AI assistant

The help assistant is an AI. It answers questions on using SentryAssets and on the IT-Grundschutz, such as “How do I create a rollout token?” or “What does the backup module require?”. It stays off until the administration switches it on.

Support, not an audit

AI answers can be wrong. Every answer names its sources; check them before relying on it.

What the model receives

• the question;
• the path of the page you ask from, without identifiers;
• the matching passages from the help texts of the web UI and, for roles allowed to read the IT-Grundschutz, from the imported Kompendium.

The model never receives inventory data. Outside its topics or without sources the assistant gives no answer. Questions about your own data are answered by the data assistant (Pro), which is switched on separately.

Setting up

1. Open Settings › AI assistant.
2. Choose the provider:
   • Mistral (data centres in the EU), with an API key;
   • an OpenAI-compatible endpoint, such as your own model through Ollama or vLLM.
3. Test the connection and switch the assistant on.

• Connections over https with TLS 1.3 and certificate verification only.
• The API key is stored encrypted and not shown again.
• A model in your own network needs its network in SA_ASSISTANT_ALLOWED_NETWORKS, such as public,192.168.10.7/32. By default only addresses in the internet.
• There are limits per person and organisation. The audit log records questions without their text.

Asking

Open the assistant through AI assistant in the navigation or the command palette (Ctrl + K, “Ask the AI assistant: …”). Each question is answered on its own; the answer is marked as AI and names its sources.

This documentation as knowledge base

Today the help texts of the web UI and the Kompendium are the knowledge base. This documentation is to be added later.