Localization Agent

⬜ Proposed — not built. Tier 1 — buildable now

What it would do

Transcreates drafts into Hindi and regional languages — idiom and tone, not word-for-word. Critically, re-runs compliance in the target language, because a safe English sentence can turn unsafe in translation.

Why an agent

Translation has APIs. Transcreation is judgment: idiom, cultural fit, and keeping a compliance-safe phrasing safe once the words change.

What it needs first

Nothing new. OpenAI is already wired. Needs a per-tenant language allow-list and a do-not-translate glossary (product names, regulatory terms).

Roadmap source

Vernac — pages 3, 4 and 6 — see roadmap_ref_extracted.md.

Design notes and open questions

Who reviews vernacular output when the reviewer does not read the language? This is an operational blocker, not a technical one.
Does the CMS support multi-locale posts, or is each language a separate draft?
Compliance must re-run on the translated body — the existing verdict does not carry over.

Subagents

Transcreator — idiom and tone per language, not word-for-word
Glossary guard — holds product names and regulatory terms untranslated
Compliance re-checker — re-runs review in the target language, because safety does not survive translation automatically
Layout fitter — vernacular text expands; catches what breaks templates and aspect ratios

Stub spec. Expand to the full overview.md + technical.md split (as the three built agents use) when this moves from proposed to scheduled.