# REOS AI: the platform's AI layer

<Lead>
All AI features of the platform run through **REOS AI**, the bundled AI layer. You never pick a model and never compare vendors: REOS AI automatically selects the fitting model per request and keeps optimizing in the background.
</Lead>

## The principle: mode instead of model

<DefinitionList>
  <DefItem term="Lite">
    The economy tier for simple tasks, considerably cheaper than Standard. Off by default; the owner enables it when wanted.
  </DefItem>
  <DefItem term="Standard">
    Everyday work: the right balance of quality and cost. Around nine out of ten requests run here.
  </DefItem>
  <DefItem term="Premium">
    For heavy tasks (analyses, long texts, tricky comparisons) with the strongest model usage, priced accordingly.
  </DefItem>
</DefinitionList>

Important: the mode changes only model choice and cost, never diligence. Replies are not artificially shortened in Lite.

Admins can pin a fixed model per mode under **Settings > Assistant > AI models** (text and image separately); the default remains REOS AI's automatic choice.

## Transparency and control

- **AI identifies itself:** where AI speaks outward in your office's name (for example the WhatsApp bot), it says so openly. AI-generated images carry a notice when shared. This matches the EU AI Act's transparency duty for limited-risk systems.
- **Changing actions only with your yes:** when the assistant wants to create or change data, you confirm it individually in the chat. Nothing happens without consent.
- **Third-party content is data, not commands:** content from CRM, files or web pages is treated as data. Instructions hidden inside are not followed; the AI never hands out credentials.
- **Humans decide:** AI proposes (findings, suggestions, drafts); the decision is yours. This holds in every tool.

## Costs

AI usage is billed in credits based on actual consumption, with automatic refunds on technical errors and the small-amounts rule (below 10 credits nothing is displayed). Details: [Understand and top up credits](/help/howto/topup-credits) and [Assistant modes and costs](/tools/agent/modi-und-kosten).

## BYOK: your own provider keys

From the **Max plan** upwards your office can store its own API keys of the major AI vendors (bring your own key). Costs then run through your provider contract instead of credits; in return you take over the data processing relationship with the vendor yourself. Everything important including the privacy consequences: [BYOK](/help/compliance/byok).

## Frequently asked questions

<Faq>
  <FaqItem q="Which model is behind my current request?">
    REOS AI decides automatically based on task and mode. You do not need to know it to get good results; admins can pin a fixed model per mode.
  </FaqItem>
  <FaqItem q="Does the platform train models on my data?">
    Your content is processed to answer your requests, not handed to third parties as training material. The binding details are in the privacy policy; an overview is in [Data and security](/help/grundlagen/datenhaltung-und-sicherheit).
  </FaqItem>
  <FaqItem q="What happens if no AI provider is configured?">
    AI features say so clearly instead of failing silently; some tools work with a deterministic fallback result so you can test the flow.
  </FaqItem>
</Faq>