transglot for localization managersstates, checks, reviewers

Every string, every language, and what state it is in.

Who is waiting on what, which rows failed which check, and what the reviewer changed. It is all in the workspace, written by the run itself, so there is no spreadsheet to keep up to date beside the product.

Start for free See the platform

Without

With

Progress lives in a spreadsheet that is out of date the moment you send it.

What is ready, by language and by key, in the workspace, with the four checks recorded on every row.

Terminology drifts, because the glossary is a document nobody opens.

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

Review happens without context, so length and layout break after sign-off.

In-context review shows the rendered screen, so length and layout are judged against it rather than guessed at.

how it works

Every string, and what state it is in. The board writes itself.

five steps: every string in every language, the workspace board, a draft and then a review pass, the glossary check, a reviewer notified in context

solutions · route consolesolutions / managersLocalization managers

what goes inEvery string, in every statewho is waiting on what, and which rows failed which check

how it gets thereThe workspace boardreadiness by locale and by key, written by the run rather than by a person

which passFlux to draft, then a review passan AI review across 5 MQM dimensions before a person sees the rowFluxDeepboth passes, chosen per key

the checkThe glossary gateenforced at translation time and again at review, up to 40 terms a chunkplaceholderspluralsglossarylengththe check that catches it here

what comes outA state change, a reviewer notifiedin context against the rendered screen, screenshots kept 90 days

an illustrative first weekLocalization managers

readiness · acme/web

de-DE · ready

fr-FR · in review · reviewer notified

ja-JP · length check · holding 2 rows

ar-EG · six plural forms · complete

glossary · enforced again at review

the plan that fits

Team$49a month

1,000,000 words a month
Unlimited seats and unlimited projects
All 41 integrations: 36 from Starter, 5 from Team

Three seats on Free and Starter, unlimited from Team. The four roles, owner, admin, developer and translator, are on every plan.

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 spreadsheet and a mailbox, versus a workflow.

A spreadsheet and a mailbox, versus
What is being compared	Transglotone board	The way it is done nowin attachments
Status	Per language and per key, in the workspace, written by the run itself rather than by a person.	A spreadsheet that is stale the moment it is sent.
Terminology	One glossary, enforced by the terminology check when the engine writes and again at review.	A document attached to the brief, and then not read.
Quality signal	Four deterministic checks on every row, plus an AI review across 5 MQM dimensions and three severities.	Whoever complained loudest, last.
Review	In-context editing against the rendered screen, from Starter, with a 15 minute session and screenshots kept for 90 days.	A diff, and a guess about where the string appears.
Reuse	Exact memory hits skip the model entirely, so an unchanged row costs nothing to keep.	The same sentence, paid for again.

Transglot

one board

Status: Per language and per key, in the workspace, written by the run itself rather than by a person.
Terminology: One glossary, enforced by the terminology check when the engine writes and again at review.
Quality signal: Four deterministic checks on every row, plus an AI review across 5 MQM dimensions and three severities.
Review: In-context editing against the rendered screen, from Starter, with a 15 minute session and screenshots kept for 90 days.
Reuse: Exact memory hits skip the model entirely, so an unchanged row costs nothing to keep.

The checks, the MQM dimensions and the retention windows above are configuration values in the shipped product, not adjectives.

the numbers

Four numbers, and where each one comes from.

Each figure names the file it comes from. A number without a source line does not belong on this site.

deterministic checks on every row: placeholders, plurals, glossary and length. Source: the QA runner pre-gate.

MQM dimensions in the AI review pass: accuracy, fluency, terminology, style and locale convention. Source: the MQM typology.

224

CLDR plural languages, so a language that declares six forms is held to six. Source: the generated plural category table.

days of in-context screenshot retention, on 15 minute review sessions. Source: the in-context configuration.

FAQs

What a localization manager and a reviewer both ask on day one.

How do I know which languages are ready to ship?

The workspace shows readiness by language and by key, written by the run rather than typed by a person. Every row carries its state, which of the four checks it passed, who changed it last and when. There is nothing to export and nothing to circulate.

Can an automatic translation overwrite a human edit?

No. A reviewed row is protected by a write guard that holds machine writes off it, and a human edit is protected against a concurrent human edit by a compare-and-set check. They are separate guards because they are separate races.

How many people can review, and what does a reviewer cost?

Three seats on Free and on Starter, unlimited from Team at $49 a month. Seats are not billed per head on any plan. The four organisation roles, owner, admin, developer and translator, exist on every plan including Free.

Can I give an outside translator access to one language only?

Yes. Grants are per project and per language, so a freelancer can be given French and see nothing else. They work in the same workspace as everyone else, with their own account.

How do I keep terminology consistent across languages?

One glossary per project. Terms are enforced by the terminology check when the engine writes, and again in the QA review pass, which is a Team feature and up. Up to 40 relevant terms are folded into each chunk the engine sees, so the term is in front of it rather than in a document beside it. The glossary is available from the Starter plan.

What happens to a row that fails a check?

It gets one automatic corrective retry. If it still fails it is refused rather than shipped, and the refusal is recorded against the run so you can see what was held and why.

How do you measure translation quality?

In two layers. Four deterministic checks run on every row: placeholders, plurals, glossary and length, and they give the same answer every time. Above them an AI review pass, a Team feature and up, scores the row across 5 MQM dimensions, accuracy, fluency, terminology, style and locale convention, at three severities. The deterministic checks cost nothing; the AI review pass meters the words it reads.

Can a reviewer see the screen the string appears on?

Yes, where in-context review is switched on: it renders the screen the string sits in, so length and layout are judged against the real thing rather than guessed at. A session lasts 15 minutes, and the screenshots behind it are kept for 90 days and then deleted. In-context is available from the Starter plan.

Can I undo a bad translation?

Every version is kept for 180 days, with a floor of the twenty most recent per string, on every plan including Free, and each one records who wrote it. Reverting restores it in place.

What happens when the English source changes after a translation was approved?

The change is caught at the moment the source is written, not by a nightly scan, and the affected translations are queued for a bounded repair pass. Repeated edits to the same key inside a short window collapse into one job so a churning file does not run up a bill. This is a Team feature and up, and it emits a source_drift.repaired webhook event.

Can I import from Lokalise, Phrase, Crowdin, Transifex, Weblate and POEditor?

Yes. All 6 have migration importers. Anything else comes in as one of the 20 file formats, read natively, and a single CSV or XLSX upload takes up to 200,000 data rows.

Does translation memory reduce what I pay?

Exact matches do. A string whose exact translation is already in memory skips the engine entirely and consumes nothing from your allowance. Fuzzy matches are reported so you can plan a job, but they still go through the engine.

Do you train AI on our content?

No. Our AI subprocessors are called for inference only, under commercial terms, and the trust page names every one of them and where it runs. The shared exact-match memory pool is anonymised and can be switched off per project on every plan.

works with what you already run