AI Code Identifier Translator
Stop context-switching just to read a variable name.
If you've ever opened a file and hit a wall of 用户数量, مدير_الإعدادات,
or текущаяПопытка, you know the drill — alt-tab to a translator, copy the
identifier, paste, alt-tab back, repeat for every symbol in the file. This
extension does that step for you, live, inline, as you scroll.
🌐 Chinese, Japanese, Korean, Arabic, and Russian identifiers are shown
as English right in the editor — 机器人控制器 reads as RobotController,
当前坐标X reads as currentCoordinateX.
🔒 Your file never changes. Not one byte. Translation is a pure visual
overlay (VS Code decorations) on top of the real text — git diff stays
empty, and moving your cursor onto a translated name instantly reveals the
original underneath.
🖥️ 100% local. Every translation runs through Ollama
on your machine. No API key, no cloud call, no code ever leaves your
computer.
⚡ Fast after the first pass. Translations are cached per-repository, so
reopening a file you've already scanned is instant.
Status: this is Phase 1 — read-side, decoration-based display. Editing
support (typing the English name and having it write back the original
script) is on the roadmap but not yet implemented; see ARCHITECTURE.md
§8 for the design and why it's a separate phase, and KNOWN_LIMITATIONS.md
for the current honest list of rough edges.
Requirements
Build & run from source
git clone <this-repo>
cd ai-identifier-translator
npm install
npm run compile
Then in VS Code: F5 (or Run → Start Debugging) to launch an Extension
Development Host with the extension loaded.
Package for local install
npm install -g @vscode/vsce
vsce package
code --install-extension ai-identifier-translator-0.1.0.vsix
Usage
- Make sure Ollama is running (
ollama serve, or it's already running as a
background service).
- Open a file containing non-English identifiers.
- Translation happens automatically for the visible portion of the file
(toggle via
AI Identifier Translator: Toggle Translation in the command
palette, or aiIdentifierTranslator.enabled setting).
- Move your cursor onto a translated (English-displayed) identifier to
temporarily reveal the real, original text underneath.
AI Identifier Translator: Clear Cache for This Repository if you want
to force retranslation (e.g. after switching models).
Settings
See package.json → contributes.configuration for the full list
(aiIdentifierTranslator.*): target language, model, Ollama endpoint,
temperature, concurrency limit, cache location, and decoration colors.
Testing
npm run compile
npm test
Unit tests cover the identifier scanner's regex fallback path and the
SQLite cache's per-repository isolation guarantee. (Decoration-rendering
behavior is inherently harder to unit test — it's covered by manual smoke
tests per VS Code release, see KNOWN_LIMITATIONS.md §1.)
Project layout
src/
├── extension.ts # activation, command registration, wiring
├── types.ts # shared type definitions
├── config/ConfigManager.ts # typed settings wrapper
├── cache/CacheManager.ts # SQLite cache + multi-repo router
├── identifiers/
│ ├── IdentifierScanner.ts # semantic-token-based + regex-fallback scanning
│ └── RepoIdentity.ts # stable per-repo hash for cache isolation
├── translation/
│ ├── OllamaClient.ts # local Ollama HTTP client, JSON-mode prompting
│ └── TranslationQueue.ts # cache-first resolution + throttled model calls
└── decorations/
└── DecorationController.ts # pure-visual rendering, never edits the buffer
Why the file is guaranteed unchanged
The extension never calls TextEdit / WorkspaceEdit on your source files.
All "translation" is implemented as editor decorations (a purely visual
layer VS Code renders on top of the real buffer). This is why git status
/ git diff will always show zero changes from this extension — there is
no code path in this project capable of writing to your files. See
ARCHITECTURE.md §1 and §6 for the mechanism.
