AI Architecture

1. Overview

Centrify 360 AI adds a conversational assistant on top of your Business Central analytics environment.
The AI service is deployed separately from the ETL orchestrator and connects to Microsoft identity, Power BI Online, and your chosen large language model provider.

AI usage requires a separate Centrify 360 AI add-on license.
The base Centrify 360 license does not include AI usage rights.

2. High-Level Architecture

At a high level, the AI flow works as follows:

  1. A user opens the Centrify 360 AI interface.

  2. The user signs in with Microsoft Entra ID.

  3. The AI service authenticates using the configured Power BI App Registration / service principal.

  4. The AI service accesses Power BI datasets and metadata according to the granted permissions.

  5. The prompt is sent to your selected model provider.

  6. The response is returned to the user together with the relevant analytics context.

3. Authentication

Centrify 360 AI uses Microsoft OAuth for user sign-in.
The same Power BI App Registration is a core prerequisite for AI.
Refer to Power BI App Registration for setup details.

4. Bring Your Own Model

Centrify 360 AI is designed as a bring-your-own-model architecture.
You provide the model endpoint, API key, and deployment or model name.

Supported provider patterns include:

  • Azure AI Foundry / Azure OpenAI

  • Anthropic

  • DeepSeek

  • Gemini

5. Recommended Models

The following models are recommended starting points:

  • GPT-5.4

  • Claude Sonnet 5

  • DeepSeek V4 Pro

  • Gemini 3.6 Flash

6. Cache Support

Centrify 360 AI requests are typically large because they include semantic model context, dataset metadata, instructions, and user prompts.
For this reason, models with prompt caching support are strongly recommended.

In practical deployments, cache support can reduce AI cost by approximately 70% to 80% for repeated usage patterns.

7. Power BI Permissions Required for AI

To use Centrify 360 AI, users must have sufficient access to the Power BI dataset being queried.
At minimum, the user must have access to the relevant workspace and dataset and must have Build permission on the dataset in Power BI Online.

8. Deployment Inputs Required

When enabling AI, you typically provide:

  • Power BI App Registration credentials

  • AI endpoint URL

  • AI API key

  • AI deployment or model name

These values are entered during:

9. Cost Planning

For model-by-model examples and monthly usage simulations, see AI Cost Simulation.