What this does

Human-in-the-loop pauses execution before or after specific nodes, allowing a human to approve, reject, or modify the agent's action before continuing.

Steps

Add a checkpointer. The checkpointer saves state snapshots for later resumption.

from langgraph.checkpoint import MemorySaver

memory = MemorySaver()

Compile the graph with the checkpointer and an interrupt point. Use interrupt_before to pause before a node.

from langgraph.graph import StateGraph, END
from typing import TypedDict

class State(TypedDict):
    input: str
    approved: bool

def process(state: State) -> dict:
    return {"approved": False}

def execute(state: State) -> dict:
    return {"approved": True}

workflow = StateGraph(State)
workflow.add_node("process", process)
workflow.add_node("execute", execute)
workflow.set_entry_point("process")
workflow.add_edge("process", "execute")
workflow.add_edge("execute", END)

app = workflow.compile(
    checkpointer=memory,
    interrupt_before=["execute"]
)

Run with a thread ID. The thread ID ties state across interruptions.

config = {"configurable": {"thread_id": "session-1"}}

# First run — pauses before "execute"
result = app.invoke({"input": "transfer $100", "approved": False}, config)
print(result)  # State is partial, execution is paused

Resume with human approval. Update state and continue.

# Human reviews and approves
state = app.get_state(config)
state.values["approved"] = True
app.update_state(config, {"approved": True})

# Resume execution
result = app.invoke(None, config)
print(result)  # Now "execute" node ran

Reject by updating state and ending. Set a field that causes the execute node to skip.

state.values["approved"] = False
app.update_state(config, {"approved": False})
# The execute node can check approved before proceeding

Verification

python -c "
from langgraph.checkpoint import MemorySaver
m = MemorySaver()
print(type(m).__name__)
# Expected: MemorySaver
"

Common failures

Missing thread_id. The checkpointer requires a thread_id in the config; otherwise, state is not persisted and interrupts fail silently.
Resuming with wrong state schema. update_state must match the state schema. Extra keys cause validation errors.
Interrupt prevents graph from reaching END. If execution is interrupted and never resumed, the graph stays in a paused state. Check app.get_state(config).next to see pending nodes.
Version mismatch - The installed package or runtime differs from the command shown; check the version first and rerun the smallest verification command.
Local environment drift - Another service, virtual environment, model, or path is being used; print the active binary path and configuration before changing the guide steps.

Related guides

How to Set Up Basic LangGraph Agent from Scratch
How to Implement Memory in LangGraph State