Studio

Give recurring translations shared product knowledge, configure how they run, and resolve the decisions they return.

A correction can improve one translation. A clear term definition or instruction can also help later translations avoid the same problem. Studio is where you maintain that reusable language knowledge and decide how Lingo.dev should use it.

A localization engine combines model choices with attached glossaries, instructions, and brand voice. Applications and integrations select a localization engine when they request translation. The saved configuration gives those requests a shared basis without requiring each caller to reconstruct the team's guidance.

Start with a localization engine #

Create a localization engine for the work you want to translate. The web app's creation form starts with recommended primary and fallback models, no selected language resources, and a translation-only pipeline. You can choose models and attach existing resources during creation or change them later. The defaults are enough to produce a first translation.

Then add the knowledge that changes what a good translation means for your content. A product term needs a definition and usage boundaries. A recurring writing requirement needs an instruction. Tone and formality need an account of the audience. Translation context helps place each kind of guidance where it belongs.

Glossaries, rulesets, and brand voices can be shared across localization engines. Check their attachments when changing them: a shared edit can affect several workflows. Generate guidance from existing content offers another way to prepare these resources, with inspection before use.

Choose how requests run #

Model selection and pipeline policy answer different questions. Models and fallbacks determines which model is attempted for a language pair and what happens when it cannot complete the request. Pipeline determines which rewriting and human-review stages a translation job uses.

Some choices concern the material being reused or returned:

  • Translation reuse explains when an existing target can be retained instead of generated again.
  • Prompt caching explains provider reuse of repeated model input, which is separate from retaining an accepted translation.
  • Length limits explains how to request shorter wording for a constrained interface and what still needs a rendering check.
  • Review policy brings language coverage, reviewers, bypass conditions, and timeout behavior into one decision.
  • Human review explains how the pipeline waits for a person and continues with their decision.
  • Decision pools explains how to manage reviewers and route review by language and directory.

Resolve the decision a result asks for #

The Inbox brings together work that needs attention. A review accepts or edits a particular translation. A suggestion proposes a change to shared guidance for future requests. Triage can connect a reported issue to language changes, tracker work, or a separately approved retranslation.

Keep those decisions distinct. Accepting a corrected sentence does not mean its wording belongs in every later translation. Changing a glossary does not rewrite output already published. Each action has a different scope, and understanding that scope helps a useful correction become useful shared knowledge.

Investigate an unexpected result #

If saved guidance appears not to have affected a translation, first check whether it was applicable to that request. Context retrieval and locale resolution explain selection. Request evidence helps distinguish what actually ran from the configuration visible now.