Skip to main content

Creating, editing and system rules

Creating a system​

To create a system, click the "New system" button on the home page.

creation_systeme

Fill in the name of your system, describe which game it is drawn from, and you can add the author.

Suggested sections​

As mentioned on the home page, rules are structured into several sections.

On the UI side, the editor offers default sections: Combat, Classes, Stats, Magic, Monsters, Progression.

regles_systeme

You are free to add others. But keep in mind:

The Lore module is used to describe the universe (factions, places, legends). The Game System should only describe the game mechanics, not the lore. This separation is intentional — the AI injects rules into contexts different from the lore, and mixing the two degrades the relevance of the prompts.

Importing a rules PDF​

Rather than retyping your rules by hand, you can import a PDF: the application extracts the text, splits it into chunks, then the AI automatically structures it into sections (Combat, Classes, Stats…) that it adds to your editor.

Starting an import​

  1. In the rules editor, click Import a PDF.
  2. Choose a file in PDF format (only this format is accepted).
  3. The analysis starts. It can take from a few seconds to several minutes depending on the size of the document and the AI provider used.

Import_IA

During processing, progress is shown in two stages:

  • an extraction phase for the PDF text (no bar yet, the size is not known at this point);
  • then an analysis bar that advances chunk by chunk, with the titles of detected sections appearing as they are found.

Here is what it looks like during the import: Import_IA_encours

And the result:

Import_IA_resultat

If your AI provider is momentarily overloaded, a status message informs you of the automatic retries: simply let the import continue.

Prefer a cloud model

Unless you really own an exceptional machine capable of running very heavy AI models, I recommend using a cloud provider. PDF processing is quite demanding and requires a lot of resources. Even a machine able to run a gemma4:26b, for example, is generally not enough — at least for now.

Sections are merged, not replaced​

Important

The import adds the extracted sections to those already present in the editor; it does not replace them.

  • A brand-new section title creates a new card.
  • An already existing title (e.g. a "Combat" section that already exists) has the imported content appended after the existing one.

Re-importing the same PDF therefore stacks the contents. If needed, clear the relevant sections before starting another import.

Nothing is saved automatically: the imported sections appear in the editor so you can review, correct or delete them. Click Save to confirm — only then are the rules stored.

Scanned PDFs (images)​

Scans and OCR

If the PDF is a scan (image pages with no text layer), the application automatically falls back to optical character recognition (OCR). Recent versions — both Docker and the desktop installer — bundle the OCR engine (French + English), so it works without any extra installation.

That said, OCR is slower and a little less reliable than native text: for the best result, prefer a PDF whose text is selectable whenever you have the choice.

Limits and error cases​

  • Size: up to about 60 MB per file.
  • Unreadable or corrupted PDF: the import stops with an error message.
  • No usable text (empty PDF, or a scan the OCR cannot read anything from): "no rules found" message, nothing is added.
  • A chunk that fails: the problematic passage is skipped and the import continues with the rest; if all chunks fail, the import is cancelled.

Local (Ollama) vs cloud provider​

The result differs slightly depending on the configured AI provider:

  • Ollama (local): the AI only detects the section boundaries and keeps your text as is — fast, no rephrasing.
  • Cloud provider: the AI rewrites and formats the content of each section — cleaner text, but slightly reworded.

Settings​

The PDF import parameters — chunk size and the AI call timeout — are configurable in Configuration → Settings.

If an import fails with a timeout or returns truncated sections, try reducing the chunk size (or increasing the timeout — and the context window for the AI you are using).