Intelligent Atelier (Plugin)
The UBIK® Intelligent Atelier Plugin provides a user interface for interactive agentic AI modelling in UBIK® Studio.
Contents
Overview
The Intelligent Atelier can be used to develop part of a UBIK® customizing, debug errors or to find answers to UBIK®-specific questions. In the To-Do List, one can always see what is being planned and done. The artefacts view lists all currently relevant objects, ready for drag & drop to other UBIK® Studio controls.
Prerequisites
The Intelligent Atelier is a UBIK® Plugin for UBIK® Studio.
Installation
Install the plugin with the name "UBIK.Module.IntelligentAtelier" using UBIK Plugin Injection.
Configuration
For the feature to work, an AI model is required. Access to this service must be provided by the administrator. In UBIK® Studio's database connection settings, configure values for the following keys:
- AI REST API URL: Azure AI Foundry deployment URL
- AI REST API Key: Azure AI Foundry deployment Key
- AI REST API Model: The name of the deployed model
When this article was created, the supported models are GPT-4.1 and GPT-5.4, presumably including any version in between. Further, only the Azure OpenAI chat completion REST API is supported.
Accessibility
The Intelligent Atelier can be opened either as a Window or as a Tab Control. Use the "Magic Wand" toolbar menu icon or the "Intelligent Atelier" View-menu entry to open it in a window, or right-click the tab header in the main working area to open the Intelligent Atelier as a tab control, using the respective context menu entry.
Usage
Paradigm
The general usage workflow is to enter and send a request, upon which the Intelligent Atelier creates a plan and works it off, providing the results. Complex requests usually require a dialog to refine the results iteratively.
Use-cases
The Intelligent Atelier supports the following use-cases:
- Support: Answer a UBIK®-related question
- Concept: Provide a functional or technical customizing concept
- Debugging: Find a configuration error
- Bugfixing: Fix a misconfiguration
- Prototyping: Implement requirements
Request
In the "Request" field (1.), the user can enter any prompt corresponding to a use-case listed above, and start processing using the Play-button.
Response
Once the Intelligent Atelier has come to a conclusion, it will output an answer in the "Response" field (2.). This can take a while, because there is a lot of expert knowledge to process for the Atelier. Further, similarly to an interaction with a coworker, complex requests usually require a dialog, with the user reviewing the (intermediate) results and providing follow-up instructions.
To-Do List
For most non-trivial use-cases, the Intelligent Atelier establishes a plan in the form of a To-Do List (3.). Hover the items to inspect its notes in a tool-tip!
Artefacts
All relevant objects accessed, created or edited by the Intelligent Atelier are listed in the Artefacts area (4.). These items can be dragged and dropped to most other UBIK® Studio controls, including tab controls like the Bulk Editor, Relation Editor or Object Explorer.
Tips and tricks
- Create a backup of the database before starting an Atelier session.
- To restart a session completely, close the Intelligent Atelier control or window and open a new one.
- The quality of the results depends, amongst other factors, on the quality of the request. The clearer and more precise a prompt is, the easier it is to process.
- Ambiguity or even contradiction are some of the most difficult obstacles for the Intelligent Atelier; the results are rarely satisfactory for requests with these properties.
- Sometimes, the modelling gets off on the wrong foot. When things just don't work out, often it's best to restart with a new session.
- Sessions are not persisted and cease to exist when closing the Atelier view. The configured REST API may or may not persist details about the interaction, this depends on the configured service.
Supported scope
When this article was created, the Intelligent Atelier had an understanding of the following concepts:
- Basic ER-models with MetaClasses, MetaProperties, References, Relations
- Basic View Items (Relation, Reference, including the reverse cases)
- MetaClass Scopes
- Basic View configuration including explicit root objects
- Not yet supported: Queries, Proxies, Plugin content like MRO, etc.
Technical background
The Intelligent Atelier uses an agentic loop with a mixture-of-experts architecture on application level. An Azure OpenAI chat completions REST API is accessed in combination with a rich set of functions for tool-calls and elaborate system instructions to enable both a deep understanding of UBIK® and an algorithm for processing complex requirements.
