This is the first article in a series about connecting external data sources for processing with Automic AI features in version v26.

Just as REST has become the standard for web services, the MCP (Model Context Protocol) standard has become established for AI models. MCP enables AI models to connect securely to external data sources, tools, and APIs. It serves, so to speak, as the gateway to data sources and APIs.

This makes it possible to give an AI model direct access to a database in order to analyze and process its data. Integration with a ticketing system is another typical use case.

For the AI features, the Automic team developed the Automation.AI component. This component provides the functionality for connecting the LLM and the MCP. Automation.AI is the name of the underlying component. The features it provides are grouped under the term “Gen AI.”

The Automation.AI component is launched separately via the Automic Service Manager, similar to an analytics backend. Configuration is done via the “application.properties” file. More on this in the following blog posts.

In my opinion, the way an Automic AI job is structured is well designed. Once an MCP has been configured in Automation.AI, it can be selected, and the list of MCP tools can be easily compiled. I’ll go into detail on this later in the series, as it’s relevant to security.

For an Automic AI job to work, the LLM (AI provider) must first be configured. Google Gemini, OpenAI ChatGPT, Azure OpenAI, Ollama, and VMware PAIS are available for this purpose. The configuration process is described in the Automic Help Center. Later in this series, I’ll go into more detail about connecting local AI models.

A new feature is the AI Job object, which uses the Connection object. As is common with AI, two types of prompts are available. Simply put, the system prompt specifies the context in which the AI is to operate. For example, if the AI Job is supposed to prepare data for a sales statistics report, the system prompt can explicitly define this. To ensure predictable results, a well-crafted system prompt is very important.

The user prompt defines what the AI job is supposed to do. Here, too, it is necessary to describe in detail what is expected and how the result should be presented. In automated environments, long and well-crafted system and user prompts are a must to ensure reliable results.

It is important to note the AI output variable, which stores the result of the AI analysis in a runtime variable. This can then be further processed in subsequent jobs.

The “Conversation ID” is also interesting because it allows you to build on the results of a previous AI job.

Outlook

In the following blog post, I explore the question of when it makes sense to use AI jobs, or whether traditional job objects are still the better solution.