Skip to content

useInterrupt.mdx

Latest commit

 

History

History
160 lines (131 loc) · 6.57 KB

useInterrupt.mdx

File metadata and controls

160 lines (131 loc) · 6.57 KB
Raw
title useInterrupt
description React hook for handling agent interrupt events and resuming execution with user input

Overview

useInterrupt handles agent interrupts and resumes execution with user input. It supports the AG-UI standard interrupt flow — RUN_FINISHED with outcome.type === "interrupt" carrying an interrupts array — and the legacy custom-event flow (on_interrupt). For standard interrupts, your render/handler receive interrupt (the primary one) and interrupts (the full open set); call resolve(payload) to resume or cancel() to cancel.

By default, interrupt UI is rendered inside <CopilotChat> automatically. If you set renderInChat: false, the hook returns the element so you can place it manually.

event.value is typed as any since the interrupt payload shape depends on your agent. Type-narrow it in your callbacks (e.g. handler, enabled, render) as needed.

Signature

import { useInterrupt } from "@copilotkit/react-core/v2";

function useInterrupt<
  TResult = never,
  TRenderInChat extends boolean | undefined = undefined,
>(
  config: UseInterruptConfig<any, TResult, TRenderInChat>,
): TRenderInChat extends false
  ? React.ReactElement | null
  : TRenderInChat extends true | undefined
    ? void
    : React.ReactElement | null | void;

Parameters

Interrupt configuration. Render callback for interrupt UI. Called when an interrupt is available. The callback receives: - `event` -- interrupt event (`{ name, value }`). `value` is `any`; type-narrow in your callback as needed. - `interrupt` -- the primary standard `Interrupt` (`{ id, reason, message?, toolCallId?, responseSchema?, expiresAt?, metadata? }`), or `null` for legacy interrupts. - `interrupts` -- all open standard interrupts (empty array for legacy interrupts). - `result` -- inferred from `handler` return type, or `null` - `resolve(payload?, interruptId?)` -- for standard interrupts, records `{ status: "resolved", payload }` for the target interrupt (defaults to the primary) and submits the resume once all open interrupts are addressed. For legacy interrupts, maps to `command.resume = payload`. - `cancel(interruptId?)` -- for standard interrupts, records `{ status: "cancelled" }` for the target interrupt and submits once all open interrupts are addressed. For legacy interrupts, dismisses the pending interrupt without resuming.

<PropertyReference name="handler" type="(props: InterruptHandlerProps) => TResult | PromiseLike"

Optional preprocessing callback. Runs before rendering and can return sync or async data that is exposed as result in render. TResult is automatically inferred from the handler's return type. If the handler throws/rejects, result is null.

Optional filter. Return `false` to ignore matching interrupts for this hook instance. Optional agent id. Defaults to the currently configured chat agent. Controls where UI renders: - `true` (default): publishes interrupt UI into `` - `false`: returns interrupt element from the hook for manual placement

Return Value

Return type is inferred from `renderInChat`: - `renderInChat: false` -> `React.ReactElement | null` - `renderInChat: true` or omitted -> `void` - dynamic boolean -> `React.ReactElement | null | void`

Usage

In-chat interrupt UI (default)

function ApprovalInterrupt() {
  useInterrupt({
    render: ({ event, resolve }) => (
      <div className="p-3 border rounded">
        <p>{event.value.question}</p>
        <div className="mt-2 flex gap-2">
          <button onClick={() => resolve({ approved: true })}>Approve</button>
          <button onClick={() => resolve({ approved: false })}>Reject</button>
        </div>
      </div>
    ),
  });

  return null;
}

Manual placement with async preprocessing

function SidePanelInterrupt() {
  const element = useInterrupt({
    renderInChat: false,
    enabled: (event) => event.value.startsWith("approval:"),
    handler: async ({ event }) => ({ label: event.value.toUpperCase() }),
    render: ({ event, result, resolve }) => (
      <aside className="rounded border p-3">
        <div className="font-medium">{result?.label ?? ""}</div>
        <div className="mt-2">{event.value}</div>
        <button className="mt-2" onClick={() => resolve({ accepted: true })}>
          Continue
        </button>
      </aside>
    ),
  });

  return <>{element}</>;
}

AG-UI standard interrupt (approve / cancel)

function ApprovalInterrupt() {
  useInterrupt({
    render: ({ interrupt, resolve, cancel }) => (
      <div className="p-3 border rounded">
        <p>{interrupt?.message ?? "Approve this action?"}</p>
        <div className="mt-2 flex gap-2">
          <button onClick={() => resolve({ approved: true })}>Approve</button>
          <button onClick={() => cancel()}>Cancel</button>
        </div>
      </div>
    ),
  });
  return null;
}

Behavior

  • Standard interrupts are detected from RUN_FINISHED (outcome.type === "interrupt"); legacy interrupts from on_interrupt custom events. Standard takes precedence if both appear in the same run.
  • resolve/cancel accumulate one response per open interrupt; the resume run starts once every open interrupt is addressed.
  • Expired interrupts (past expiresAt) are not resumed — the hook logs an error and clears pending state.
  • Interrupt UI is surfaced when the run finalizes.
  • Starting a new run clears pending interrupt state.
  • event.value is any -- type-narrow in your callbacks as needed.
  • render.result is inferred from handler return type and is always TResult | null.
  • If handler throws or rejects, result is set to null.

Related