glot

Your coding agent localizes your app. Deterministic code guarantees that nothing broken or too long lands in your catalogs.

The real glot translate binary. The LLM is scripted to get three strings wrong on its first pass; every refusal and message is glot's. Reproduce it, or rerun it with a real model doing the retries: docs/launch/record/refusal.sh.

npx glot-i18n init        # detect your i18n stack, write glot.toml + rules for your coding agent
npx glot-i18n i18nize     # hardcoded UI strings -> keys, components rewritten (next-intl, react-i18next)
npx glot-i18n translate   # translate only the delta; invalid output is refused, retried, never written

Or all of it in one idempotent step: npx glot-i18n localize --yes --targets fr,de (init if needed → i18nize → translate → check).

Guarantees

Checked by code on every LLM output, on glot set, in the overlay and in CI (glot check). A refused string goes back to the model with the exact reason (2 retries, then the fallback model); if it still fails, nothing is written.

Placeholders	{name}, {{name}}, tags: same set as the source, none invented, none translated. ICU, i18next and MessageFormat 2 syntax parsed.
CLDR plurals	Every category the target locale needs (pl: one, few, many, other; ar: six of them).
Glossary	glossary.toml in your repo: required and forbidden translations, do-not-translate terms.
Pixel width	glot measure records each key's box and font from your running app; translations are shaped with that font and refused if wider. glot fit shortens what already overflows.
Lockfile	glot.lock hashes source and target: only the delta is translated, and a hand edit is never overwritten without --force.
MCP for your agent	glot mcp gives Claude Code, Cursor or Codex your keys, their usages (file:line, UI role), the glossary, and tools that add strings through the same validator.

Works with the i18n library you already use, adds no runtime, bring your own LLM key (Anthropic, OpenAI, any OpenAI-compatible endpoint). MIT.

More: Docs (concepts, CLI, MCP, integrations & GitHub Action, agent skill, AI-translation provenance, FAQ & costs) · Labs (SEO, database content, emails/images, task parity, plural coverage…) · Self-hosted server (experimental) · Demo app · Architecture