Setting up AI (Ollama, whisper.cpp, privacy)

Goal

The AI features run – preferably on your own server – and every website only uses what is enabled.

Prerequisites

• Feature “ai” in the feature set.
• An AI provider: Ollama (own server), whisper.cpp for captions or an external service with a data processing agreement.

Step by step

Transcript: every step matches one subtitle in the video.

1. Grundeinstellungen › KI: switch on per website and choose capabilities – text, images, search, captions.
2. A daily limit, notes on audience and style, and a glossary steer the assistant.
3. “Fähigkeiten” (capabilities): provider and model are set by the agency in the configuration – e.g. Ollama and whisper.cpp.
4. “Bleibt auf dem Server” (stays on the server): with local models content never leaves your premises. External services are marked.
5. Configuration and commands (ai:test, search:index) are in the technical documentation › KLXM Ai.

Commands & configuration

// config/config.local.php (Installation) oder config/sites/{website}.php
'ai' => [
    'provider' => 'ollama',
    'base_url' => 'http://127.0.0.1:11434',
    'models'   => ['text' => 'gemma3', 'embed' => 'bge-m3', 'vision' => 'gemma3'],
    'transcribe' => ['backend' => 'whisper_cpp', 'model_path' => '…/ggml-large-v3-turbo-q5_0.bin'],
],

php bin/console ai:test --cap=text
php bin/console search:index --all

Tips & pitfalls