Meeting Summary from Audio

The crm-report-demo reaktor (in operaide-apps/apps/op-demo-crm-report) converts a recorded meeting into a structured CRM report, stores it via a mock ERP REST API, and returns the created report URL.

Reaktor registration

import { aktorSetting, registerReaktorDefinition } from '@operaide/aktor';
import { z } from 'zod';
import { aktorCreateCRMReport, aktorGenerateAnswer, aktorSaveReport } from './aktors';

registerReaktorDefinition({
    reaktorDefinitionId: 'crm-report-demo',
    label: 'CRM Report',
    description: 'A CRM report for a given company',
    createReaktor({ audio, customerName }) {
        const apiKey = aktorSetting(z.string().describe('API Key for the ERP'), '5440aa7de35c3be:681d56e565bd709');
        const baseUrl = aktorSetting(z.string().describe('Base URL for the ERP'), 'https://erp.case.operaide.ai');

        const report = aktorCreateCRMReport({ audio, customerName });
        const savedReportUrl = aktorSaveReport({ reportJson: report, baseUrl, apiKey });
        return aktorGenerateAnswer({ reportUrl: savedReportUrl, report });
    },
    inputSchema: z.object({
        audio: z
            .string()
            .describe('[file-upload] Audio file in base64 format')
            .refine((val) => val.startsWith('data:') && val.includes(';base64,'), {
                message: 'Must be a data URL with base64 encoding',
            })
            .openapi({
                example: 'data:audio/mp3;name=recording.mp3;base64,//uQxAAA...',
                description:
                    'Base64 encoded audio file. Use data URL format (data:type;name=filename;base64,data) for auto-detection of MIME type and filename',
            }),
        customerName: z.string().openapi({ example: 'John Doe' }),
    }),
    outputSchema: z.string().describe('[textarea] JSON output'),
});

Aktor pipeline

aktorCreateCRMReport orchestrates the heavy lifting. It:

1. Transcribes the uploaded audio with aktorSpeechToText.
2. Generates a summary, follow-up tasks, and an email subject via dedicated LLM prompts.
3. Builds the final CRM payload and returns it as JSON.

export function aktorCreateCRMReport(input: { audio: Aktor<string>; customerName: Aktor<string> }) {
    return defineAktor(
        'aktorCreateCRMReport',
        {
            createAktor() {
                const transcript = aktorSpeechToText({ audio: input.audio });

                const summary = aktorCreateSummary({ transcript });
                const todos = aktorCreateTodos({ transcript });
                const subject = aktorCreateSubject({ transcript });

                const startDate = aktorGetDate({});
                const endDate = aktorGenerateEndDate({ transcript, startDate });

                return aktorCreateReport({
                    summary,
                    todos,
                    customerName: input.customerName,
                    subject,
                    startDate,
                    endDate,
                });
            },
        },
        input
    );
}

aktorSaveReport posts the generated JSON to the mock ERP endpoint, while aktorGenerateAnswer formats the final response (including the saved report URL).

Sample request

Send a request similar to the following payload (trimmed for brevity):

POST http://localhost:8811/api/v1/aktor/crm-report-demo/run
Content-Type: application/json

{
  "audio": "data:audio/mp3;name=call.mp3;base64,//uQxAAA...trimmed...",
  "customerName": "Contoso GmbH"
}

The response body contains the generated CRM report JSON together with the URL returned by the mock ERP API.