Hello

Your whole product, translated automatically.

See how it works Browse the integrations

Your product, in every language you sell in. Edited on the page itself.

A real checkout, running below in the language you pick. Change any word on it and the fix is live, reviewed and versioned, without a release.

Start for free

the overlay is off. Turn it on and the strings become clickable

northbound.shop/checkout?tg_locale=de-DEyour app, running

northboundBrauchst du Hilfe?

Bestellung prüfen

2 items · arrives TuesdayKostenlose Rücksendung

Alta shell jacketgraphite · M€98,00

Merino base layersand · M€20,00

product names come from the catalog, not the string file. The overlay leaves them alone.

Zwischensumme€118,00

Expressversand€10,40

Heute fällig€128,40

Cancel within 30 days, no questions.By paying you accept our terms.

overlay off4/10 approved in de-DE

in the room4 live

Yyouidle

MMiraediting line.total

TThéomoving around the page

RRinlength check

saves · de-DEnothing yet

nothing saved yet_
save a string and it lands here, and in the project as a new version

overlay offalt+click opens a stringesc closesdrag ⣿ to move the panelarabic flips the whole surface, not just the text

Running in your browser. Nothing is sent anywhere, and the teammates on the surface are an illustration.

We make it easy to sell to customerswho do not speak your language.

Connect a repository or a CMS. Transglot finds the strings, drafts every locale you have configured in parallel, and writes them back as a pull request or straight into your CMS.

Every string, every language, who changed what and when. Four roles, per project and per language, on every plan including Free.

Placeholders, plurals, glossary and length are checked on every translation. A row that fails is repaired once, and if it still fails it is refused rather than shipped.

Load the terms that must stay put and the ones that must never appear. They are locked into the prompt, and then checked again afterwards.

A missing translation fails the pull request rather than the launch. There is a CLI, a GitHub Action, a REST API, signed webhooks and an MCP server your own agents can call.

Translations publish to an edge bundle your app pulls over the air. A correction is live inside the minute, with no app store review and no redeploy.

The editor after a first pass: every row drafted into French, the machine-written ones badged AI.

The translation grid: keys down the side, locales across the top, each cell showing its state.

The QA screen: languages down the side, checks across the top, one failing cell opened onto the strings behind it.

The glossary: terms with their per-locale targets and their do-not-translate flags.

transglot/locales · required checkpull request #482

$ transglot push --ci

diff: 6 strings changed, 3 locales mapped

drafted 18 rows, Glossa Flux

gates: placeholders, plurals, glossary, length

de-DE length 67 > 48 budget

repaired once, re-gated 41 pass

check transglot/locales passed

committed to the branch, bundle queued for merge

transglot dispatch · ja-JPacme/checkout

$ transglot publish --locale ja-JP

one string changed since v41

bundle v42 built, 12.4 kB

immutable url minted, cached for a year

latest url revalidates within the minute

v41 still pinnable, 20 versions kept

no redeploy, no app store review

Two AI models. One fast, one careful.

Glossa Flux drafts every language the moment your English changes. Glossa Deep takes the slow route on the strings your brand is judged on.

Two AI models.
What happens	Glossa Fluxdrafts everything, always on	Glossa Deepthe careful pass, where it counts	Agencies and TMSthe old way
What it is for	The whole locale set, drafted the moment the English changes.	The strings a customer judges you on: the pricing page, the checkout, the error a payment throws.	Whatever you remembered to put in the file you sent.
Why it is fast	Every configured locale runs at the same time, not one after another, and a sentence you have already approved is reused rather than re-written.	Slower on purpose: several steps per string, and it can look at a screenshot of where the string renders.	One round trip per release, per language, in business hours.
What it is told first	Your glossary, your do-not-translate list and the plural rules for the language, before it writes anything.	All of that, plus the surrounding screen, the key name and the length the interface can hold.	A spreadsheet, and a style guide nobody opened.
On a failure	Repaired once, re-checked, and refused if it still fails. A broken row does not ship.	Escalated to a human with the rendered screen attached, so the reviewer can see the problem.	Found by your users.
What it costs	Included on every plan, Free included. Unchanged strings cost nothing to re-run.	Included from the Team plan. No per-word surcharge for using it.	Quoted per word, per language, per round.

Glossa Flux

drafts everything, always on

What it is for: The whole locale set, drafted the moment the English changes.
Why it is fast: Every configured locale runs at the same time, not one after another, and a sentence you have already approved is reused rather than re-written.
What it is told first: Your glossary, your do-not-translate list and the plural rules for the language, before it writes anything.
On a failure: Repaired once, re-checked, and refused if it still fails. A broken row does not ship.
What it costs: Included on every plan, Free included. Unchanged strings cost nothing to re-run.

Both engines run the same four checks on every row they write: placeholder integrity, plural completeness, glossary terms and length. How the engine decides which is which

The seven things that decide it.

Every one of these is a mechanism you can check rather than a promise. The figures come from the code, and the pages behind each tile show it working.

quality

four gates

It refuses.

Everyone else flags a bad translation and ships it anyway. Placeholders, plurals, glossary and length are checked before a model is asked for an opinion. A row that fails is repaired once, re-checked, and then held rather than published.

gate placeholders pass

gate plurals pass

gate glossary pass

gate length 67 > 48 budget

repaired once, re-gated

still over: held for review

translation memory

$0

A sentence you settled costs nothing again.

An exact match is served from your memory instead of being re-written, and it is never metered.

file formats

20

In and back out, intact.

JSON, XLIFF, PO, Android XML, iOS strings and stringsdict, ARB, YAML and the rest, with plurals and placeholders held both ways.

integrations

41

32 of them two way.

Your CMS, your repository, your help desk, your design tool. Every connector is included rather than sold as an add-on.

plurals

224 rule sets

Six forms in Arabic, one in Japanese.

Every CLDR plural category for the language, held as a set. A missing form is a failed gate, not a shipped bug.

zero
one
two
few
many
other

mcp server

16

Tools your agents can call.

Read a project, add a language, translate what is missing, run the checks. An agent cannot ship a broken string either.

in context

hold alt, click anything

Fix the wording on the running page, not in a spreadsheet.

The overlay draws in a shadow root so your CSS is untouched, the key in the page cannot write, and every change carries the name of the person who made it.

Read it before you buy it. Nothing is gated.

No form, no email wall, no call. The reference pages and the calculator are live now; the two long-form titles say plainly that they are still being written.

The Boring Localization Book
long formbeing written
The Boring Localization Book
The argument in full: why language should ship on the same rails as code, and what changes in a team when it does.
read the outlinein preparation
A First Multi-Locale Launch
long formbeing written
A First Multi-Locale Launch
Sequencing, right-to-left, pricing pages and support load, in the order a first launch actually meets them.
read the outlinein preparation
The Locale Reference
reference708 ICU locales
The Locale Reference
Every locale the validator accepts, how a tag is canonicalised, and which languages carry their own plural rule set.
open the referencelive page
The Cost Model
modelyour numbers
The Cost Model
The calculator with its assumptions visible and editable, so you can disagree with ours and still get a ledger out.
open the calculatorlive page
The Lexicon
specimenone page
The Lexicon
The whole product set as a dictionary entry: senses, a conjugation table, specimen plates and attested figures.
read the entrylive page
The Dispatch
dispatchpublished as written
The Dispatch
Shorter than a book: pieces on i18n setup, glossary enforcement, QA in CI, and migrating off another tool.
read the dispatchlive page

FAQs

What people ask before they connect anything.

What is Transglot?

Software that translates your app or website into every language you sell in, and keeps it translated. It finds the strings in your code or your CMS, writes them in each language, checks every one against four automatic checks, and puts them back where they came from as a pull request, a write back to your CMS, or a bundle your app fetches.

How does the AI translation actually work?

Every string goes through Glossa Flux the moment the English changes: all your languages at once, with your glossary, your past approved translations and the screen the string renders on already in hand. The strings your brand lives on can go through Glossa Deep, the careful pass. Both end at the same four checks.

What stops it from shipping a bad translation?

Four checks run on every row before a person sees it: placeholder integrity, plural completeness, glossary terms and length against the space the text has. A row that fails is repaired once and re-checked. If it fails again it is held rather than published, and a required check on your pull request stays red.

Which languages does Transglot support?

There is no fixed list. The validator accepts any BCP-47 language tag that ICU recognises, which is 708 locales in the current ICU table, and plural handling comes from the 224 CLDR plural languages rather than from a guess. Right-to-left scripts are handled, including the six plural forms Arabic needs.

Which file formats does it support?

Twenty, and they round-trip in both directions: JSON flat and nested, Android XML, iOS .strings, .stringsdict and .xcstrings, Flutter ARB, Laravel PHP and JSON, XLIFF 1.2 and 2.0, gettext .po, YAML, CSV, XLSX, Java properties, .NET resx, Markdown, Unity CSV and Unreal PO. Seven of them carry plurals, and plurals and placeholders survive the trip.

Is there a free plan, and what is in it?

100,000 words to start and 10,000 a month after, one project, 15 language rows including your source so 14 targets, three seats, and every role and permission. It is a hard cap rather than a meter, so it cannot produce a bill. No credit card.

How much does it cost after that?

Starter is $19 a month for 300,000 words, Team $49 for 1,000,000, Business $149 for 4,000,000. Past a paid plan’s allowance it meters at $0.0008 a word, which is $80 for another hundred thousand. Annual is ten months for twelve.

Do you train AI models on my content?

No. Your strings are sent to the subprocessors listed in our Trust Center to be translated and nothing else, and a translation memory hit is answered without calling any of them. Exact matches are held in a shared pool that records no person, no team and no workspace, and any project can leave it with one switch.

Is Transglot open source? Can I self-host it?

No. The platform is hosted by us. Our command line tool and our GitHub Action are public on GitHub, but the platform itself is not open source and there is no self-host path today. Public open-source projects can apply for a free Team-level grant instead.

Do I need a developer to set it up?

To connect a repository, once, for about as long as it takes to add a config file. After that a marketer or a translator works in the browser: they can edit the words on your running site by holding ALT and clicking them, with no ticket and no deploy.