HITL Overview

Allow your agent and users to collaborate on complex tasks.

What is this?

Human-in-the-loop (HITL) lets an agent pause mid-run to collect input, confirmation, or a choice from the user, then resume with that answer folded back into its reasoning. It's what turns an autonomous workflow into a collaborative one: the agent keeps its context, the user keeps the steering wheel.

When should I use this?

Use HITL when you need:

  • Quality control: a human gate at high-stakes decision points.
  • Missing information: the agent asks the user for a value it cannot find.
  • Expert input: the user supplies domain knowledge the model lacks.

Two patterns for HITL in CopilotKit

Install the OpenAI Agents AG-UI adapter

pip install ag-ui-openai-agents openai-agents

Ask for the human with a frontend tool

The simplest route is an ordinary frontend tool. Register it in the browser with useHumanInTheLoop, or with useFrontendTool and an async handler that resolves once the user has answered. The adapter hands the tool to the model on every run. When the model calls it, the run stops and the browser asks the user. The answer goes back with the next request as the call's result, and the model continues from there. Nothing waits on the server in between.

Name the tool in the agent's instructions

That leaves the agent with one job: knowing the tool exists and when to call it. A model that is never told about book_call will answer in prose and the picker will never mount.

hitl.py
from agents import Agent

from demo_agents._common import MODEL, chat_settings

HITL_IN_CHAT_PROMPT = (
    "You help users book an onboarding call with the sales team. "
    "When they ask to book a call, call the frontend-provided "
    "`book_call` tool with a short topic and the user's name. "
    "Keep any chat reply to one short sentence."
)


hitl_in_chat_agent = Agent(
    name="booking_assistant",
    instructions=HITL_IN_CHAT_PROMPT,
    model=MODEL,
    model_settings=chat_settings(),
)

Or gate a server tool with needs_approval

When the backend decides it needs the user, mark the tool needs_approval=True. The SDK pauses before the tool runs, and the adapter ends the AG-UI run with an interrupt for that call, which useInterrupt renders. The interrupt's reason is "tool_approval", its metadata carries tool_name and the parsed tool_input, and it asks for {"approved": boolean}.

interrupt_agent.py
@function_tool(needs_approval=True)
def schedule_meeting(ctx: ToolContext[Any], topic: str, attendee: str = "") -> str:
    """Ask the user to pick a meeting time, then confirm what was scheduled.

    Args:
        topic: Short description of the meeting purpose.
        attendee: Who the meeting is with, if known.
    """
    # Runs once the user approved, with the time they picked beside it.
    answer = _answers.get({}).get(ctx.tool_call_id, {})
    label = answer.get("chosen_label") or answer.get("chosen_time")
    if not label:
        return f"User did not pick a time. Meeting NOT scheduled: {topic}"
    return f"Meeting scheduled for {label}: {topic}"

The next request resumes the interrupt. {"approved": true} runs the tool before the model is called again. Any other answer gives the model the SDK's rejection text, and a cancelled interrupt answers the call as cancelled. If the user sends a new message instead of deciding, the approval stays open: the run ends with the same interrupt again, and the new message reaches the model once the approval is decided.

Hand the tool more than a yes or no

An Agents SDK approval is only a yes or a no, so an approved tool runs with its own arguments and nothing else. The showcase's picker sends the chosen time beside approved. A small OpenAIAgentsAgent subclass keeps each request's answers, by tool call id, where schedule_meeting can read them:

interrupt_agent.py
class SchedulerAgent(OpenAIAgentsAgent):
    """Lets ``schedule_meeting`` read the time the user picked.

    The adapter runs an approved call with its arguments alone: an Agents SDK
    approval is a yes or a no, with nothing to hand the tool. The picker
    sends the time beside the approval, so this keeps each answer where the
    tool can read it.
    """

    async def run_streamed(self, input: RunAgentInput) -> AsyncIterator[BaseEvent]:
        _answers.set(
            {
                entry.interrupt_id: entry.payload
                for entry in input.resume or []
                if isinstance(entry.payload, dict)
            }
        )
        async with aclosing(super().run_streamed(input)) as events:
            async for event in events:
                yield event

Nothing is kept on the server

The paused call lives in the conversation the page holds, and the server trusts the history it is sent. Replaying the same resume request runs an approved tool again, so key side effects on ToolContext.tool_call_id where a retry must not repeat them, and authorize the decision yourself if that matters.

CopilotKit ships two complementary ways to pause an agent turn and ask the human something. They look similar from the outside (the chat pauses, a custom component appears, the user answers, the run resumes), but they're wired differently on the backend, and each has its own niche.

PatternWho decides to pause?Backend surface
useHumanInTheLoopThe LLM, by calling a registered client-side toolA frontend-only tool description (Zod schema + render)
useInterruptYour agent code, at a step that must stop for a personAn interrupt the agent emits, for example interrupt() in a LangGraph node

Pick useHumanInTheLoop when the pause is an agent-initiated decision (the model chose to ask the user) and you want the picker UI inlined into the normal tool-call flow.

Pick useInterrupt when the pause is a checkpoint in your agent code: the code path always requires a human answer, and the server-side contract is an interrupt. On the Built-in Agent, interrupts are temporarily unavailable, so use useHumanInTheLoop there.

Pattern 1: useHumanInTheLoop (tool-based)

The app registers a HITL tool on the client with useHumanInTheLoop. When the LLM calls that tool, CopilotKit renders your Vue component and gives it a respond callback during the executing phase. The agent sees the answer as the tool result and continues from there.

src/components/MeetingApproval.vue
<script setup lang="ts">
import { z } from "zod";
import { useHumanInTheLoop } from "@copilotkit/vue/v2";
import TimePickerCard from "./TimePickerCard.vue";

useHumanInTheLoop({
  name: "schedule_meeting",
  description: "Ask the user to select a meeting time.",
  parameters: z.object({
    topic: z.string(),
    attendee: z.string(),
  }),
  render: TimePickerCard,
});
</script>

<template><p>Meeting approval tool registered.</p></template>

TimePickerCard.vue receives the tool arguments and a respond callback. respond is defined only while the tool call is executing:

src/components/TimePickerCard.vue
<script setup lang="ts">
const props = defineProps<{
  args: { topic?: string; attendee?: string };
  respond?: (result: unknown) => Promise<void>;
}>();
</script>

<template>
  <section>
    <p>Pick a time for {{ props.args.topic ?? "the meeting" }}</p>
    <button
      :disabled="!props.respond"
      @click="props.respond?.({ chosen_time: '09:00' })"
    >
      Tomorrow at 09:00
    </button>
  </section>
</template>

The picker UI can use static candidate slots, live availability, a calendar API, or any other application data. Mount the registration component beneath CopilotKitProvider so the composable can reach the runtime.

Pattern 2: useInterrupt (graph-paused)

With a runtime interrupt, your agent code enforces the pause:

  1. The agent interrupts with a serializable payload.
  2. The run stops.
  3. The client renders your UI with the payload.
  4. The user answers, and the client resumes the run with the answer.

CopilotKit's useInterrupt composable and CopilotChat's #interrupt slot form the Vue render contract.

See the useInterrupt deep dive for the full walkthrough, including the backend contract and Vue slot wiring.

Going headless

Both patterns above can drive custom UI. For runtime interrupts, set renderInChat: false and use the reactive state returned by useInterrupt; for a frontend-tool pause, let the Promise-returning HITL tool drive application state. See the headless interrupts guide for the interrupt path.

How is this guide?