Skip to main content

Human in the Loop

interrupt() pauses a graph mid-run, persists its full state via the checkpointer, and waits — no polling, no separate queue. A Command(resume=...) call later continues execution from that exact point with the human's input folded in.

from langgraph.checkpoint.memory import InMemorySaver
from langgraph.graph import END, START, StateGraph
from langgraph.types import Command, interrupt

def human_review(state: State):
    decision = interrupt({"proposed_action": state["action"]})
    return {"approved": decision["approved"]}

graph = (
    StateGraph(State)
    .add_node("human_review", human_review)
    .add_edge(START, "human_review")
    .add_edge("human_review", END)
    .compile(checkpointer=InMemorySaver())
)

config = {"configurable": {"thread_id": "run-1"}}
graph.invoke({"action": "delete_record"}, config)  # pauses at human_review

# ... a human reviews, then:
graph.invoke(Command(resume={"approved": True}), config)  # resumes
tip

Gate on the tool's blast radius, not on the model's confidence. A confident model can still be confidently wrong — what matters is whether the action it's about to take (delete, send, spend) can be undone.

Requires a checkpointer — see Checkpointing — since the whole point is pausing execution and resuming it later, possibly in a different process.

See also​