Fill from Your Website

Fill a localization engine's glossary, rules, and brand voice from your website - Lingo.dev reads your pages and drafts them for you to review.

Use your website to draft a glossary, translation rules, and a brand voice for your localization engine. Lingo.dev reads up to five pages of your site and uses your existing content to identify terminology, writing conventions, and tone.

In the Studio 2.5 navigation, open Studio 2.5 → Localization engines instead of Infrastructure in the steps below. Existing engine links still work.

The generated resources are attached to your engine. You can review and edit them like any glossary, ruleset, or brand voice you create yourself.

Fill an existing engine #

  1. Open Infrastructure and select your engine.
  2. Select Fill missing context. The button appears while the engine is missing a glossary, ruleset, or brand voice.
  3. Enter your website URL. Check the screenshot to confirm you've selected the right site. Lingo.dev also detects the site's language.
  4. Under What to fill, choose the resources to generate. Resource types already attached to the engine, or that your role can't create, are unavailable.
  5. Select Start.

Generation runs in the background. You can leave the page while it works.

Other ways to start #

  • When creating an engine: Enter your URL at the website step. Generation starts after the engine is created.
  • To create one resource: Select Create from website on the glossary, ruleset, or brand voice list. Choose the engine to attach it to.
  • From your AI assistant: Use MCP to create and fill a new engine or fill an existing one. Through MCP, you can also provide up to ten links or pasted texts instead of one website.

Which pages it reads #

Lingo.dev reads the page you provide, then selects additional pages that describe your product, such as documentation, pricing, product, and about pages. It reads up to five pages in total.

When selecting additional pages, it skips those disallowed by your site's robots.txt. An HTTP 4xx response is treated as an absent file, so additional pages may still be read. A timeout, fetch error, HTTP 5xx response, or successful response without text prevents reading additional pages on that site.

What it creates #

Generation uses the source and target languages configured on your engine.

Glossary #

The glossary records your own brand, company, and product names so translations keep them as written. For example, a product named "Acme Cloud" would keep that name across languages.

It also records feature and UI names when your website provides an established translation. It does not invent translations for those names.

Other companies' product names and standards are left out because translations keep them as written without a glossary entry.

Rules #

Rules describe writing conventions found in your content, such as how to format units or which wording to avoid. The rules are grouped in a ruleset.

Each rule applies to all target languages or to the specific languages it concerns. For example, a convention found only in your French content can become a French-specific rule.

Brand voice #

The brand voice describes how translations should sound, including tone, formality, and how to address the reader.

When the voice is the same across target languages, Lingo.dev creates one shared description. It adds language-specific descriptions only where your content supports a difference.

What happens to existing content #

Generation creates new resources rather than editing or merging into existing ones:

  • Fill missing context creates only resource types the engine does not already have. An attached glossary counts as present even if it contains no terms.
  • Create from website can add a new glossary or ruleset alongside those already attached to the engine.
  • An existing brand voice is never replaced. Brand voice generation is skipped if the engine already has one.

Progress and retries #

Progress remains visible across the app while the job runs. Each engine can run one generation job at a time.

When the job finishes, each resource is marked as completed, skipped with a reason, or failed. Retrying starts a new job using the same sources and generates only failed resource types with no set attached to the engine. If none qualify, the retry returns NothingToProvision.

Your organization can start up to 10 jobs in any 24-hour period, including retries.

Review the results #

Treat the generated content as a first draft. Review the glossary terms, rules, and brand voice descriptions. Correct or remove anything that does not fit your product.

Next Steps #

For a server-created engine, the Provisioning API documents source payloads, polling for per-resource progress, and recovery after a partial result.