Memory saver doesn't work in react agent

[cris-m]

I am new and I still learning about langgraph. I saw the example about langgraph react agent and I am playing with it. I wanted to add memory to it like thread-level persistence I add Memory Saver but it doesn't work. I am missing something? here is the graph:

from datetime import datetime, timezone
from typing import Dict, List, Literal, cast

from langgraph.graph import StateGraph
from langgraph.prebuilt import ToolNode
from langchain_core.messages import AIMessage
from langchain_core.runnables import RunnableConfig
from langgraph.checkpoint.memory import MemorySaver
from langchain_core.prompts import ChatPromptTemplate

from agents.chatbot.configuration import Configuration
from agents.chatbot.state import InputState, State
from agents.chatbot.tools import TOOLS
from agents.chatbot.utils import load_chat_model

async def call_model(
    state: State, config: RunnableConfig
) -> Dict[str, List[AIMessage]]:
    configuration = Configuration.from_runnable_config(config)

    configurable = config.get('configurable', {})
    user_info = configurable.get('user_info', {})
    current_time = datetime.now(tz=timezone.utc)
    prompt = ChatPromptTemplate.from_messages(
        [
            ("system", configuration.system_prompt),
            ("placeholder", "{messages}")

        ]
    ).partial(
        date=current_time.isoformat(),
        time=current_time.strftime("%H:%M:%S")
    )
    model = load_chat_model(configuration.model).bind_tools(TOOLS)
    message_value = await prompt.ainvoke(
        {
            "messages": state.messages,
            "user_info": user_info,
        },
        config,
   
    response = cast(AIMessage, await model.ainvoke(message_value, config))
    if state.is_last_step and response.tool_calls:
        return {
            "messages": [
                AIMessage(
                    id=response.id,
                    content="Sorry, I could not find an answer to your question in the specified number of steps.",
                )
            ]
        }

    return {"messages": [response]}

workflow = StateGraph(State, input=InputState, config_schema=Configuration)

workflow.add_node(call_model)
workflow.add_node("tools", ToolNode(TOOLS))
workflow.add_edge("__start__", "call_model")


def route_model_output(state: State) -> Literal["__end__", "tools"]:
    last_message = state.messages[-1]
    if not isinstance(last_message, AIMessage):
        raise ValueError(
            f"Expected AIMessage in output edges, but got {type(last_message).__name__}"
        )
    if not last_message.tool_calls:
        return "__end__"

    return "tools"


workflow.add_conditional_edges(
    "call_model",
    route_model_output,
)

workflow.add_edge("tools", "call_model")

graph = workflow.compile(
    checkpointer=MemorySaver(),
    interrupt_before=[],
    interrupt_after=[], 
)
graph.name = "Chatbot Agent"

it doesn't remember the last previous conversation, I can even confirm by checking in langsmith. maybe it is due to call_model method. here is the state:

from __future__ import annotations

from dataclasses import dataclass, field
from typing import Sequence

from langchain_core.messages import AnyMessage
from langgraph.graph import add_messages
from langgraph.managed import IsLastStep
from typing_extensions import Annotated


@dataclass
class InputState:
    messages: Annotated[Sequence[AnyMessage], add_messages] = field(
        default_factory=list
    )


@dataclass
class State(InputState):
    is_last_step: IsLastStep = field(default=False)

[vbarda]

@cris-m how are you invoking it? i tested your code and it works as intended:

thread_config = {"configurable": {"thread_id": "1"}}

await graph.ainvoke({"messages": [("user", "hi, my name is John")]}, thread_config)
await graph.ainvoke({"messages": [("user", "what's my name?")]}, thread_config)
# the model correctly remembers the name for the same thread

perhaps you're looking for cross-thread persistence, where the model remembers information from one thread in another? in that case, check out this how-to guide https://langchain-ai.github.io/langgraph/how-tos/cross-thread-persistence/

[cris-m]

I don't undestand what happened. it seems I have issue with langsmith. I have just change the key and everything back to normal.
I don't understand how langsmith can affect the agent.

[vbarda]

hm, langsmith shouldn't affect it. please let me know if you run into issues again

[cris-m]

I have identify the issue. The model is called in fast api and every time it was generating a new thread_id when the route is called. That why it was behaving perfect when use LangGraph Studio but it was not responding when calling the model from the fastapi route.

[vbarda]

makes sense!