Internationalization (i18n) Guide

This document explains how language support works in the Beledarians LM Studio Toolbox plugin and how to add new locale translations.

---

Supported Languages

Locale ID	Language	Script
en	English	Latin
zh-CN	Chinese (Simplified)	简体中文
zh-TW	Chinese (Traditional)	繁體中文
de	German	Latin

---

Architecture: Dual-Layer Language Mechanism

The plugin uses a two-layer system to work around an architectural constraint in the LM Studio SDK: context.withConfigSchematics() can only be called once at plugin startup and cannot be updated afterwards.

Layer 1 — Config UI Language (Static, Boot-Time)

Layer 2 — Message Language (Dynamic, Per-Turn)

• What it covers: All runtime text — delegation hints injected into prompts, plan mode instructions, retrieval status messages, citation prefixes/suffixes, document injection text.
• How it works: A "Message Language" dropdown is exposed in the plugin settings (part of Layer 1). On every user message, promptPreprocessor.ts reads the current value of this dropdown and looks up the matching dictionary.
• When it changes: Immediately — on the very next user message, with no plugin restart required.

Example Scenarios

OS Language	Message Language Dropdown	Result
zh-TW	zh-TW	Full Traditional Chinese UI and runtime text
en-US	zh-CN	English config panel, Simplified Chinese runtime hints
de-DE	en	German config panel, English runtime hints

---

File Structure

---

Adding a New Language

---

Chinese Locale Notes

Sinophone Variant Routing

The resolveLocale() function maps the following OS tags:

System Tag	Routes to
zh-TW, zh-HK, zh-MO, zh-Hant	zh-TW
zh-CN, zh-Hans, zh-SG, zh	zh-CN

Both dictionaries ship as full, independent translations — there is no silent fallback between zh-CN and zh-TW. Each receives its own complete string set appropriate to its regional vocabulary.

Recommended Embedding Model for Chinese

For RAG features with Chinese-language documents, the default nomic-ai/nomic-embed-text-v1.5-GGUF model has degraded performance on multi-byte Chinese characters. The zh-CN and zh-TW config subtitles explicitly recommend:

Additionally, for Chinese documents, lowering the Retrieval Affinity Threshold to 0.35–0.45 (from the default 0.5) typically improves recall on Chinese technical text.

---

Contributing Translations

To contribute a translation or correction:

• Fork the repository.
• Create or update a locale file under src/locales/.
• Run npx tsc --noEmit to confirm zero type errors.
• Open a Pull Request with a description of the changes.

Chinese (zh-CN / zh-TW) review from native speakers is particularly welcome — please open an issue or PR if you spot unnatural phrasing in technical terms.