Translate with AI
Have Glossa draft a language's missing translations, or only the rows you pick, and know what it will cover before it starts.
Glossa, transglot's translation engine, drafts translations for you. Glossa Flux is the default model; projects with Quality mode on use Glossa Deep instead. You can start a run yourself from the editor, and transglot also starts runs on its own when new source text arrives (see Automatic translation).
Starting a run needs the Developer role or above. Translators and Reviewers can edit and approve what Glossa writes, but cannot start a run.
Before you start
Set up the project's glossary and brand voice first if you plan to use them. Glossa follows both on every run, and a translation that is already written is not redone when you change them later.
Translate everything missing in a language
Open the language
In the editor, choose the target language.
Start the run
Select Translate missing. It covers every string in the language that has no translation, plus Glossa drafts whose source text has changed since they were written.
Check what it covers
A dialog asks Translate into language? and shows how many strings and words the run covers, and how that compares with the words left in your plan this cycle. Under Which languages, keep language only, or choose the option that lists every language with strings waiting.
Confirm
Select Translate. Cancel starts nothing.
Translate selected strings
Select the rows
In the editor's Table view, tick the strings you want translated. See Act on many strings at once.
Start the run
Select Translate in the bar above the table, check the figures in the dialog, and confirm.
What happens during a run
- Translate missing reads Translating…, and a progress card shows how many strings are done. Rows fill in as strings land, each in the draft state.
- You can keep working. Leaving the page does not stop the run.
- Stop ends the run early. Strings already translated are kept; the rest need starting again.
- When the run ends, the card says what happened: how many strings were translated, how many came from translation memory, and how many failed. Retry failed re-runs only the failed strings.
Every run, with what started it and which strings failed, is listed on the project's Translation runs page: select Runs in the project's navigation, or Translation runs on the project overview.
Only one run goes at a time in a project. Starting another while one is running is refused with "A translation run is already in progress on this project. Wait for it to finish, then start this one."
What Glossa does and does not touch
Your own work is safe. A translation you have edited or reviewed is never overwritten by AI. A run skips it even when its source text has changed, and the end-of-run summary counts strings "left as they are because someone had already edited or reviewed them".
Repeated text is reused. When the exact same source text has been translated before, transglot reuses that translation at no cost instead of translating it again. See Translation memory.
Placeholders, plurals and terms are checked. Glossary terms are matched against source text and enforced during translation and during quality checks; a translation that departs from a glossary term is flagged. Plural keys get every plural form the target language needs. See Plural forms.
Review is routed for you, if you want. Each AI translation carries a confidence signal. Translations that pass the automatic checks and come back confident can be marked as not needing review, so your reviewers only see the ones worth their time. Everything else is flagged for review. Automatic checks always win: a translation with a broken placeholder, the wrong plural forms, or a glossary conflict is always sent to review no matter how confident it was. This routing is off by default; turn it on with Route AI translations by confidence in the project's Settings, under Checks (Team and above).
Automatic translation
With Auto-translate on (the default), transglot starts a run without anyone pressing a button when:
- Source text changes. New or changed source strings that arrive by push, upload, API or an integration are translated into every target language. See AI translation settings.
- You add a language. Everything already in the project is translated into the new language. The console asks first, with the word count: Add language and translate into it? See Add a target language.
When the first strings arrive in a new project, the project's setup page asks whether to translate them now, with the word count, and offers Keep translating new text automatically. Clearing that box, or choosing Not now, turns Auto-translate off for the project; you can turn it back on in the project's settings.
Runs started automatically follow the same rules as runs you start: edited and approved translations are never overwritten, and memory is reused first.
Usage
Your plan includes an allowance of AI translation each month. Usage is measured against that allowance, and what happens when you exceed it depends on whether usage-based billing is turned on for your workspace. Text reused from translation memory does not count against it. The confirmation dialog tells you before a run starts whether it fits. See What counts towards my usage? and pricing.
If it doesn't work
- Translate missing is not shown: you need the Developer role or above, or the project has no source strings yet.
- "Nothing to translate. No string in language is waiting for this run.": every string in the language is translated and current, or has been edited by a person.
- "The figures for this run could not be loaded, so nothing was started. Try again in a moment.": nothing was started or charged; try again.
- A refusal that names your plan, payment or spend cap: AI translation is paused for the workspace's account. The message says why, and Plan & billing says what to do. See Usage-based billing.
- The run finished with failures: open the run on Translation runs to see which strings failed and why, then select Retry failed.