transglot for game studiosunity, unreal, gettext

Ship the patch in every language, without breaking the UI.

Unity Localization CSV and Unreal PO are read and written directly, with no converter in the middle. The glossary holds your character names and your lore, and a string that would overrun a button is refused before it reaches the build.

Start for free See the language coverage

Without

With

Character names and item lore drift between builds, and between vendors.

One glossary, enforced when the engine writes and again at review, with up to 40 terms folded into every chunk.

German and Russian overrun a button that was measured in English.

The length check holds every row to the maximum length you set on the key, and flags an overrun for review.

A CSV round trip through a spreadsheet, and the plurals do not come back.

Unity Localization CSV and Unreal PO are read and written natively, and gettext PO carries plurals intact.

how it works

Your string table goes in. It comes back fitting the buttons.

five steps: Unity CSV or Unreal PO, the CLI on the patch branch, the translation pass, the length check, the table back in the build

solutions · route consolesolutions / gamesGame studios

what goes inUnity Localization CSV, Unreal POnative codecs, no converter step, keys and comments intact

how it gets thereThe CLI on every patch branchor the GitHub Action, from Starter

which passDeep on names and loreFlux for the rest of the string tableFluxDeepboth passes, chosen per key

the checkThe length gateGerman and Russian overrun a button measured in English; refused before the buildplaceholderspluralsglossarylengththe check that catches it here

what comes outThe string table, back in the buildwritten natively; one CSV or XLSX upload takes 200,000 rows

an illustrative first weekGame studios

build · acme/rpg · patch branch

format · unity localization csv

glossary · names and lore · held

length check · 2 refused · 1 repaired

placeholders · matched on every row

string table written · patch ready

the plan that fits

Starter$19a month

300,000 words a month
Unlimited projects and languages, three seats
The scoped token Free cannot mint: CLI, Action, API, MCP

The length check is one of four deterministic gates, and they cost nothing, on every plan. The CLI on the patch branch starts here.

See all eight pages

every one of them starts on the same free plan

A fixture, in your browser. The five steps, the two passes, the four checks and the plan rows are the shipped mechanism and the shipped price list; the first-week tape is illustrative and nothing on this plate is sent anywhere.

A converter and a spreadsheet, versus reading the file directly.

A converter and a spreadsheet, versus
What is being compared	Transglotnative codecs	The usual pipelineexport, edit, pray
Engine formats	Unity Localization CSV and Unreal PO are two of the 20 native formats. Nothing converts in the middle.	A CSV export, a macro, and a merge nobody wants to review.
Names and lore	One glossary, enforced by the terminology check when the engine writes and again on review.	A style guide attached to the brief.
Text that overruns the UI	A row that overruns is held for review, and vision grounding reads the rendered screen where it is enabled.	Found in QA, on a console, on a Friday.
Plurals	224 CLDR plural languages, so a language missing a form is refused rather than shipped.	One string, with an s bolted on the end.
Patch cadence	Every patch branch drafts itself from Starter, through the CLI or the GitHub Action.	A translation drop, once a milestone.

Transglot

native codecs

Engine formats: Unity Localization CSV and Unreal PO are two of the 20 native formats. Nothing converts in the middle.
Names and lore: One glossary, enforced by the terminology check when the engine writes and again on review.
Text that overruns the UI: A row that overruns is held for review, and vision grounding reads the rendered screen where it is enabled.
Plurals: 224 CLDR plural languages, so a language missing a form is refused rather than shipped.
Patch cadence: Every patch branch drafts itself from Starter, through the CLI or the GitHub Action.

Vision grounding is a Team feature and up, switched on per project, and it spends no word allowance. The four deterministic checks cost nothing and run on every plan.

what breaks, and what catches it

Four ways a translation breaks a build.

These four checks run on every row, on every plan, on both passes. They are deterministic, which is why the answer is always the same one.

a lost token

Placeholders

A missing or renamed placeholder is refused rather than guessed at. Braces, positional tokens and printf shapes are all checked.

deterministic

a missing form

Plurals

Every plural category the target language declares has to be present. Arabic declares six, and six is what the check expects.

224 plural languages

a renamed hero

Glossary

Names, items and legal terms are held. Up to 40 terms are folded into each chunk, and the check runs again at review.

terminology

an overrun button

Length

The maximum length you set on the key is enforced on every row. A row that overruns is written and held for review rather than shipped.

held for review

The four checks are deterministic. That is the whole reason a broken row does not reach a build, and they run on every plan.

the QA runner pre-gate

FAQs

What a studio asks before the next patch.

How do I translate a Unity or Unreal game?

Point Transglot at the string table you already export. Unity Localization CSV and Unreal Engine PO are read and written directly, so nothing converts in the middle, and the table comes back in the same shape with the translations in it. From the Starter plan the CLI runs on your patch branch and the GitHub Action does it on every push.

Which engine formats are native?

Unity Localization CSV and Unreal Engine PO, alongside gettext PO, CSV, XLSX and both JSON shapes, out of 20 formats in all.

How do I stop character names and lore drifting between builds?

Put them in the glossary. Up to 40 relevant terms are folded into each chunk the engine sees, the terminology check enforces them when the engine writes, and it runs again in the QA review pass. A name that has been agreed once stops being renegotiated by whoever is translating that week.

How do I stop German and Russian breaking my UI?

Set a maximum length on the keys that sit in fixed furniture. Every row is measured against it, and a row that overruns is written and held for review rather than shipped. Where vision grounding is switched on, which is a Team feature and up, the screenshot on the key goes to the engine with the string so it can see the space it has to fit, and that pass spends none of your word allowance.

Do plurals survive a spreadsheet round trip?

They do here, which is the point. CSV and XLSX are two of the seven formats that carry plurals, and every plural category the target language declares has to be present or the row is refused. Arabic declares six forms and is held to six.

Can translators see the screen the line appears on?

In-context review renders the screen the string appears on, so a line is judged in its own frame rather than in a column. A session lasts 15 minutes and the screenshots behind it are kept for 90 days. In-context is available from the Starter plan.

How large can a string table be?

One CSV or XLSX upload takes up to 200,000 data rows, which is the shipped ceiling rather than a soft recommendation. There is no cap on the number of keys a project can hold.

Can I translate one patch instead of doing a localization drop?

Yes, and that is the intended shape. Every patch branch drafts itself through the CLI or the GitHub Action, so the strings that changed are translated with the change rather than collected for a milestone.

What does it cost to localize a game?

Count the words in your string table, multiply by the number of languages, and read that against the plan allowances: 100,000 words to start and 10,000 a month after on Free, 300,000 on Starter at $19, 1,000,000 on Team at $49 and 4,000,000 on Business at $149. Words past the allowance are $0.0008 each, about $80 per 100,000. The first pass on a large back catalogue is usually one big month; patches after that are small.

Do you handle right-to-left languages?

Yes, the same way as any other. Direction comes from the language tag rather than from a special mode, and the locale table records it. The plural check is what actually does the work for Arabic, holding it to all six of its declared forms.

Do you train AI on our script?

No. Your strings reach our AI subprocessors for inference only, under commercial terms, and the trust page names every one of them. The shared exact-match memory pool is anonymised and can be switched off per project, which is worth doing for an unannounced title.

Can we keep our existing translators?

Yes. They work in the same workspace with their own accounts, and access can be scoped to one project and one language. The translator role exists on every plan, including Free.

works with what you already run