issue with Resuming Interrupted Tool in LangGraph Multi-Agent Architecture

[Yash-Makwana1]

Hi LangGraph team,

I'm currently using LangGraph's multi-agent architecture. I'm facing an issue where, if I use interrupt() inside a tool of an agent (say, in the second tool of a node that calls multiple tools sequentially), then upon resuming, the entire node is re-executed. This results in the first tool being executed again, even though the interrupt was triggered in the second tool.

Is there a recommended way to resume from the exact point of interruption, without rerunning the previous tools in the same node?

Appreciate any guidance or workaround for this!

Thanks!

[RockeyDon]

A critical aspect of using interrupt is understanding how resuming works. When you resume execution after an interrupt, graph execution starts from the beginning of the graph node where the last interrupt was triggered.
https://langchain-ai.github.io/langgraph/concepts/human_in_the_loop/?h=resume#using-with-invoke-and-ainvoke

I prefer to add a field in state to store 'last end', when resume, skip the code executed

[Yash-Makwana1]

The link provided gives a 404 , and will it solve the issue ?

[vbarda]

@Yash-Makwana1 if you are using prebuilt create_react_agent, you can specify version="v2". this will execute each tool call in a separate "instance" of ToolNode and will store tool results for non-interrupted tools, so that when you resume those tools are not re-executed

see implementation here https://github.com/langchain-ai/langgraph/blob/main/libs/prebuilt/langgraph/prebuilt/chat_agent_executor.py#L880
and a new guide for adding interrupts to tools here https://langchain-ai.github.io/langgraph/agents/human-in-the-loop/

hope this helps!

[Yash-Makwana1]

@vbarda
So i tried this

tool with interrupt:

def subtract(a: float, b: float) :
    """ Subtract two numbers."""
    interrupt_data = interrupt(
        {"test": 'test'})
    return {'success': True, 'data': interrupt_data}

Agent creation:

checkpointer = InMemorySaver()
calculator_agent = create_react_agent(
    model,
    tools=[add,  subtract],
    name="calculator_agent",
    prompt="""some prompt""",
    version='v2',
    checkpointer=checkpointer
)

Supervisor creation:

from langgraph.checkpoint.memory import InMemorySaver

check_pointer = InMemorySaver()
model = ChatOpenAI(model="gpt-4o", temperature=0.1)

# Create supervisor workflow
workflow = create_supervisor(
    [calculator_agent ],
    model=model,
    prompt=("with some prompt"),
    output_mode="full_history",
)
graph = workflow.compile(checkpointer=check_pointer)

But still the same issue persists

langgraph version 0.3.18

[Yash-Makwana1]

@vbarda
i tried with the latest version of langgraph as well still the same issue persisted

Any solution for this?

[ykkaixin]

maybe you do not need to specific the checkpoint for subgraph (calculator_agent), cause the parent's checkpointer will handle all state.

[Yash-Makwana1]

@hinthornw @vbarda can anyone provide a solution for this ?

[chenweisomebody126]

@Yash-Makwana1 Have you tried v0.4 as in https://changelog.langchain.com/announcements/langgraph-v0-4-working-with-interrupts?