General Setup zapHome OnPrem AI Search

1. General Information zapHome OnPrem AI Search

zapHome OnPrem AI Search is the secure, local technical platform for all zapliance solutions including modern MCP agents. It is designed to analyze SAP data directly within your organization’s IT environment, ensuring maximum data privacy by enabling seamless integration with your own Azure OpenAI services. Controlled outbound connections are used for specific external lookups. All installations are performed on-premise using Docker, so your data always remains within your infrastructure and zapliance never has access to your zapHome instance.

The zapliance agents form an expert team dedicated to SAP data analysis. Specialized agents handle tasks such as SAP data extraction, structured data labeling, and comprehensive risk and opportunity assessments based on the zapliance indicators selected for you. All analysis is conducted exclusively within zapHome and can be flexibly processed further—for example, using Power BI Desktop or through collaboration with other zapliance agents. This can be combined with the targeted use of selected external data sources and services.

Additionally, some agents are designed to interact directly with business professionals, understanding your unique business context. Using a chat interface, you can provide instructions to these agents, who independently select and execute the most suitable tools from their toolkit, document the results, and share their insights with you in the chat. While these agents can work together using each other’s outputs, they are also able to function as regular chatbots, leveraging integrated algorithms and targeted web search capabilities to answer your questions. As the customer, you retain full responsibility for the final evaluation of results.

With zapHome OnPrem AI Search, you benefit from a secure, customizable, and collaborative environment for advanced SAP data analysis—fully under your control.

You need to set up the following components to get zapHome alive:

2. Virtual Machine/Server

A Linux-based virtual machine (VM) or server that meets the specified requirements is prepared as the host environment for Docker.

Details can be found in the Knowledge Base article: Setup Guide Virtual Machine / Server

We recommend Docker as the container management tool. For other deployment scenarios, please refer to this article.

3. Docker Setup for zapHome OnPrem AI Search

zapHome OnPrem AI Search is deployed using Docker on the dedicated server or VM. zapliance publishes and manages its images through image registries, ensuring access to the latest verified versions.

Details can be found in the Knowledge Base article: Setup Guide Docker

4. Docker Containers

Docker containers provide the ability to run applications together with all required components in isolated environments, ensuring consistent and independent operation across systems.

4.1 NGINX Reverse Proxy Server

To enable secure browser-based access, all communication between the user's web browser and the containers is routed through a reverse proxy server. It acts as a single gateway, allowing for centrally managed SSL/TLS encryption. Thus, only one certificate is needed to secure the entire environment, simplifying the setup.

4.2 zapAnalytics

The zapAnalytics software is provided as a Docker container.

Details can be found in the Knowledge Base article: Setup Guide zapAnalytics

4.2.1 RFC-User Access

zapAnalytics accesses the SAP system for data processing via RFC using read-only permissions.

Details can be found in the Knowledge Base article: Setup Guide SAP

4.2.2 Content Bundles

A content bundle wraps all analysis logic into a .zip file. After the relevant company codes, fiscal years, and client information are provided, zapliance generates and shares the content bundle (.zip) via an exchange folder in OneDrive. You must upload the content bundle .zip file in zapAnalytics to create a project.

Details can be found in the Knowledge Base article: Content Bundles

4.2.3 Parquet Files

After data processing is completed, result files specific to each solution are made available for download. The generated parquet files serve as the foundation for creating Power BI dashboards.

Details can be found in the Knowledge Base article: Setup Guide Power BI Reports

4.2.4 MCP Server

zapAnalytics can run integrated MCP server for each project to access the results database. Once launched, the MCP server can be connected to the Wilma Agent in LibreChat for further processing and analysis.

Details can be found in the Knowledge Base article: Setup Guide zapAnalytics

4.2.5 Selected External Web-Services (e.g. VAT ID Check)

zapAnalytics can make use of external web services for automated or manually triggred agent tasks. E.g. an automated check of the registred VAT ID at official authorities is possible. Thus, zapAnalytics will set up https connections to specific external services. The firewall must allow such a connection.

4.3 LibreChat (Chat interface)

LibreChat is provided as a Docker container by zapliance, which means that a separate installation of LibreChat on your target system is not required.

Details can be found in the Knowledge Base article: Setup Guide LibreChat
Details on zapAgent Wilma can be found in the Knowledge Base article: Agent Onboarding Wilma

5. Azure OpenAI

The use of Azure OpenAI is required for operating LibreChat, as it enables the integration of advanced language models and ensures compliance with enterprise security and data privacy standards.

Details can be found in the Knowledge Base articles:

- Setup Guide Azure OpenAI (including Web Search)
- Setup Guide Document Intelligence (depending on the solution)

6. Power BI Desktop (on End-User Device)

Once data extraction and processing are completed within zapAnalytics, the Power BI report templates (.pbit) need to be opened using Power BI Desktop for importing the results from the zapAnalytics result files (.parquet). This will create Power BI reports (.pbix) that are accessible locally by the client, serving as the basis for the workshop session.