How transglot works
You bring your source strings in once, transglot translates them with AI and your own past translations, your team reviews what needs a person, and the finished languages go back out to your product.
Every screen in transglot belongs to one of four stages: content in, translation, quality and delivery. They run in that order for each string, and they keep running as your source text changes.
Where your work lives
A workspace holds your team and your projects, and belongs to an account that holds the plan and billing. A project holds one set of keys (your source strings), the source language they are written in, and the target languages you ship. See Workspaces and projects and Keys and strings.
1. Content in
Strings reach a project in one of these ways:
- From your codebase, with the CLI, the GitHub Action or the API.
- From a connected repository, which extracts strings when you push. See Connect GitHub.
- From a content platform such as a CMS, a store or a help centre. See Connect a content platform.
- From a file you upload, or a project you migrate from another tool. See Import content.
The project records which language the strings are written in and which languages you translate into. See Languages and locales.
2. Translation
Glossa, the transglot translation engine, drafts the missing translations. It reads the context you give it: the project description, each key's description, your glossary, your brand voice and, when you turn it on, similar translations your team has already approved. When the exact same source text has been translated before, transglot reuses that translation at no cost instead of translating it again. See AI translation and Translation memory.
New and changed strings are translated as they arrive while the project's Auto-translate setting is on, or you start a run yourself from the editor. Everything lands in the editor, where your team can change any translation. A translation you have edited or reviewed is never overwritten by AI.
3. Quality
Automatic checks look for broken placeholders, missing plural forms, translations longer than a key allows, and glossary conflicts, so a broken translation is never written. On Team and above, an AI review also gives each translation a 0-100 quality score with specific findings, and flags the ones a person should check. Reviewers approve or flag translations, and can see a string in context through screenshots and the in-context editor. See Quality checks and Translation states.
4. Delivery
Finished languages go back out in the way that suits your product:
- Pulled into your build with the CLI or the GitHub Action.
- Proposed as a pull request by a connected repository.
- Published to the CDN, which your app reads at runtime. See The CDN.
- Published back to the content platform the strings came from.
- Exported as files from the project.