Cross-Border Commerce Dictionary
as a Service

Designed for 49 languages.
9 production-ready today, scaling to 12-15 by Q3 2026.
Sub-5ms latency. HTTP API only.

Try sandbox curl →Get API keyRead docs

One concept, every language

Your customer searches 「シミ取り」 in Japanese. mogan-i18n resolves to canonical skin_care_spot_remover and returns equivalents across 49 designed languages (9 production-ready today):

JP シミ取り
ZH 除斑
KR 기미 제거
FR anti-tache
IT rimuovi macchie
BR removedor de manchas
ES quitamanchas
DE Fleckenentferner

One API call. Sub-5ms. No external service roundtrips.

1,474,726 Redis keys | 76,000 canonical entries | < 5ms p50 latency · Live grep at /healthz

Built for shopping carts. Not for travel phrases.

Generic translation services translate words.
mogan-i18n translates commerce intent.

"tênis Brasil" → sneakers (BR pt-BR — not the device "tennis")
"geladeira" → refrigerator
"máquina de lavar" → washing machine
"シミ取り" → skin care spot remover (not "stain remover")

Each entry is type-tagged (brand / category / composite / integrated) with confidence scores and back-translation Jaccard validation.

25K+ leaf categories mapped to Google Product Taxonomy

The dictionary learns every night.

Every time a customer searches a term we don't know yet, mogan-i18n logs the miss (PII-scrubbed per GDPR Art. 5).

At 3AM, our cron pipeline:

  1. Scrubs PII (email / phone / SSN / national ID patterns)
  2. Calls Claude Haiku for forward + back-translation
  3. Validates via Meilisearch reverse-match (≥5 hits required)
  4. Auto-approves if Jaccard similarity ≥ 0.85
  5. Writes to Redis with provenance metadata (_source: 'dict-grow-auto', _model: claude-haiku-4-5, _approved_at, _similarity)

Next time anyone searches that term — Redis fast-path, < 5ms, with AI provenance audit trail.

30-day pending retention | 90-day translated retention | $10/night budget cap | per-IP rate limit 10/hr