Translate

Take source content through generation, retrieval, and inspection in the workflow that will use it.

Lingo.dev translates source content with language models and shared guidance from a localization engine. The result returns to the caller: a downloaded JSON document for application data, target files for a file workflow, or the destination supported by an integration.

Start with a small piece of the content you actually need to translate. Follow it through to a retrieved result and inspect both its meaning and the structure the receiving system uses. This establishes a complete first task before you automate recurring work.

Start where the source already lives #

SourceFirst complete task
Files in a projectTranslate your first file with the CLI, from installation and authentication through target-file retrieval.
Data in an applicationMake your first API request, translate a structured payload, and save the returned messages locally.
A connected repository or content systemChoose an integration that reads the intended source and returns changes to the right destination.

Translation jobs accept inline JSON or uploaded files. Create the job, upload any missing files, start it, and wait for the requested targets to complete. Each completed target supplies a download URL. The engine determines which pipeline stages run.

If the translations already exist and the question is their quality, go directly to Measure. They do not need to be generated again or moved from their current production system.

Give the first translation a useful test #

Choose one target locale and a source you can examine. For an application catalog, include a label, a sentence with a variable, and a message whose meaning depends on context. For a document, include a heading, a link, and a paragraph containing a product term. Those examples test more of the real task than an isolated greeting.

Keep the source revision and intended output destination with the work. Check Formats for what a file handler translates and preserves, and Languages for language, region, and script identifiers.

The chosen localization engine supplies model choices and shared translation guidance. A new localization engine can translate before you attach guidance, but a model cannot infer every product convention from a short sentence. Supply the context needed to understand your example.

Inspect the result you will use #

Follow the chosen quickstart through retrieval. If the work pauses for review or fails, use that workflow's recorded state to choose the next action. Repeated submission without knowing what completed can duplicate work and cost.

Then inspect two different properties of the output:

  1. Meaning. Compare the intended action, claim, terminology, and tone with the source and guidance. Use a qualified reviewer or an applicable measurement when you cannot judge the target language yourself. Successful execution does not establish language quality.
  2. Usable structure. Confirm the target locale, path, message keys, variables, links, and protected content. If an application already exists, load the output there and check substitution and layout. If the task only created a file, inspect that file directly as the quickstart describes.

Keep an accepted target with the source it describes. Once the author changes that source, the earlier translation may need revision even though it was correct when accepted.

Make repeated work predictable #

After the first result is useful, decide what starts another run, what changed content it should include, and where acceptance happens. Integrate develops the connection; translation reuse explains how existing output can survive updates; Plan a multilingual release follows the work through readiness for publication.