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
useHumanInTheLoop: nothing to wire on the agent
The AG-UI Mastra adapter forwards each tool you register with
useHumanInTheLoop to the agent as a client tool. When the model calls
it, the run hands control to the browser, and the user's answer comes
back as the tool result. Do not declare the tool on the Mastra agent.
useInterrupt: return suspend() from a tool
A Mastra tool pauses with suspend(), not interrupt(). Declare a
suspendSchema for the payload your UI reads and a resumeSchema for
the answer it sends back. The @ag-ui/mastra bridge sends the pause to
the frontend as an AG-UI interrupt. When the user answers, Mastra runs
execute again with the answer in resumeData.
// Native interrupt backend for the `gen-ui-interrupt` + `interrupt-headless`
// cells (OSS-383). Unlike the old Strategy-B workaround (a frontend-only
// `schedule_meeting` satisfied by `useHumanInTheLoop`), this is a REAL Mastra
// suspend tool: the model calls `schedule_meeting`, the tool `suspend()`s with
// a time-picker payload, and the @ag-ui/mastra bridge maps that to an AG-UI
// interrupt (legacy `on_interrupt` CUSTOM event + the standard `RUN_FINISHED`
// interrupt-outcome, on by default in the v1 bridge). The frontend `useInterrupt`
// hook renders the picker and `resolve(...)`s, which resumes the run — the tool's
// `execute` is re-invoked with `resumeData` carrying the user's selection.
//
// Load-bearing (see the Mastra capability-map memory):
// - `return suspend(...)` DIRECTLY. `await suspend(); return x` completes the
// tool, so under fast streaming the agentic loop continues past the pause.
// - Resume needs instance `storage` (configured on `new Mastra({ storage })`)
// and the suspend-chunk runId — both handled by the bridge + src/mastra/index.ts.
import { createTool } from "@mastra/core/tools";
import { z } from "zod";
export interface InterruptTimeSlot {
label: string;
iso: string;
}
// Mirror of `src/app/demos/_shared/interrupt-fallback-slots.ts` so the backend
// supplies future-relative candidate slots inside the suspend payload (the
// frontend only falls back to its own generator if these are absent).
function atLocal(date: Date, hour: number, minute = 0): Date {
return new Date(
date.getFullYear(),
date.getMonth(),
date.getDate(),
hour,
minute,
0,
0,
);
}
function nextMonday(from: Date): Date {
const day = from.getDay();
let offset = (1 - day + 7) % 7;
if (offset <= 1) offset += 7;
const next = new Date(from);
next.setDate(from.getDate() + offset);
return next;
}
export function generateCandidateSlots(
now: Date = new Date(),
): InterruptTimeSlot[] {
const tomorrow = new Date(now);
tomorrow.setDate(now.getDate() + 1);
const monday = nextMonday(now);
const candidates: Array<[string, Date]> = [
["Tomorrow 10:00 AM", atLocal(tomorrow, 10)],
["Tomorrow 2:00 PM", atLocal(tomorrow, 14)],
["Monday 9:00 AM", atLocal(monday, 9)],
["Monday 3:30 PM", atLocal(monday, 15, 30)],
];
return candidates.map(([label, date]) => ({
label,
iso: date.toISOString(),
}));
}
export const scheduleMeetingInterruptTool = createTool({
id: "schedule_meeting",
description:
"Ask the user to pick a meeting time. Surfaces an interactive time-picker " +
"to the user and returns their selection. Call this whenever the user asks " +
"to book or schedule a meeting.",
inputSchema: z.object({
topic: z
.string()
.describe("What the meeting is about (e.g. 'Intro with sales')."),
attendee: z
.string()
.optional()
.describe("Who the meeting is with (e.g. 'Alice'), if known."),
}),
suspendSchema: z.object({
topic: z.string(),
attendee: z.string().optional(),
slots: z.array(z.object({ label: z.string(), iso: z.string() })),
}),
resumeSchema: z.object({
chosen_time: z.string().optional(),
chosen_label: z.string().optional(),
cancelled: z.boolean().optional(),
}),
// Mastra createTool execute signature is `(inputData, executionContext)`:
// the validated INPUT is the first arg; the ToolExecutionContext is the
// second. `suspend` / `resumeData` live under `executionContext.agent`
// (the `AgentToolExecutionContext` sub-object), NOT at the top level.
// Destructuring them off `executionContext` directly yields `undefined`, so
// `return suspend(...)` throws `suspend is not a function`; the model gets a
// tool-error, re-calls, and the loop spins to the step cap with no
// `tool-call-suspended` chunk. Mirrors the proven @ag-ui/mastra dojo tool.
execute: async (inputData, executionContext) => {
const { suspend, resumeData } = executionContext?.agent ?? {};
// Second pass: the user resolved the interrupt — the run resumes here with
// their selection. Return a short confirmation the agent narrates.
if (resumeData) {
if (resumeData.cancelled) {
return "The user cancelled — no meeting was scheduled.";
}
const when =
resumeData.chosen_label ?? resumeData.chosen_time ?? "the chosen time";
return `Scheduled "${inputData.topic}" for ${when}.`;
}
// First pass: suspend with the picker payload. Returned directly so the
// agentic loop pauses here instead of continuing.
return suspend?.({
topic: inputData.topic,
attendee: inputData.attendee,
slots: generateCandidateSlots(),
});
},
});Two details fail without an error:
suspendandresumeDataare onexecutionContext.agent, the second argument ofexecute. Read from the top level of the context, they areundefined, and the model retries the tool until the run reaches its step limit.return suspend(...)directly. Withawait suspend()followed by anotherreturn, the tool completes and the agent keeps going past the pause.
Give the Mastra instance storage
Resume reloads the paused run from storage, so the Mastra instance that
registers the agent needs storage:
import { Mastra } from "@mastra/core/mastra";
import { LibSQLStore } from "@mastra/libsql";
export const mastra = new Mastra({
agents: { interruptAgent },
storage: new LibSQLStore({
id: "mastra-storage",
url: ":memory:",
}),
});Read the payload under suspendPayload
The bridge wraps your suspend() value, so useInterrupt does not
receive it as is:
- In the custom-event flow,
event.valueis a JSON string. Parse it and readsuspendPayload. - In the standard AG-UI flow, the payload is in
interrupt.metadata.mastra.suspendPayload.
useInterrupt({
render: ({ interrupt, event, resolve }) => {
const payload = interrupt
? interrupt.metadata?.mastra?.suspendPayload
: JSON.parse(event.value).suspendPayload;
return (
<div>
<p>Pick a time for {payload?.topic}</p>
<button onClick={() => resolve({ chosen_time: "Tomorrow 10:00 AM" })}>
Tomorrow 10:00 AM
</button>
<button onClick={() => resolve({ cancelled: true })}>Cancel</button>
</div>
);
},
});Whatever you pass to resolve(...) arrives in the tool as resumeData,
so shape it to match the tool's resumeSchema.
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.
| Pattern | Who decides to pause? | Backend surface |
|---|---|---|
useHumanInTheLoop | The LLM, by calling a registered client-side tool | A frontend-only tool description (Zod schema + render) |
useInterrupt | Your agent code, at a step that must stop for a person | An 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.
<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:
<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:
- The agent interrupts with a serializable payload.
- The run stops.
- The client renders your UI with the payload.
- 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?