AI Translation Stack adds guarded, memory-assisted localization workflows

Summary

AI Translation Stack is a framework-independent TypeScript toolkit for translating and serving JSON message catalogs with AI providers. It tracks source changes, invalidates stale locale values, preserves placeholders and HTML tag structure, validates glossary terms, and detects likely hard-coded user-facing strings. Its translation runner combines exact translation memory with optional semantic retrieval, sends glossary-aware requests through provider adapters, validates returned JSON, checkpoints each chunk atomically, and records locale, model, keys, response identifiers, errors, and usage without storing prompt text. The package includes a next-intl adapter that loads the source catalog first, overlays only non-empty configured translations, and falls back to the source locale for unknown locales or incomplete values. Audit, synchronization, baseline, scanning, testing, linting, and build commands do not require provider credentials; an API key is needed only for translation. Six selectable adapters are documented for OpenAI, Claude, DeepSeek, GLM, Grok, and Gemini, with optional embeddings where supported. The project requires Node.js 22.17 or newer and TypeScript 7, and is distributed under the Apache License 2.0. The repository warns that it does not redact sensitive source strings or guarantee linguistic correctness, so human review remains appropriate for high-risk or brand-sensitive content.