AiPBX
Xüsusiyyətlər Quraşdırma Videolar Mobil Sənədlər Canlı demo Əlaqə GitHub-da sponsor olun UpCloud ($250 Pulsuz Kredit) YouTube GitHub
ODOO AiPBX-i Odoo CRM & ERP ilə birləşdirin — zəngləri və müştəriləri tək mərkəzdən idarə edin. Odoo-nu Sına →
Ana Səhifə / Documentation / AI Screens (`/ai-local`, `/ai-cloud`, `/ai-voicemail`, `/ai-call-analysis`)

Texniki sənədlər yalnız ingilis dilindədir; GitHub və AiPBX portalındakı (Admin → Help) mətnlə eynidir.

AI Screens (/ai-local, /ai-cloud, /ai-voicemail, /ai-call-analysis)

The AI menu splits local models, cloud providers, and per-feature settings into dedicated pages:

  • Local models (/ai-local): Install or remove the local AI runtime (aipbx-ai) and Ollama, download/run/stop Piper TTS, Faster-Whisper STT, ONNX embedding, and Ollama LLM models, and assign local defaults, languages, and Try playground.
  • Cloud AI (/ai-cloud): Configure cloud AI provider accounts (OpenAI, Google Gemini, Groq, Deepgram, ElevenLabs, Azure, Amazon Polly, or OpenAI-compatible servers), monthly quotas, and test connections.
  • PBX Assistant (/ai-assistant): Assistant model binding, knowledge base (/ai-knowledge), and chat retention (defaults to 10 days; nightly pruning).
  • AI Voicemail (/ai-voicemail): Enable/disable voicemail transcription and summary, choose per-feature engines, and inspect recent ai_jobs entries.
  • Call Quality Analysis (/ai-call-analysis): Configure automated Call Quality Review (Auto-QA), review criteria, and inspect recent ai_jobs entries.

Legacy URLs /ai-models and /ai-settings permanently redirect (301) to /ai-local (with ?tab=cloud → /ai-cloud, ?tab=assistant → /ai-assistant, ?tab=voicemail → /ai-voicemail, ?tab=cdr → /ai-call-analysis).

See also: Cloud AI · Local AI

Local language models with Ollama

Under AI Settings → Local models, choose Install / start Ollama. Setup runs in the background and shows its progress or an error. It installs the official Linux archive and starts a service bound to 127.0.0.1:11434, with Ollama cloud features disabled. Existing Ollama models are retained. This button does not download a model.

Models already installed in Ollama appear in the model list and Assistant model selector without requiring the Python speech runtime. Install the speech runtime separately to use the Hugging Face catalog. GGUF and safetensors language models are verified by SHA-256, imported into Ollama, and marked ready only after successful loading. Model architectures must be supported by Ollama; unsupported models show an error rather than a ready state. Select a ready local language model under PBX Assistant → Model (or by language under Engines).

Hugging Face access tokens are used only in memory for inspection and the initial download. They are never saved in manifests. Retrying a protected model download after a service restart requires supplying the token again.

Chat history defaults to ten days and is pruned nightly, including old messages in active conversations. Choosing Off deletes saved chats immediately and prevents subsequent chat persistence. PBX change audit records remain separate.

The settings merge preserves old plugin settings and KV records, including existing voicemail summaries, under their original namespaces.

View or edit this page on GitHub