AI assistant

The assistant is a local AI tool inside Studio. You use it in two places:

• The Assistant panel on the right: you type requests that do work in the project, or ask questions.
• The Assistant panel in the code area: you ask questions while writing a script and try the code from the answer before adding it.

The assistant exists only in Studio. It does not run in Live.

Writing to tables goes through the same validation as Excel import: unique names, selection values and PLC address format. For timed scripts a task is added as well. The panel design wizard opens with the requested settings, and you press Create yourself.

Everything done by one request is a single undo step. The Undo button or Ctrl+Z reverts the whole action.

• Code that deletes something asks first. The box names what will be deleted, and you choose Apply or Cancel.
• If the code fails, Studio reverts the whole action and asks the model for corrected code. This happens at most twice, and it stops if the same error comes back.
• If the code changed nothing in the project, the warning No change was made to the project. appears. The model saying it is done does not count.
• Show code opens the code that ran.

The Assistant button in the header of the script code area opens the panel to the right of the code. The model answers in the language of the open script: VBS, JavaScript, Python or C. This panel does not change the project. Code enters the editor only when you press a button.

• Insert at cursor: inserts the code at the cursor, or replaces the selected code.
• Replace all: replaces all the code in the editor.
• Copy: copies the code to the clipboard.

The code in the answer goes through the Issues check: syntax, tag and screen names in the project, and script permissions. This check does not run the code.

Studio runs the code from the answer in a simulated tag environment with two scenarios. The code writes nothing to the PLC, and the project’s tags do not change.

The result shows the tags read and written, notifications, opened screens, tags that are not in the project and runtime errors. If there is a runtime error, a timeout or a missing tag, Studio asks the model for one correction. You still add the corrected code to the editor yourself.

Studio starts the model as a separate program on this computer. That program listens only on 127.0.0.1, so other computers cannot connect to it. Studio sends requests only to this program, and the project is not sent out. The assistant needs no internet connection. When Studio closes, the model closes too.

The model receives a project summary: connections, tags, screens, scripts, the open screen and the selected objects. In the code area it also receives the open code and the selected code.

The model does not load when Studio starts. It loads the first time a panel opens, and a request typed in the meantime waits. Response speed depends on the model size and the hardware.

• The model can misunderstand a request. Review what it did, and test the project before going live.
• The assistant does not deploy the project to Live, does not run it and does not connect to a PLC.
• It does not cover recipe and report design screens, menu and toolbar design, object animations or creating faceplates.
• A dry run tries only two fixed sets of values and does not read live tag values.
• In module scripts a procedure is defined but not called. Object properties, CreateObject, file and database operations are not tried in a dry run.