceholder("chat_history", optional=True),
        ("user", "{input}"),
        MessagesPlaceholder(variable_name="agent_scratchpad"),
    ],
)

agent = create_openai_tools_agent(llm, tools, prompt)

agent_executor = AgentExecutor(
        agent=agent,
        tools=tools,
        verbose=cfg.agent_verbose,
        handle_parsing_errors=True,
)

I have this system prompt

SYSTEM_INIT_PROMPT = """
You are a helpful assistant.

Today is {today}.
"""

And when call the AgentExecutor, I just pass the parameter like this

now = tz.localize(datetime.now()).strftime("%Y/%m/%d (%A)")

res = agent_executor.invoke(
        {
            "input": last_message.message_content,
            "chat_history": history,
            "today": now,
        },
        {
            "callbacks": callbacks,
            "run_id": run_id,
            "metadata": {
                "thread_id": thread_id,
            },
        },
    ).get("output")

How can I pass the today variable to the SYSTEM_INIT_PROMPT?
Besides the today variable, we have others, so if we can't pass them, it seems impossible to migrate.

[HoangNguyen689]

@hinthornw
Can you look a little bit into the above problem?

[vbarda]

@HoangNguyen689 thanks for flagging this! we're looking into making create_react_agent more flexible to allow customizing graph state (which would enable passing other input keys in your example)

in the meantime, you can use below example to achieve the desired behavior. for additional input keys just modify the AgentState

from typing import Annotated, Literal, TypedDict

from langchain_core.messages import HumanMessage, BaseMessage
from langchain_core.prompts import ChatPromptTemplate
from langchain_openai import ChatOpenAI
from langchain_core.tools import tool
from langgraph.checkpoint import MemorySaver
from langgraph.graph import END, StateGraph
from langgraph.graph.message import add_messages
from langgraph.prebuilt import ToolNode


# Define the tools for the agent to use
@tool
def search(query: str):
    """Call to surf the web."""
    # This is a placeholder, but don't tell the LLM that...
    if "sf" in query.lower() or "san francisco" in query.lower():
        return ["It's 60 degrees and foggy."]
    return ["It's 90 degrees and sunny."]


tools = [search]
tool_node = ToolNode(tools)

model = ChatOpenAI(model="gpt-3.5-turbo").bind_tools(tools)

SYSTEM_INIT_PROMPT = """
You are a helpful assistant.

Today is {today}.
"""

class AgentState(TypedDict):
    # add any other input variables you need to keep track of here
    today: str
    messages: Annotated[list[BaseMessage], add_messages]


# Define the function that determines whether to continue or not
def should_continue(state: AgentState) -> Literal["tools", END]:
    messages = state['messages']
    last_message = messages[-1]
    if last_message.tool_calls:
        return "tools"
    return END


def call_model(state: AgentState):
    prompt = ChatPromptTemplate.from_messages(
        [
            ("system", SYSTEM_INIT_PROMPT),
            ("placeholder", "{chat_history}"),
        ]
    )
    model_runnable = prompt | model
    response = model_runnable.invoke(
        {
            "today": state["today"],
            "chat_history": state["messages"]
        }
    )
    return {"messages": [response]}


workflow = StateGraph(AgentState)
workflow.add_node("agent", call_model)
workflow.add_node("tools", tool_node)

workflow.set_entry_point("agent")
workflow.add_conditional_edges(
    "agent",
    should_continue,
)
workflow.add_edge("tools", 'agent')

react_agent = workflow.compile()

react_agent.invoke({
    "messages": [("human", "what's the weather in sf? make sure to mention today's date")], 
    "today": "July 12, 2024"
})

[HoangNguyen689]

@vbarda
First, thank you for your nice response.

For the thread_id, I have it both in the configuration and state for the application's purpose.

Next, I want to clarify the difference between the 2 cases.

With AgenExecutor: input message + chat_history(dynamodb) -> output
With Langgraph: input message + checkpointer(dynamodb?) -> output

Is that right and equal?

Because as I tested as below, it didn't work. Although I put the history inside the messages, the agent can't recognize the preceding conversation. (As of now, I still haven't implemented the checkpointer with dynamodb yet)

from abc import ABC, abstractmethod

from boto3.session import Session
from langchain_community.chat_message_histories.dynamodb import (
    DynamoDBChatMessageHistory,
)
from langchain_core.chat_history import BaseChatMessageHistory


class ChatHistoryStore(ABC):
    @abstractmethod
    def get_thread_store(self, thread_id: str) -> BaseChatMessageHistory:
        pass


class DynamoDBChatHistoryStore(ChatHistoryStore):
    def __init__(self, boto3_session: Session):
        self.boto3_session = boto3_session

    def get_thread_store(self, thread_id: str) -> BaseChatMessageHistory:
        return DynamoDBChatMessageHistory(
            table_name="ChatHistoryTable",
            session_id=thread_id,
            key={"thread_id": thread_id},
            boto3_session=self.boto3_session,
        )


chs = DynamoDBChatHistoryStore(aws_session)

store = chs.get_thread_store(thread_id)

res = react_agent.invoke(
     {
        "messages": store.messages + [("human", st_input)],
        "today_date": now_jst,
        "thread_id": thread_id,
    },
    {
        "configurable": {
            "thread_id": thread_id,
        },
    },
)

[vbarda]

@HoangNguyen689 you can't use langchain's BaseChatMessageHistory as a checkpointer, you would need to implement a custom LangGraph checkpointer or use one of prebuilt ones. please see more information here: https://langchain-ai.github.io/langgraph/how-tos/persistence/

[vbarda]

btw, @HoangNguyen689 we just released version 0.1.9 that should make the above very easy to achieve with create_react_agent. see an example here https://langchain-ai.github.io/langgraph/reference/prebuilt/#create_react_agent (search for "Add complex prompt with custom graph state"). hope this is useful!

[HoangNguyen689]

@vbarda
Thank you! I will read about it.
However the create_react_agent is used to create the entire graph? not only the node?

[vbarda]

that's correct!