Architecture Overview

ABA follows a clean layered architecture that keeps each component independent and replaceable.

For a detailed technical deep-dive, see System Architecture.

High-Level Flow

flowchart TB
    subgraph Client["Client Layer"]
        WebChat["Web Chat"]
        Teams["MS Teams"]
        Slack["Slack"]
        Custom["Custom App"]
    end

    subgraph AI["AI Layer"]
        LLM["Large Language Model"]
    end

    subgraph MCP["MCP Server"]
        FI["FI Tools"]
        CO["CO Tools"]
        MM["MM Tools"]
        SD["SD Tools"]
        More["..."]
    end

    subgraph SAP["SAP System"]
        OData["OData Services"]
        Auth["Authorization"]
    end

    Client -->|"Natural language"| AI
    AI -->|"MCP Protocol"| MCP
    MCP -->|"OData / RFC"| SAP
    SAP -->|"Data (authorized)"| MCP
    MCP -->|"Formatted result"| AI
    AI -->|"Conversational answer"| Client

Core Principles

Security First

All SAP calls use the authenticated user's credentials. ABA never uses a service account to access business data.

Standards-Based

Built on SAP's standard OData services and the open MCP protocol. No proprietary lock-in.

Modular

Each SAP module is implemented as independent MCP tools. Enable only what you need.

LLM-Agnostic

Works with Claude, Gemini, Azure OpenAI, or self-hosted models. Switch providers without changing the integration.

Enterprise-Grade Privacy

ABA only connects to models hosted by trusted hyperscalers (such as Microsoft Azure, Google Cloud Platform, or AWS) or via SAP AI Core when applicable. When using these providers, your prompts and SAP data are processed securely and remain logically isolated: your data is not used to train public models nor exposed to other customers.