Translation Manager
AI drafts and what comes after
AI can write a first draft of every page on your site in minutes. Translation Manager uses AI for that too, and then handles the rest of the job. That means agencies, review, approval, reuse and keeping more than one site in step.
Many ways to translate
Translation Manager sends content to a connector and brings the result back into Umbraco. You pick the connector, so the same workflow works with AI, machine translation or people:
- The AI connector. Use OpenAI, Azure OpenAI, Anthropic Claude, Google Gemini, Ollama and other providers, set up in Translation Manager.
- The Umbraco.AI connector. If your site already has AI profiles set up in Umbraco.AI, this connector uses them, so you don't configure a provider twice.
- DeepL, Google Cloud Translation and Microsoft Translator. These are machine translation services with their own pricing and their own strengths for particular language pairs.
- XLIFF files. You export a job as XLIFF 1.2 or 2.0, send it to any agency or translation tool, and import the file that comes back.
- Agency and platform connectors, for services that take the content through their own API.
Each translation set has a default connector, and you can choose a different one when you create a job. So you can send marketing pages to an agency and product pages to DeepL from the same site. The full list is on the connectors page.
Working with an agency
Most agencies work in XLIFF. Translation Manager exports a job as an XLIFF file with the configuration data stripped out, so the translator sees only the text. When the file comes back you import it into the job, and the translations go into review like any other connector's.
If your agency has an API connector, the exchange happens without files.
Review, approval and notifications
A job moves through stages: created, submitted, received, reviewing, approved. The dashboard shows where each job is. When some content comes back before the rest, the job is marked partial, and you can approve what has arrived and check the rest back in later.
Approving and publishing are separate permissions. You can give a translator or an outside reviewer the right to approve translations, which saves them into the content, without the right to publish. Someone on your team publishes.
Each translation set can send email when content is waiting to be sent, when a job is submitted, when it comes back, and when it is approved. It can also email the person who created the job.
Large jobs can be approved in the background, so the editor doesn't wait on a browser request while hundreds of pages save and publish. If an item fails, background approval skips it, records the error in the job history and carries on with the rest. A translation set can also approve automatically, and either publish the result or save it as a draft for someone to check.
Translation memory
Translation Manager stores every approved translation. When the same text comes up again, in a later edit or on another page, the stored translation is used and the text isn't sent to the provider.
With an AI or machine translation provider you pay per character or per token, so text that comes out of memory costs nothing. On a site where most edits change a sentence or two, that is most of the text. You can view and edit the memory in the backoffice.
Translation Manager also tracks what changed in the source since the last translation, so a new job only includes the changed content.
Glossary
The glossary is a list of terms with the translation to use in each language, or a flag saying the term must not be translated. Product names are the usual example.
You can import and export it as CSV, one language pair at a time, or as TBX, the standard termbase format, for every language in one file. That means you can bring in a termbase from an agency or another tool and keep the two in step.
The AI connector adds the glossary to its prompt. The DeepL connector pushes it to DeepL as a DeepL glossary and only uploads it again when a term changes.
Sites in separate trees
Some multilingual Umbraco sites use language variants on one content tree. Others keep a separate site for each language, often with its own domain. Translation Manager handles both.
With separate trees, it creates and updates pages in the target site, and the link updater rewrites links in rich text and Multi URL Picker properties to point at the matching page on the target site. Without that, links on the French site would point back to the English pages.
Writing your own connector
The connector API is documented and public. If your translation service has an API, you can write a connector for it using the same base classes as our connectors, including translation memory and glossary support. See the connector documentation.
What a translation workflow gives you
- A choice of who translates. Send each job to AI, a machine translation service or an agency, and change that later without changing how editors work.
- Sign-off before anything goes live. Translations go into review, and approving and publishing can sit with different people.
- Everyone knows where things are. The dashboard shows every job's stage, and emails tell the right people when work is sent, comes back or is approved.
- You pay once for each piece of text. Translation memory reuses approved translations, and change tracking sends only what changed.
- The same words every time. The glossary keeps product names and key terms consistent across languages and providers, and can be shared with an agency as TBX.
- Sites kept in step. Separate-tree sites get new and updated pages, with links pointing at the right language.
- A record of what happened. Each job keeps its history, including anything that failed to approve and why.
You can try all of this before you buy. An unlicensed install translates between one language pair with no time limit.
Keeping content private
Content goes from your site straight to the service you configure, using your own account and keys. It does not pass through Jumoo. The data handling page has the detail for each connector.