Catch localization bugs
before your players do.

• A translator deletes the {count} placeholder — the reward never pays out.
• An untranslated English string ships to the Japanese build.
• The German text overflows the button and gets cut off.
• "Mana" becomes three different words across the game.

1. 01

   Upload your strings

   Drop in CSV or JSON. Loclint auto-detects the encoding and your placeholder format — {curly}, {{double}}, %s, printf.

2. 02

   Every update gets checked

   A deterministic rule engine and a confidence-scored AI pass flag issues by category and severity — in seconds, not a full manual QA cycle.

3. 03

   Fix it before you ship

   Issues come back with suggested fixes — correct your file, re-run the check, and ship clean.

You could paste ten strings into ChatGPT, once. Here's why that stops working.

Placeholders need certainty, not a guess

A missing {count} is a bug, not a judgment call. A chat model checks probabilistically — it misses some and invents others. Loclint's rule engine catches every one, deterministically.

It has to run every time

You won't paste 5,000 strings into a chat on every release. With Loclint, every update is one upload — checked against the same rules, in seconds.

A report, not a wall of chat

Issues come back categorized, severity-ranked, and deduplicated, with suggested fixes — not paragraphs to re-read.

Placeholders, tags, character limits, terminology, character voice — checked as game text, not generic strings. The AI review pass goes further, catching what native Japanese players feel but bug reports never name — honorific drift, translationese, a line that breaks character.

vs. manual LQA

Manual checking is slow and can't cover every update — or every language. Loclint checks every locale you ship against the same rules, in seconds.

vs. Lokalise & Crowdin

They manage strings. Loclint checks whether those strings actually work in-game — placeholders, limits, terminology, tone.

vs. spreadsheets

A spreadsheet can't tell you a placeholder went missing. Loclint can, on every row.

Built in Japan.

Dropped subjects, shifting honorifics, character voice — Japanese is brutal to QA from the outside. Loclint is built where that bar is set. And the same checks run on every language you ship.

• Under NDA? Turn AI review off and your text never reaches any external AI — the rule checks run in full on Loclint's own infrastructure. It's a per-check toggle, so you decide run by run.
• Your text is never used to train AI models — not by us, and not by the AI providers we call.
• Your strings live on Cloudflare, encrypted in transit and at rest. Delete a project and the files, the extracted strings, and the results go with it.

See exactly where your text goes, and when it's deleted →

Will it understand my placeholder format?

Yes — formats are auto-detected per project: {curly}, {{double}}, %s, printf, and more.

Is the AI noisy?

The rule engine is deterministic. AI issues carry a confidence score, and low-confidence ones are hidden by default.

How much setup?

Upload a file and go. Set per-project character limits and a glossary to enable overflow and terminology checks; GitHub integration for per-push checks is on the way.

Is it safe to upload text from an unannounced title?

AI review sends strings to the OpenAI API — never used for training, deleted within 30 days — and it's a per-check toggle, so text from an unannounced title need never reach it. Deleted data clears our database backups within 30 days too.

What does it cost?

Free to start — 500 strings checked a month, no credit card. Re-checking a file only counts the strings that changed since the last check, so running it on every build costs almost nothing. Paid plans start at ¥2,900/mo (Creator), with Studio and Agency as your volume grows.