AiPBX
Özellikler Kurulum Videolar Mobil Belgeler Canlı demo İletişim GitHub'da Sponsor Olun UpCloud ($250 Ücretsiz Kredi) YouTube GitHub
ODOO AiPBX santralinizi Odoo CRM & ERP ile entegre edin — tek platformdan yönetin. Odoo'yu İncele →
Ana Sayfa / Documentation / AI Voicemail Summaries & Transcripts

Teknik dokümantasyon yalnızca İngilizcedir; GitHub ve AiPBX portalındaki (Yönetim → Yardım) metinle aynıdır.

AI Voicemail Summaries & Transcripts

The community ai-voicemail plugin displays completed voicemail transcripts and summaries at /plugins/ai-voicemail. AiPBX 1.72.12 includes the corrected event subscription and full transcript payload. See the plugin developer guide.

Enable transcription

  1. Enable voicemail transcription under AI → Cloud services → Voicemail, and configure a speech engine through Cloud AI or Local AI Models.
  2. Install AI Voicemail Transcription & Executive Summary from the community catalog under Integration → Plugin Catalog.
  3. Open /plugins/ai-voicemail to view completed results.

The plugin consumes the core's voicemail.transcribed event. It does not transcribe audio itself. Without a completed transcript or summary, it creates no summary entry and makes no LLM request.

Event flow and storage

The core's VoicemailAiService transcribes the recording and optionally summarizes it, storing its result in voicemail_ai. The detached voicemail notification script emits voicemail.transcribed with the extension, caller, duration, summary preview and full transcript when available.

The plugin optionally sends that transcript to its configured LLM, then stores the transcript and summary in the summaries collection of sys_plugin_kv. These entries appear in the plugin admin page. The same completed event is skipped on redelivery. The core caps each KV collection at 500 items; this is not a configurable time-based retention policy.

The result is plain summary text. There are no structured urgency fields, synchronized transcript timestamps or separate action-item records. If the LLM fails or returns nothing, the plugin preserves the core summary, or uses the first 200 transcript characters when no summary is available.

Optional LLM settings

SettingBehavior
API Base URLOpenAI-compatible base URL, for example https://api.openai.com/v1. The plugin appends /chat/completions. Administrator-configured HTTP endpoints are also accepted for internal services.
API KeyA non-empty value enables the LLM request and is sent as a Bearer token. Empty means reuse the core result without an LLM call. For a local server without authentication, use a non-empty placeholder only if that server accepts it.
Summarization LLM ModelModel available on the configured server, for example gpt-4o-mini.
Summary Output LanguageTurkish, English or the language of the message. Applies to the optional LLM request.

The LLM request has a 12-second total timeout and a 5-second connection timeout; redirects are disabled.

Notifications and privacy

The voicemail player, email and mobile notification integrations use the core voicemail AI result. The plugin's optional custom summary is stored for its admin page; it does not replace the core result or send a second email/push notification. See voicemail and mail settings.

Core transcript redaction runs when enabled by the privacy settings. The plugin does not apply a separate redaction filter or check a cloud AI lock before its optional HTTP request. Configure its endpoint and credentials according to your data policy. Choosing a local LLM alone does not ensure all processing is local: the core's speech and summary engines must also be configured accordingly.

Moving from the bundled version

The next core upgrade moves an installed bundled copy to /var/lib/aipbx/plugins/ai-voicemail using an offline transition package. It preserves its enabled/disabled state, encrypted settings and stored summaries. Existing uploaded copies are preserved. New installations do not install it automatically. Later plugin updates come from the community catalog. The core transcription service remains in AiPBX.

View or edit this page on GitHub

On This Page