onSummary | SignalWire

Lifecycle hook invoked when a post-prompt summary is received after a conversation ends. Override this method in a subclass to process summaries — for example, saving them to a CRM, triggering follow-up workflows, or logging call outcomes.

A post-prompt must be configured via setPostPrompt() for summaries to be generated. With a dynamic config callback set, this runs on the request’s configured copy of the agent. A summary delivery must carry the token the SDK adds to the post-prompt URL for that call, or it is refused with 403.

The default implementation does nothing. You must override it in a subclass or set a setPostPromptUrl() to receive summaries at an external endpoint.

Parameters

summary

PostPromptData | nullRequired

The summary object generated by the AI based on your post-prompt instructions. null if no summary could be extracted from the response.

rawData

PostPromptRequired

The complete raw POST data from the post-prompt request, including metadata like call_id and the full AI response.

Returns

void | Record<string, unknown> | Promise<void | Record<string, unknown>> — For a request whose action is fetch_conversation, return the conversation (for example { conversation_summary: '...' }); the post-prompt endpoint sends it back instead of { success: true }. Otherwise the return value is ignored.

Example

import { AgentBase } from '@signalwire/sdk';
class SupportAgent extends AgentBase {
  constructor() {
    super({ name: 'support', route: '/support' });
    this.setPromptText('You are a helpful customer support agent.');
    this.setPostPrompt(`
      Summarize this call as JSON:
      - intent: caller's primary intent
      - resolved: yes/no/partial
      - sentiment: positive/neutral/negative
    `);
  }
  override async onSummary(
    summary: Record<string, unknown> | null,
    rawData: Record<string, unknown>,
  ): Promise<void> {
    if (summary) {
      const callId = (rawData.call_id as string) ?? 'unknown';
      console.log(`Call ${callId} summary:`, summary);
      // await saveToDatabase(callId, summary);
    }
  }
}
const agent = new SupportAgent();
await agent.serve();