Skip to content

Docs

What we translate, and what we do not

A GM about to hand over a decade of notes deserves a straight answer about language, not a feature list. Here is what Canonry promises, and where that promise stops.

The interface and the copilot speak your language

Chrome, buttons, empty states, error messages, dates and the numbers in your credits panel follow the language on your account - English or Italian today - wherever you're signed in, phone at the table included. Everything the copilot says to you follows the same preference: an Ask answer, a propagation plan's reason, an audit flag's rationale, all in your language, whatever language the canon underneath happens to be written in.

Your canon keeps its own language

This is the rule that actually matters, and it runs the other way. An entry's language is detected from what you already wrote in it, per entry, the moment it's saved - and you can override that if the detector gets it wrong. When the copilot drafts a paragraph that lands inside an entry, it writes in that entry's language, not yours: an Italian interface does not get to start writing Italian paragraphs into an English entry just because you're the one reading the screen. A propagation proposal can carry two languages at once, and both are correct - the drafted text is in the target entry's language, the reason it exists is in yours.

Names are never translated, by an import or by a proposal. "The Gilded Rat" stays "The Gilded Rat" in an Italian sentence, the way a person’s name would.

Retrieval crosses the gap

None of the above is worth much if search only works within one language. Ask an Italian question and it has to find the English entry that answers it - that is what makes the embedding model a multilingual choice rather than an incidental one, and it is why "chi gestisce il Ratto Dorato" and "who runs the Gilded Rat Tavern" can point at the same inn.

Nothing rewrites what you wrote

The copilot proposes; it does not rewrite your prose on its own initiative, in any language. A drafted paragraph sits in your inbox like any other proposal until you accept it - there is no background job translating your world while you're not looking.

What this does not include

Stated plainly, because the limits are the part worth trusting:

  • Two languages today. English and Italian, nothing else, at launch - a third is not on this page because it has not been decided, not because it would be hard.
  • No bulk translation of an existing world. There is no button that rewrites everything you have already written into another language. Your prose is yours, and a mass rewrite is a background job Canonry does not run - a translated paragraph is a proposal like any other, one entry at a time, and it still needs your accept.
  • Quotations stay in their original language, always. The evidence behind a proposal is quoted verbatim from the entry it came from - translate it and it is no longer the sentence actually in your canon, and you cannot check it against the text. A translation can sit beside a quotation, clearly marked as ours, but never in place of it.
  • When the copilot talks to you, it follows your interface language, not the entry's. Ask your Italian account about an English entry and the answer comes back in Italian - that is the opposite of the rule above, on purpose: one rule governs what gets written into your canon, this one governs what gets said to you about it.