An AI Dungeon-style adventure in your own language: what to look for

Updated: October 14, 2026 13 min read

If you searched for AI Dungeon in your own language, you probably want something simple: an AI Game Master that tells the story, waits for you to say what you do, and answers in the language you actually think in. Every large language model can write fluent German, Turkish or Spanish in 2026, so the prose itself is rarely the problem. What breaks is everything around it: names, dice words, buttons, error messages, and a narration that slides back to English halfway through a scene. This page explains what goes wrong in English-first tools, what to check before you commit to a campaign, and exactly how Thavernia handles language, including where it still falls short.

Can you play it in your language? The short answer

Yes, but "it answers in my language" and "it was built in my language" are two different things. Many AI story apps only work properly in English: you can type in French or Japanese and you will usually get French or Japanese back, but that is the language model's doing, not the product's. The interface, the billing screen, the error messages, the scenario library and the community stay English, and a long scene often drifts back to English on its own.

For a player who lives in English, none of that matters. It starts to matter the moment you want a table where everyone reads the narration comfortably, or you want the Game Master to sound like a storyteller from your own shelf rather than a manual that went through a translator. If you are comparing options more broadly, our page on AI Dungeon alternatives looks at the question from the rules and memory side; this page stays on language.

Writing a language vs. being built in it

You can't settle whether a platform "speaks a language" with a single question, because there are three tiers and all three put readable text in the chat box. Telling them apart means looking outside the box.

The three tiers

  • Tier 1: one language only. Interface and content are English. Write in another language and the model improvises; the product doesn't.
  • Tier 2: translated. The interface has been bulk-translated, and the narration is thought in English and then spoken in other words. The sentences are correct but the voice is imported, the tone of a manual, not a storyteller.
  • Tier 3: written natively. Narration is generated directly in the chosen language, interface and system strings were written in it, errors and support come in it too. Very few tools sit here.

Five checks tell you which tier you're looking at, and none takes a minute. First, break something on purpose: trigger an invalid action and read the language of the error, the translation layer collapses there before anywhere else. Second, open the billing or account screen; marketing pages get localised, invoices almost never do. Third, throw an idiom into the narration and watch not whether the model understands yours, but whether it ever coins one of its own unprompted. Fourth, check voice consistency: if the Game Master slips between present and past tense, or between second and third person, every few turns, there is no style plan behind it. Fifth, look for strings that don't fit their boxes, clipped buttons, "1 items", a date rendered 08/20/2026 on a page whose readers write 20/08/2026. Those small seams tell you the interface was fitted afterwards.

These five checks measure the product, not the marketing site. Don't conclude a tool is localised because its blog is; blogs get translated, error messages don't.

What goes wrong when a tool is English-first

The failures are small one by one, but a role-playing game is hours of reading, and small seams add up to a table that never quite feels like yours. These are the four that show up most.

The translated feel

The sentences are grammatical, yet nobody in your language would write them. English word order survives, English images survive ("a chill runs down your spine" rendered word for word), and the register swings between a novel and a user manual. In languages with formal and informal address, the Game Master may call you "tu" one turn and "vous" the next. None of this is wrong enough to report, and all of it pulls you out of the scene.

Names

English-first tools hand you characters and towns with English names, then narrate them in another language, so a German scene is full of Bobs and Brightwaters. The reverse also happens: a hero named in your language can make the model switch languages, because it treats the name as a hint. And in languages that inflect names, a Turkish suffix after an apostrophe or a German genitive, the model has to know the name is a name and bend it correctly every time.

Dice and rules words

Even when the story is in your language, the game layer often isn't. "Roll a Dexterity check", "saving throw", "advantage", "critical hit" stay in English on the buttons, or get translated so literally that a player who knows the game in their own language doesn't recognise them. The rules vocabulary is the part a player reads most often, so it is the part that most needs to be right.

Mixed languages

The worst case is a scene that starts in Italian and is half English by turn twenty: suggestion chips in English, one NPC who suddenly answers in English, a quest log that was never translated. It happens because the instructions the model receives are written in English, and over a long context English is the language it falls back to when nothing pins it down.

What to look for before you start a campaign

  • The first turn is in your language. The opening is where drift is most likely, because there is no earlier text to anchor the language yet.
  • Dice requests and results in your language. The roll button, the ability name and the result label, not just the prose around them.
  • Characters and places that belong to your language. Ready heroes and starting adventures written for it, not English ones passed through a translator.
  • One language per campaign, held to the end. NPC dialogue included, twenty turns in.
  • Interface and narration chosen separately. Useful when your group is mixed: each player reads menus in their own language while the story runs in the one the table agreed on.
  • A voice that matches. If the narration is read aloud, the narrator should be a native speaker of that language, not an English voice reading foreign words.

Why playing in your mother tongue is worth this effort at all is its own topic, and we wrote it out separately: playing tabletop RPGs in your own language.

Telling a chat assistant "be my Game Master" in your language

This is the real default for most people, and on the language axis it does well: a general chat assistant writes fluent prose in most major languages, there is nothing to install and the first hour is genuinely enchanting. What breaks afterwards isn't the language, it is the game. Memory keeps your preferences (that you like a harsh Game Master, that you want long description), not your campaign ledger. Your hit points, your gold, the number of potions in your pack and which step of the quest you're on are numbers written inside a conversation. As the chat grows, the past gets quietly summarised, and any number outside the summary is gone. On turn thirty you say "I had two potions", and the model says "of course, you have two potions", because agreeing with you costs it less than staying consistent.

The same structure applies to dice: whoever says you rolled 17 is also whoever decides what 17 means. There are ways to prop this setup up, and we wrote them out separately: playing D&D with a chat assistant.

How Thavernia handles language

  • Six narration languages: English, Turkish, German, Italian, French, Spanish. The narration is generated directly in the campaign's language, not written in English and translated.
  • Interface localised separately: menus, system strings, dice buttons, ability names and error messages exist in the same six languages, and the interface language is chosen independently of the campaign language.
  • A language lock on every turn: the instruction to write the whole narration and every line of NPC dialogue in the campaign language sits at the very end of the Game Master's instructions, where it carries the most weight, and says plainly that names which look like another language never change the language of the text.
  • Ready heroes and premises per language: the quick start's heroes and opening adventures are written separately for each language, name, origin, personality and appearance included. The Turkish fighter is Demirhan, the English, German, Italian, French and Spanish one is Doran; the caravan adventure is "Kayıp Kervanın Sırrı" in Turkish and "The Lost Caravan" in English.
  • A narrator voice per language: when you have the narration read aloud, the voice is picked from the table's language, a German table gets a German narrator, a Spanish table a Spanish one.
  • Rules that hold in any language: the server produces the dice number, the player throws the d20, the model only interprets. Hit points, inventory, gold, quest step, NPCs and places live in database rows, not in the prose.
  • Multiplayer: friends join one table by invite link and every turn streams to everyone live; each player keeps their own interface language.
  • Free trial: three turns with no account and no card, /trial.

Here's the mechanism rather than the sales pitch. First, language: a campaign's narration language is chosen when you create it and is independent of the interface language; the text is generated in that language rather than translated into it. Second, dice: the server produces the number with random_int(1,20) before the model sees any text, the 3D d20 displays that number, the player throws it, and the model only turns the result into story. Third, the ledger: hit points, inventory, gold, quest step and NPCs sit in tables, and after every turn a background pass re-applies any world-state lines the model dropped. Fourth, refereeing: a boss is split into three phases, a single blow is capped at a share of its maximum health, and a boss death the numbers don't support is refused and the turn rewritten. And one thing we don't soften: friendly fire is real, and your character can die.

The language test: run the same scene yourself

Comparison claims are easy to write, so here is a standard opening you can paste into any tool you're considering, in whatever language you want to play. It's deliberately idiom-free but concrete: mud, a torch, a door, and a hesitation. Translate it into your language first, then start the scene with it.

Standard test opening
I'm following the muddy bootprints leading out of the inn's back door. I'm holding a torch, the door is ajar, and I can hear wet breathing from inside. Before I go in, I want to work out how many people left those tracks.

Be my Game Master: narrate the scene in English, don't make decisions for me, and ask me for a roll if one is needed.

What to watch for: whether the text holds one language and one register from start to finish; whether system strings (buttons, warnings, suggestion chips) are in the same language as the narration around them; whether the prose builds its own imagery or recycles translated stock phrases; whether the mode of address stays consistent; whether it decides for you by writing "you step inside and see three goblins"; and finally, whether it actually asks for a roll, or quietly rolls inside the narration and describes the result in the same breath. That last one is the subject of the next section.

The dice question: can an AI award itself a success?

When one language model is both narrator and referee, you get a structural conflict of interest. Models lean toward pleasing the user and toward building a coherent story; when those two pull apart, the story usually wins. And because the number is produced inside the text, "you rolled 14" and "you force the door open with difficulty" are written at the same moment, in the same place, each seeing the other. The model isn't trying to cheat; it's picking the number that fits the scene. The result is the same either way: tension leaks out, because losing was never really on the table.

The fix isn't a better model, it's a different architecture. On one side everything happens inside a single language model, narration, dice, inventory, refereeing. On the other, the work is split: the server produces the number, a database keeps the record, and the model only narrates. In Thavernia the die is rolled before the model sees any of the turn's text; the player throws the 3D d20 and watches that number land; the model can't change it, only interpret it. The same logic applies to world state: your hit points don't live in the prose, they live in a row, so they don't vanish when the model forgets them.

This architecture doesn't solve everything, and claiming otherwise would be a lie. The model still drops metadata lines sometimes, and narration and record still drift apart occasionally; that's what the post-turn correction pass and the boss-death check exist for. We wrote out exactly where the refereeing is solid and where it gives on a separate page: is the AI Game Master fair?

Where we are still limited

This section isn't a footnote. Knowing these before you sign up is in your interest.

  • Six languages, not sixty. If your language isn't English, Turkish, German, Italian, French or Spanish, the narration can't be set to it.
  • One narration language per campaign. A mixed group can each read the interface in their own language, but the story itself is told in one language the table agrees on.
  • A fantasy table. Thavernia runs fantasy adventures with rules; if you want to jump genres mid-sentence with nothing refereeing you, a free-form story tool will suit you better.
  • Our prose isn't flawless. Long scenes can fall into repeated sentence shapes, and the narration occasionally over-decorates, in every language.
  • The referee bends in one known place. If the model keeps insisting across turns that a boss already below half health has died, the server eventually concedes so that narration and record don't contradict each other. A fresh boss doesn't fall to insistence.

Playing a turn without an account

Here's the friction in numbers: no account, no card, three turns. Three turns in total, not per day. What happens, in order: you pick a ready hero, the Game Master narrates the scene in your chosen language, you write what you do, at some point it asks you for a roll, and you throw the d20 yourself, the number you see on screen is the number the server generated before you threw it. The model can't change it; it can only tell you what it means. When the three turns are up your character pauses; sign up and you can claim it and carry on, with 20 Game Master turns on a free account before a plan is needed.

If you'd rather read how the Game Master side works first, the AI game master page walks through the mechanism step by step. If you'd rather just play: a three-turn trial.

About this page

Updated 14 October 2026: the page was rewritten to focus on playing in your own language. The language details (six narration languages, the language lock, ready heroes and premises per language, the narrator voice per language) were checked against Thavernia's current code on that date.

For transparency: this page was written by Thavernia, so we are an interested party. Instead of hiding the bias we wrote the checks out in the open, so you can run them on any tool including ours, and stated our own limits (six languages, one narration language per campaign, the give on boss insistence, the repetition in long scenes) without softening them. If you spot an error we'll correct it, and when the page is updated the date in this section changes with it.

Frequently asked questions

I came here searching for AI Dungeon in my language. Is that what this page covers?

It covers the experience behind that search: a free-form AI text adventure with a Game Master that narrates, asks you what you do and reacts. This page doesn't review other products. It explains what to check in any tool before you play in a language other than English, and how Thavernia handles it.

Why does the narration in many AI story apps feel translated?

Because the product was designed in English and other languages were added afterwards. The model can write the language, but the instructions behind it, the example sentences, the buttons and the rules vocabulary were all English, so the prose tends to follow English sentence shapes, keeps English game terms and drifts back to English as a scene grows.

Can an AI game master make up its own dice results?

In any setup where the model is both narrator and referee, yes it can. Because the number is produced inside the text, nothing stops it from picking whatever fits the story. The way to close that is to move the number outside the model: the server rolls before the model sees any text, and the model only interprets the result it's given.

Does the campaign language have to match the interface language?

Not in Thavernia. The narration language is chosen when the campaign is created and is independent of the interface language; the prose is generated directly in that language rather than translated from English. So you can run the interface in English and read narration in Spanish, or the reverse. Supported narration languages: English, Turkish, German, Italian, French, Spanish.

Do the ready-made heroes and adventures change with the language?

Yes. The quick start offers ready heroes and opening premises, and each one is written separately in all six languages: name, origin, personality, goal, flaw and appearance. The Turkish fighter is Demirhan, the English one is Doran, and the opening adventure has its own title in each language. Nothing is sent to the Game Master in the wrong language.

Do I need an account to try it free?

No. At /trial you can play three turns with no account and no card. Those three turns are the total, not a daily allowance: you pick a character, the Game Master narrates in your chosen language, at some point it asks for a roll and you throw the d20 yourself. If you like it, you can claim that character when you sign up, and a free account then gets 20 Game Master turns.