How it works

Install once. Then four things happen.

The script finds your text. Your rules steer the AI. Two checks review every translation. Then you fetch the finished files into your own site, so your pages stay translated even if BeMyWords vanished tomorrow.

Pass 01 — Discover

It reads the site you already have.

Paste the tag and the script reads your sitemap. Then it picks up new text as visitors open pages, including text inside components and CMS fields. If you work from a repository, you push the keys yourself. An AI assistant can also register them over MCP while it writes them.

  • Works on Tilda and hosted sites, Astro, Rails, React, Vue, iOS, Android, Flutter.
  • It leaves alone anything you mark as not to be translated.
Pass 02 — Your rules

Your rules go in before the AI writes a word.

Your glossary, tone, fixed substitutions and product names go to the AI with every sentence, as rules it must follow. "Skiwo" stays "Skiwo" in every language.

  • Different instructions per language, if you want them.
  • A sentence translated once is reused in every project.
Pass 03 — Checkup

Two checks, then a short list.

The checkup reads every translation against your rules. The language check finds text that ended up in the wrong language, like an English line in your Norwegian tab. You get one score and a short list of sentences worth a look.

99.2% on an example run. 3 sentences flagged, none blocking.
Pass 04 — Ship

Your own site serves the files.

Your build fetches the finished translation files and ships them with your site. Your visitors make no extra request, and nothing of ours sits between them and your pages. You can also serve translations live from the script, or sync them over the API.

  • JSON, YAML, PO, .strings — plus a drop-in converter for Android XML and Flutter ARB.
  • Export everything, any time.
  • Connect GitHub, and one click rebuilds your site with the fresh translations. An edit never waits for your next deploy.
  • Files in your build give your site a real address per language, which is what search engines index.
Bring your own model

Point it at your own inference endpoint and the price becomes 199 kr.

Any OpenAI-compatible endpoint works, hosted or self-hosted, with a separate vision endpoint if you want one. Paste the endpoint and key, and the flat plan is one click. The price stays flat however many words you translate.

See how the switch works
A quiet Nordic desk