anchorBeta overview

AI Summaries Beta is available in @webex/contact-center@3.12.0-next.159 and supports two workflows:

The SDK provides summary generation, response reporting, capability updates, and an event for received mid-call summaries. Your application provides the review UI, editing, notifications, feedback, and usage tracking. Consult, transfer, and task wrap-up use their existing task APIs.

anchorBefore you begin

• Configure AI Assistant. An administrator must enable AI Assistant access and the required summary features for the applicable queues. See Enable AI Assistant features in Control Hub for Webex Contact Center for licensing, user-profile, and recording prerequisites. Its Agent Desktop layout and notification instructions apply to Agent Desktop; SDK applications provide their own UI.
• Check language availability. See the Supported languages for Summaries table in Multi-language Support for AI Assistant Features. Availability depends on the feature, locale, and SKU.
• Initialize an agent session. Complete authentication, SDK initialization and readiness, registration, and station login. See Quickstart and Authentication. Use the pinned Beta package below in place of the Quickstart's unversioned installation command.
• Obtain the agent's task. Follow the task lifecycle to obtain an ITask for new and restored interactions. The examples in this guide begin with an initialized session and an existing task.

Install the Beta package with npm or Yarn:

npm install @webex/contact-center@3.12.0-next.159
# Alternative
yarn add @webex/contact-center@3.12.0-next.159

anchorCheck summary availability

Generation requires both the organization-level configuration flag and the corresponding task capability:

The SDK rejects a request if either gate is missing or false. Configure organization flags through Contact Center; use the task capability snapshot and updates to control generation actions in your UI. The summary APIs add no station or destination restrictions beyond those of the underlying call flow and tenant configuration.

Both task capabilities initially default to false. Read task.aiSummaryCapabilities when binding a task and subscribe to TASK_EVENTS.TASK_FEATURE_ENABLEMENT for subsequent changes. Reading the snapshot also handles an update that arrived before listener registration.

Bind the received-summary listener at the same time, independently of generation controls. Attach it before answering an offered call: a late listener or restored session is not guaranteed to receive previously delivered summaries.

import {TASK_EVENTS} from '@webex/contact-center';
import type {
  AISummary,
  AISummaryFeatureEnablement,
  ITask,
} from '@webex/contact-center';

export function bindSummaries(
  task: ITask,
  updateAvailability: (midCall: boolean, postCall: boolean) => void,
  showReceivedSummary: (summary: AISummary) => void
): () => void {
  const refresh = (_event?: AISummaryFeatureEnablement): void => {
    const {midCallEnabled, postCallEnabled} = task.aiSummaryCapabilities;
    updateAvailability(midCallEnabled, postCallEnabled);
  };

  task.on(TASK_EVENTS.TASK_FEATURE_ENABLEMENT, refresh);
  task.on(TASK_EVENTS.TASK_MID_CALL_SUMMARY_RECEIVED, showReceivedSummary);
  refresh();

  return () => {
    task.off(TASK_EVENTS.TASK_FEATURE_ENABLEMENT, refresh);
    task.off(TASK_EVENTS.TASK_MID_CALL_SUMMARY_RECEIVED, showReceivedSummary);
  };
}

Call the cleanup function when replacing the task or unmounting its view. Associate displayed summaries and pending results with their originating task so an old result cannot update another interaction.

anchorIntegrate mid-call summaries

Generate and share a summary

Use the initiating agent's active task when preparing a consult or transfer:

1. Check task.aiSummaryCapabilities.midCallEnabled and the normal task-state and destination requirements for the call operation.
2. Call task.requestMidCallSummary(action) with an AISummaryAction of 'CONSULT' or 'TRANSFER'. The promise resolves to an AISummary for your application to display.
3. Let the agent review and edit the content. Build an AISummaryResponse with the reviewed summary, feedback, response state, and actual view, edit, and copy counts. See Summary payloads and responses.
4. Dispatch task.sendMidCallSummaryResponse(response, action) with the same action used for generation, and handle its result.
5. Start the consult or transfer through the existing task API.

Summary reporting and call control are separate operations. A reporting acknowledgement confirms the report was accepted; it does not start the consult or transfer or confirm delivery to the receiving agent. The call operation can proceed independently of that acknowledgement, provided your application handles reporting failures.

If the agent cancels the consult or transfer selection, report the appropriate response state. Sending 'MID_CALL_CANCELLED' does not cancel generation or an active call operation. After a successful consult or transfer submission, closing its dialog should not send a second cancellation response.

Display a received summary

The receiving task emits TASK_EVENTS.TASK_MID_CALL_SUMMARY_RECEIVED with an AISummary. The receiving agent does not need to request generation to view it.

Display the event payload for that task, provide a notification and a way to reopen it, and keep received content distinct from summaries the agent subsequently generates. Your application controls summary display and any retained history. Previously delivered summaries are not guaranteed to replay.

Report receiving-agent feedback

Receiving-agent card actions use webex.cc.apiAIAssistant.sendEvent(). Use this event path for feedback on a received card; task.sendMidCallSummaryResponse() reports the initiating agent's consult or transfer response.

Use likeButton, dislikeButton, and copyButton as the action IDs for positive feedback, negative feedback, and copying.

Require a card ID and associate each submission with the displayed summary and its receiving interaction, including retained history. Prevent overlapping submissions for the same card. Confirm feedback only after a successful send and preserve the previous selection on failure; a later summary must not receive an earlier card's acknowledgement.

Beta typing limitation: SDK 3.12.0-next.159 exposes sendEvent(), but its event-name type union omits MID_CALL_SUMMARY_USER_ACTION. A direct TypeScript integration needs a narrow type assertion on that argument.

anchorIntegrate post-call summaries

Request a post-call summary on the task being wrapped up:

1. Check task.aiSummaryCapabilities.postCallEnabled, then await task.requestPostCallSummary().
2. Display the returned AISummary. Let the agent review and edit it, record feedback and usage, and select a configured wrap-up reason.
3. Build an AISummaryResponse. If reporting the selected reason, set the optional response.wrapUpCode to its configured code identifier (auxCodeId).
4. Dispatch task.sendPostCallSummaryResponse(response) before task cleanup and handle its result.
5. Complete wrap-up separately with task.wrapup({wrapUpReason, auxCodeId}), using the selected reason's label and configured identifier.

Wrap-up can start without waiting for summary-reporting acknowledgement. Handle both results independently, including when your application performs automatic wrap-up. Post-call summaries are returned by the request promise; there is no post-call-summary-received task event.

Use suggested wrap-up reasons

summary.suggestedWrapUpCodes optionally contains suggested reasons, and summary.suggestedWrapUpCodesMessage optionally provides explanatory text. Each suggestion declares a name; it may not contain a configured code identifier. Match the suggestion to a valid configured reason before using it.

Suggestions do not select a reason or complete wrap-up automatically. Keep manual reason selection available when suggestions are absent or have no configured match. Summary content and wrap-up suggestions can be available independently.

anchorSummary payloads and responses

Task API reference

Render available content

AISummary requires conversationId; its content fields are optional. Render the fields present in the payload rather than assuming a fixed sentence or format.

Check other optional fields before use. For example, the type does not require a sender name or a consult/transfer action field.

Build the response

Every AISummaryResponse requires:

The SDK reports the supplied values; your application manages response state and usage counts. The table below is an example UI policy for the supported state values, rather than behavior applied automatically by the SDK:

An exclusion choice is local until reporting. Keep the content available and preserve the choice across generation or retry. EXCLUDED records the choice for that submission; it does not establish a recipient access policy.

wrapUpCode is optional for post-call reporting. The optional summaryReceived field is not transmitted by these response methods in this SDK version and cannot serve as a delivery acknowledgement. Your application does not need to supply transport timestamps.

This helper creates an initial response from structured or plain text. Replace the draft with agent-reviewed content and update feedback, state, and counts before reporting:

import type {AISummary, AISummaryResponse} from '@webex/contact-center';

export function createSummaryResponse(
  summary: AISummary
): AISummaryResponse | undefined {
  const hasSectionText = Object.values(summary.sections ?? {}).some(
    (value) => typeof value === 'string' && value.trim().length > 0
  );
  const content = hasSectionText ? {...summary.sections} : summary.summaryText;
  if (!content || (typeof content === 'string' && content.trim().length === 0)) {
    return undefined;
  }

  return {
    summary: content,
    feedback: 'none',
    state: 'DEFAULT',
    numberOfTimesViewed: 0,
    numberOfTimesEdited: 0,
    numberOfTimesCopied: 0,
  };
}

The helper returns undefined when it finds no usable structured or plain text, including for card-only payloads. Handle supported card content separately or show an unavailable-content state; do not submit an empty default response.

anchorBeta limitations and error handling

• Concurrent operations: Allow one pending mid-call generation request per conversation, shared across consult and transfer actions, and one pending post-call request. Prevent duplicate reports while a submission is pending. Restore controls after success or failure.
• Cancellation and lifecycle: There is no dedicated public summary-request cancellation method. Remove listeners when their view is released, and ignore late results after the task changes or the session ends. Manage displayed content and retained history explicitly, and clear conversation content on logout.
• Retries: Handle generation and reporting rejections explicitly. A failed refresh can leave previously displayed content available. Offer a user-triggered generation retry without overlapping requests; do not assume automatic reporting retries.
• Inputs and diagnostics: Validate caller-supplied actions and response values. Treat rejected values as unknown, since not every error contains data.errorCode. Keep tokens and conversation content out of diagnostics.

anchorTry the Beta and verify your integration

Use the Contact Center SDK sample with an eligible test tenant, an authorized caller, two eligible agent sessions for consult and transfer, and a configured wrap-up reason. The hosted sample's deployed SDK version can differ from this guide's pinned Beta; check its build before testing.

In Auth & Registration, choose the environment and SDK toggles, supply an authorized test agent's access token, select webex.init(), then webex.cc.register(), and wait for completion. In Agent Station Login/Logout, select a permitted team and station option, select Login, and set the agent to Available.

Watch sdk ai summaries — 1 minute 34 seconds, by Ravi Chandra Sekhar Sarika, for a walkthrough of consult-summary generation, receiving-agent display, and post-call review, editing, feedback, and wrap-up. Your application must invoke the generation methods; binding listeners alone does not generate summaries.

Validate the following in your configured Beta environment. Record the SDK build, setup, and observed result, then complete the test calls and session cleanup.