Translate (Translation) — foundations RFC

Status

*ccepted*— 20260509. Sector bootstrap (skeleton + 5 impl tickets) landed as part of /k-go services/ai audit wave (Modo C). Q1 resolved: NMTfirst (NLLB200 v1) with LLM fallback for lowconfidence/longform. Q2 resolved: glossary lives in this sector — koder-official.toml versioned in repo + per-tenant overlays via API.

Summary

Tradução ptBR ↔ enUS ↔ outros como service — separado de libras/ (sign language).

Motivation

Stack inteira é bilingual (chat ptBR, produtos enUS). Hoje tradução acontece ad-hoc com gateway+LLM custoso. Service dedicado: cache, qualidade controlada, custo menor.

Scope

In

• Text translation (ptBR/enUS prioritário, +eszhja secundário)
• Document translation (preserve format)
• Terminology glossary

Out (yet)

• Real-time speech translation (escopo synth+voice combinados)

Initial design

Surfaces

• backend/ — Go API + worker
• app/ — não aplicável v1

Key APIs

• POST /v1/translate/text — texto direto
• POST /v1/translate/document — preserva format
• POST /v1/translate/glossary — gerencia terminologia

Dependencies

• services/ai/gateway — LLM-based fallback
• services/ai/cache — cache de traduções repetidas
• services/ai/prompt — templates de translation

Relation to existing sectors

• Distinto de libras/ (sign language)
• Consumido por i18n/ workflows internos

Selfhostedfirst analysis (5 gates)

• *1 Feature parity* zero
• *2 Performance* N/A
• *3 Stability* N/A
• *4 Capability* NLLB/M2M-100 + LLM via gateway viáveis
• *5 Critical-path readiness* destrava i18n efficient

Open questions

• Q1: NMT (modelo dedicado) vs LLM only?
• Q2: Glossary — onde mora terminologia Koder oficial?

Next steps

1. Ratificar esta RFC (1 round de comments).
2. Criar sector dir services/ai/translate/ com koder.toml, README.md, skeleton.
3. Abrir tickets de implementação em services/ai/translate/backlog/pending/.
4. Registrar em meta/docs/stack/registries/self-hosted-pairs.md se substituir externo.