Know your app says the right thing in every language, before your customers do.
i18nProof scans your locale files and watches what your running app shows and sends. Several AI models review each translation and argue about it; a person sees only the ones they doubt. Nothing is reviewed twice.
How it works
Connect a repo and add the SDK to your test build. Your translators, your pipeline and your providers stay exactly where they are.
Scan
Every string in every locale file is paired with its source. Untranslated text, dropped placeholders, missing plural forms, broken markup and hardcoded English are caught by rules before any model is called.
Watch
The SDK sits where your app renders templates, generates text with AI, or sends email and SMS. Each piece of text is fingerprinted at the edge, and only text never seen before goes any further.
Argue
A panel of models reviews each translation from different angles: meaning, tone, terminology, fit. A judge weighs the disagreement. Only what the panel cannot settle reaches a person, in that person's own language.
Fix
Each finding names the file and key, or the template or prompt that produced the text, with the fix ready to apply. Open a pull request from the panel, or hand it to your AI coding assistant over MCP.
What teams catch with it
Strings left in English
A checkout button still reading "Checkout" in Spanish, found in seconds, with no model cost.
Dropped placeholders
A German order email that lost {orderId}, so no German customer ever saw an order number.
Wrong tone
One informal "du" in an app that uses "Sie" everywhere else. Rules miss it; the panel does not.
Translated brand names
"Harbor Goods" turned into "Bienes del Puerto". Your glossary says what must never be translated.
Text that does not fit
A six-word German label on a one-word button, flagged before a designer finds it in a screenshot.
AI text about the wrong thing
A generated Japanese product description for a blanket that describes a desk lamp. Caught in traffic, traced to the prompt.
Reviewed once, remembered for good
The ledger is the product
Every verdict is stored against a fingerprint of the text. A string that has a verdict is never sent to a model again, across builds, branches and runs. You pay for new text, not repeated text.
Disagreement on purpose
Models that agree are not evidence. The panel is built to disagree, and the panel's confidence is checked against what your reviewers actually decide.
Your reviewers or ours
Invite your own translators at no seat cost, or route a locale to i18nProof's experts at a published per-word rate. Switch per locale.
Built for agents too
Everything the panel does is available over an API, a CLI and an MCP server with 26 tools, so a coding assistant can scan, review and fix without a person in the loop.
Try it on your own app
Start on the Starter plan with one repo and three locales. Every plan includes unlimited reviewer seats and a monthly spend cap that is on by default.