Skip to main content

Writing & Running Scripts

The assistant can write Python that drives Volvicon through its Scripting API and run it for you. You approve each script before it runs.

Running a script​

  1. Describe what you want in plain language — for example, "write a script to create a sphere", "export every surface to STL", or "smooth the active mask". The assistant replies with the script.
  2. Tell the assistant to run it — say "run this", "execute it", or "apply it".
  3. Volvicon shows an approval dialog with the exact script before anything runs. Review it, then approve or cancel.
  4. The script's output appears directly in the chat.

You can also copy any script into Volvicon's Scripting Editor and run it there with Ctrl + R.

Review before running

Treat generated scripts as drafts to check, especially anything that deletes, overwrites, or exports data. If a script will modify or remove data, the approval dialog states this. When in doubt, cancel and ask the assistant to explain what the script does.

When a generated script has an error​

Local models can make small mistakes in generated scripts — a wrong argument, an incorrect method name, or the wrong calling style. This is more common with smaller models. When it happens:

  • Ask the assistant to fix it. Paste or describe the error ("that failed with …, please fix it"), or say "rewrite it". The assistant uses the error to correct the code.
  • If the assistant mentions keyword arguments, ask it to "rewrite it using positional arguments". Volvicon's Scripting API expects values in order, not by name.
  • For stubborn cases, edit the script yourself in the Scripting Editor.

Best practices​

  • Review generated scripts before running them.
  • Use a capable model. Gemma 4 and Gemma 3 12B produce more reliable scripts; smaller models are faster but need more fixes. See Local AI.
  • Keep to one goal at a time. Long, multi-part requests are harder for smaller models to complete correctly. Break large workflows into steps.
  • Iterate. If the first script is not right, describe what went wrong; the assistant improves with feedback.
  • For code only, turn off "Allow the assistant to run scripts" in Settings. The assistant still writes scripts for you to run manually.

Response focus (temperature)​

Settings includes a Response focus control (the temperature), which governs how focused or varied the answers are:

  • Lower (around 0.2, the default): precise, consistent answers and more reliable scripts. Recommended for documentation questions and code, and the default for that reason.
  • Higher (toward 1.0): more varied, but less predictable and more prone to mistakes in code and API names.

Most users should leave this at the default. Raise it only if you want more varied phrasing and accept less consistency. It applies to every provider, local and cloud.

Grounded in the Scripting API

When retrieval is available, the assistant uses the Scripting API reference as context for generated scripts. Review each script before running it. If a name is not recognized, tell the assistant the error and it can revise the code.