LangGraph Studio weird behavior with Human-in-the-loop (HITL)

[theprobugmaker]

I got this code somewhere and in the video it's working for the person but not for me now, using the latest versions it's not. I'm getting this after filling the input:

That is the return:

Here is the code:

from dotenv import load_dotenv
from langchain_core.messages import HumanMessage, SystemMessage
from langgraph.checkpoint.memory import MemorySaver
from langgraph.graph import END, StateGraph, MessagesState
from langgraph.prebuilt import ToolNode
from typing import Literal
from langchain_core.tools import tool
from langchain_openai import ChatOpenAI

load_dotenv()

@tool
def return_foo():
    """
    Return foo.
    """
    return "foo"

tools = [return_foo]
tool_node = ToolNode(tools)

model = ChatOpenAI(temperature=0, model_name="gpt-4o-mini")
model = model.bind_tools(tools=tools)

def should_continue(state: MessagesState) -> Literal["tools", "human_feedback"]:
    last_message = state['messages'][-1]

    return "tools" if last_message.tool_calls else "human_feedback"

def should_continue_with_feedback(state: MessagesState) -> Literal["agent", "__end__"]:
    last_message = state['messages'][-1]

    if isinstance(last_message, HumanMessage):
        return "agent"

    return "__end__"

def agent_node(state: MessagesState):
    system_message = SystemMessage(content=f"You are a helpful assistant.")

    messages = [system_message] + state['messages']
    response = model.invoke(messages)

    return {"messages": [response]}

def human_feedback_node(state: MessagesState):
    pass
    
workflow = StateGraph(MessagesState)

workflow.add_node("agent", agent_node)
workflow.add_node("tools", tool_node)
workflow.add_node("human_feedback", human_feedback_node)

workflow.set_entry_point("agent")

workflow.add_conditional_edges("agent", should_continue)
workflow.add_conditional_edges("human_feedback", should_continue_with_feedback)

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

checkpointer = MemorySaver()

graph = workflow.compile(checkpointer=checkpointer, interrupt_before=['human_feedback'])

Just to give a little bit of context:

The idea is that the user can have a multi-turn conversation with the AI.

---

I saw this about the Studio upgrade that it's now using interrupt and resume but not even that example is working:

langchain-ai/langgraph-studio#20 (comment)

from typing import Literal

from langchain_openai import ChatOpenAI
from langgraph.graph import MessagesState, StateGraph, END
from langgraph.types import Command, interrupt


model = ChatOpenAI(temperature=0, model_name="gpt-4o")


def call_model(state: MessagesState) -> Command[Literal["human", END]]:
    messages = state["messages"]
    response = model.invoke(messages)

    return Command(
        goto="human",
        update={"messages": [response]},
    )


def human_feedback(state: MessagesState) -> Command[Literal["agent"]]:
    """A node for collecting user input."""
    print("Waiting for user input...")
    user_input = interrupt(value="Ready for user input.")

    print("user input:", user_input)

    return Command(
        goto="agent",
        update={
            "messages": [
                {
                    "role": "human",
                    "content": user_input,
                }
            ]
        },
    )


workflow = StateGraph(MessagesState)
workflow.set_entry_point("agent")
workflow.add_node("agent", call_model)
workflow.add_node("human", human_feedback)

graph = workflow.compile()

After the user input it's not going back to the agent, it just stops.