OpenAI

Introduction

Within the Asset Manager, OpenAI offers the ability to convert human questions formulated in German or English into Structured Query Language (SQL). This process utilizes OpenAI’s internal API. Since any query can generally be submitted to OpenAI, the AI is instructed to respond within a defined context using an SQL script. This script is then executed, and the result is displayed. To ensure data protection, the following procedure is applied:

No data is transferred from the Asset Manager database to OpenAI. However, to enable the AI to generate meaningful responses, the following information ais transmitted:
- The system prompt, which contains general instructions on how the AI should behave
- The database structure with table and column information (without data)
- The user’s queries to the AI

For this to work, an API key is required. This must be stored in the Asset Manager settings, as described here.

Input Options

Requests to OpenAI can be made in written or spoken form.

• In writing: To do this, go to the “Users” section and enter your question in the language set for the Asset Manager (German or English). After entering your question, simply click the “Chat” button (or press F6), and OpenAI will provide an answer shortly.
• Verbally: It is also possible to formulate a question verbally. To do this, Microsoft Speech Services must be configured within the Asset Manager as described here. Additionally, an input device (microphone) must be selected from the list beforehand. Recording can then be started by clicking the Record button (or pressing F5). The query can be transmitted to OpenAI in two ways:
  
  • You stop speaking, which is automatically detected. Then simply click the Chat button (or press F6), and OpenAI’s response will appear shortly.
  • You stop speaking, and the query is automatically sent to OpenAI. This requires that the End-of-speech detection checkbox be enabled.

Asset Manager AI

The Asset Manager AI provides local AI capabilities for the Asset Manager. It runs on a separate server within the local LAN. Unlike the OpenAI integration, the models are not accessed via an external API but are executed locally on the server.

Server Requirements

The server requires high-performance hardware:

• fast CPU
• at least 32 GB of RAM
• at least 16 GB of VRAM
• high-performance graphics card

More RAM and more VRAM are recommended. The Large Language Models (LLMs) that can actually be used depend particularly on the available resources of the graphics card.

Installation and Components

Asset Manager AI can be installed on the following systems:

The following components are also used:

• Ollama for local execution of various Large Language Models (LLM)
• Poppler tools for processing PDF files

This combination allows various Large Language Models to be run locally, provided that the CPU, RAM, and especially the graphics card provide sufficient resources.

Data Protection

The Data never leaves the server. All requests, document content, intermediate results, and outputs remain locally on the server within the LAN. No data is transferred to external AI services. This ensures that the data remains within your own infrastructure. This enables GDPR-compliant processing, as personal data is not transferred to external AI providers.

Features

Asset Manager AI supports the following features:

• German ↔ English Translation Assistant
  Texts and terms can be translated from German to English and from English to German. For more information, see: Translations
• Text-to-SQL
  Natural language queries can be converted into SQL queries, as described at the beginning of this chapter. Processing takes place locally via Asset Manager AI.
  
• Convert PDF Invoices to SQL Tables
  PDF invoices can be processed, analyzed, and converted into SQL tables. The Poppler tools are used for PDF processing.