Jta - JSON Translation Agent

AI-powered Agentic JSON Translation tool with intelligent quality optimization

Jta is a production-ready command-line tool that uses AI to translate JSON internationalization files with exceptional accuracy and consistency. It features an agentic reflection mechanism where AI translates, evaluates, and refines its own work, along with automatic terminology detection and robust format preservation for production-grade translations.

📑 Table of Contents

🔥 Agent Skills
Key Features
Installation
Quick Start
Documentation
Supported AI Providers
Supported Languages
Architecture
Examples
Configuration
Troubleshooting
FAQ
Contributing
License
Acknowledgments
Support

🔥 Agent Skills

Jta can be used as an Agent Skill to enable AI agents like Claude to automatically translate JSON i18n files.

Quick Start with Skills

For Individual Users:

# Copy the skill to your Claude skills directory
cp -r skills/jta ~/.claude/skills/

# Or create a symbolic link (recommended for development)
ln -s $(pwd)/skills/jta ~/.claude/skills/jta

For Project Teams:

# The skill is already in the repository at skills/jta
# When team members clone the repo, they can use it immediately
cp -r skills/jta .claude/skills/

Using the Skill:

Once installed, simply ask your AI agent:

"Translate my en.json to Chinese, Japanese, and Korean"

The agent will automatically:

Install Jta if needed
Verify API key configuration
Execute translation with optimal settings
Show results and statistics

What's Included

The skills/jta directory contains:

SKILL.md - Complete skill definition and instructions for AI agents
examples/ - Step-by-step use cases:
- Basic translation workflow
- Incremental translation mode
- CI/CD integration
scripts/ - Installation helpers

Learn More

See skills/README.md for complete documentation on using Jta as an Agent Skill.

✨ Key Features

🤖 Agentic Translation with Self-Optimization

Agentic Reflection Mechanism: AI acts as its own quality reviewer through a two-step process - first translating, then critically evaluating and refining its own work
Multi-Dimensional Quality Evaluation: The AI examines translations across 4 key dimensions: accuracy (no mistranslations), fluency (natural grammar), style (cultural appropriateness), and terminology (consistency)
Self-Generated Improvements: Rather than relying on predefined rules, the AI generates specific, contextual suggestions and applies them to produce better translations
Iterative Refinement: Each translation goes through translate → reflect → improve cycles, ensuring higher quality output
Trade-off: 3x API calls per batch for significantly improved translation quality

📚 Intelligent Terminology Management

Automatic Detection: Uses LLM to identify important terms in your content
Preserve Terms: Brand names, technical terms that should never be translated
Consistent Terms: Domain-specific terms translated uniformly across all content
Editable Dictionary: Saved to .jta-terminology.json for manual refinement

🔒 Robust Format Protection

Automatically preserves:

Placeholders: {variable}, {{count}}, %s, %(name)d
HTML Tags: , , <a href="...">
URLs: https://example.com, http://api.example.com/v1
Markdown: **bold**, *italic*, [link](url)

⚡ Smart Incremental Translation

Only translates new or modified content
Preserves existing high-quality translations
Automatically removes obsolete keys
Saves time and API costs (typically 80-90% reduction on updates)

🎯 Flexible Key Filtering

Glob Patterns: settings.*, user.**, *.title
Precise Control: Include or exclude specific sections
Recursive Wildcards: Translate entire subsections with **

🌍 RTL Language Support

Proper bidirectional text handling for Arabic, Hebrew, Persian, Urdu
Automatic direction markers for LTR content in RTL context
Smart punctuation conversion for Arabic-script languages

🚀 Production-Ready Performance

Batch processing with configurable concurrency
Retry logic with exponential backoff
Graceful error handling and recovery
Progress indicators and detailed statistics

🎨 Multi-Provider Support

OpenAI: All models including GPT-5, GPT-5 mini, GPT-5 nano, GPT-4o, etc.
Anthropic: All Claude models including Claude Sonnet 4.5, Claude Haiku 4.5, Claude Opus 4.1, etc.
Gemini: All Gemini models including Gemini 2.5 Flash, Gemini 2.5 Pro, etc.

📦 Installation

Homebrew (macOS/Linux) - Recommended

The easiest way to install Jta on macOS or Linux:

# Add the tap
brew tap hikanner/jta

# Install Jta
brew install jta

# Verify installation
jta --version

Upgrade:

brew upgrade jta

Uninstall:

brew uninstall jta
brew untap hikanner/jta

Download Binary

Download the latest release for your platform from GitHub Releases:

macOS: jta-darwin-amd64 or jta-darwin-arm64 (Apple Silicon)
Linux: jta-linux-amd64 or jta-linux-arm64
Windows: jta-windows-amd64.exe

# macOS/Linux example
curl -L https://github.com/hikanner/jta/releases/latest/download/jta-darwin-arm64 -o jta
chmod +x jta
sudo mv jta /usr/local/bin/

Using Go Install

go install github.com/hikanner/jta/cmd/jta@latest

From Source

git clone https://github.com/hikanner/jta.git
cd jta
go build -o jta cmd/jta/main.go

🚀 Quick Start

Installation

# Install via Homebrew (recommended for macOS/Linux)
brew tap hikanner/jta
brew install jta

# Or download binary from GitHub Releases
# See Installation section for details

View Supported Languages

# List all supported languages
jta --list-languages

Basic Usage

# Translate to a single language
jta en.json --to zh

# Translate to multiple languages
jta en.json --to zh,ja,ko

# Specify output directory
jta en.json --to zh --output ./locales/

With AI Provider Configuration

# Using environment variables (recommended)
export OPENAI_API_KEY=sk-...
jta en.json --to zh

# Or specify directly
jta en.json --to zh --provider anthropic --api-key sk-ant-...

Advanced Usage

# Incremental translation (only translate new/modified content)
jta en.json --to zh --incremental

# Skip terminology detection (use existing)
jta en.json --to zh --skip-terminology

# Disable terminology management completely
jta en.json --to zh --no-terminology

# Re-detect terminology (when source language changes)
jta zh.json --to en --redetect-terms

# Translate specific keys only
jta en.json --to zh --keys "settings.*,user.*"

# Exclude certain keys
jta en.json --to zh --exclude-keys "admin.*,internal.*"

# Non-interactive mode (for CI/CD)
jta en.json --to zh,ja,ko -y

# CI/CD with incremental translation
jta en.json --to zh --incremental -y

📖 Documentation

Terminology Management

Jta automatically detects important terminology in your source file and ensures consistent translation:

Preserve Terms: Brand names, technical terms that should never be translated (e.g., API, OAuth, GitHub)
Consistent Terms: Domain terms that must be translated uniformly (e.g., credits, workspace, premium)

File Structure:

Terminology is stored in a dedicated directory (default .jta/):

.jta/
├── terminology.json       # Term definitions (source language)
├── terminology.zh.json    # Chinese translations
├── terminology.ja.json    # Japanese translations
└── terminology.ko.json    # Korean translations

terminology.json (source language terms):

{
  "version": "1.0",
  "sourceLanguage": "en",
  "detectedAt": "2025-01-26T10:30:00Z",
  "preserveTerms": ["API", "OAuth", "JSON"],
  "consistentTerms": ["credits", "workspace", "prompt"]
}

terminology.zh.json (translations):

{
  "version": "1.0",
  "sourceLanguage": "en",
  "targetLanguage": "zh",
  "translatedAt": "2025-01-26T10:31:00Z",
  "translations": {
    "credits": "积分",
    "workspace": "工作空间",
    "prompt": "提示词"
  }
}

Workflow:

First run: Detects terms → saves to terminology.json → translates to target language
Subsequent runs: Loads existing terms → translates missing terms only
New language: Uses existing terminology.json → creates terminology.{lang}.json

Custom Terminology Directory:

# Use a shared terminology directory
jta en.json --to zh --terminology-dir ../shared-terms/

# Multiple projects can share the same terminology
jta projectA/en.json --to zh --terminology-dir ~/company-terms/
jta projectB/en.json --to ja --terminology-dir ~/company-terms/

Incremental Translation

Default behavior: Full translation

Jta translates all content by default for maximum quality and consistency
Simple and predictable: jta en.json --to zh always produces a complete translation

Incremental mode (optional): When you use --incremental flag, Jta intelligently:

Detects new keys
Identifies modified content
Preserves unchanged translations
Removes deleted keys

This saves time and API costs (typically 80-90% reduction on updates).

Usage:

# First time: Full translation
jta en.json --to zh

# After updates: Incremental translation (saves cost)
jta en.json --to zh --incremental

# Re-translate everything (if not satisfied with existing translation)
jta en.json --to zh

Best practice:

Development: Use --incremental for frequent updates
Production release: Use full translation for maximum quality
CI/CD: Use --incremental -y for automated updates

Format Protection

Jta automatically protects:

Variables: {variable}, {{count}}, %s
HTML tags: , 
URLs: https://example.com
Markdown: **bold**, *italic*

🎯 Supported AI Providers

Provider	Models	Environment Variable
OpenAI	All OpenAI models (GPT-5, GPT-5 mini, GPT-5 nano, GPT-4o, etc.)	`OPENAI_API_KEY`
Anthropic	All Claude models (Claude Sonnet 4.5, Claude Haiku 4.5, Claude Opus 4.1, etc.)	`ANTHROPIC_API_KEY`
Gemini	All Gemini models (Gemini 2.5 Flash, Gemini 2.5 Pro, etc.)	`GEMINI_API_KEY`

You can specify any model supported by these providers using the --model flag.

🌍 Supported Languages

Jta supports 27 languages with full metadata including flags, scripts, and number systems:

List All Supported Languages

# View all supported languages
jta --list-languages

Output:

🌍 Supported Languages

Left-to-Right (LTR):
  🇬🇧  en      English (English)
  🇨🇳  zh      中文(简体) (Chinese (Simplified))
  🇹🇼  zh-TW   中文(繁体) (Chinese (Traditional))
  🇯🇵  ja      日本語 (Japanese)
  🇰🇷  ko      한국어 (Korean)
  🇪🇸  es      Español (Spanish)
  🇫🇷  fr      Français (French)
  🇩🇪  de      Deutsch (German)
  🇮🇹  it      Italiano (Italian)
  🇵🇹  pt      Português (Portuguese)
  🇷🇺  ru      Русский (Russian)
  🇮🇳  hi      हिन्दी (Hindi)
  🇧🇩  bn      বাংলা (Bengali)
  🇹🇭  th      ไทย (Thai)
  🇻🇳  vi      Tiếng Việt (Vietnamese)
  🇮🇩  id      Bahasa Indonesia (Indonesian)
  🇲🇾  ms      Bahasa Melayu (Malay)
  🇳🇱  nl      Nederlands (Dutch)
  🇵🇱  pl      Polski (Polish)
  🇹🇷  tr      Türkçe (Turkish)
  🇱🇰  si      සිංහල (Sinhala)
  🇳🇵  ne      नेपाली (Nepali)
  🇲🇲  my      မြန်မာ (Burmese)

Right-to-Left (RTL):
  🇸🇦  ar      العربية (Arabic)
  🇮🇷  fa      فارسی (Persian)
  🇮🇱  he      עברית (Hebrew)
  🇵🇰  ur      اردو (Urdu)

Total: 27 languages

RTL Language Support

Special support for Right-to-Left languages (Arabic, Persian, Hebrew, Urdu):

Automatic bidirectional text markers
Smart punctuation conversion for Arabic-script languages
Proper handling of embedded LTR content (URLs, numbers, code)

Language Examples

# Translate to Chinese (Simplified)
jta en.json --to zh

# Translate to Chinese (Traditional)
jta en.json --to zh-TW

# Translate to multiple Asian languages
jta en.json --to zh,ja,ko,th,vi

# Translate to RTL languages
jta en.json --to ar,fa,he

# Translate to European languages
jta en.json --to es,fr,de,it,pt,nl

🏗️ Architecture

Jta follows a clean, modular architecture with clear separation of concerns:

System Architecture

graph TB
    subgraph "🖥️ Presentation Layer"
        CLI[CLI Interface<br/>Cobra + Viper]
        UI[Terminal UI<br/>Lipgloss + Spinner]
    end
    
    subgraph "🔧 Application Layer"
        APP[App Controller<br/>Workflow Orchestration]
    end
    
    subgraph "⚙️ Domain Layer"
        subgraph "Translation Engine"
            ENGINE[Translation Engine<br/>Core Orchestrator]
            BATCH[Batch Processor<br/>Concurrent Processing]
            REFLECT[Reflection Engine ⭐<br/>Agentic Quality Control]
        end
        
        subgraph "Supporting Services"
            TERM[Terminology Manager<br/>Auto-detection + Dictionary]
            INCR[Incremental Translator<br/>Diff Analysis]
            FILTER[Key Filter<br/>Pattern Matching]
            FORMAT[Format Protector<br/>Placeholder Preservation]
            RTL[RTL Processor<br/>Bidirectional Text]
        end
    end
    
    subgraph "🔌 Infrastructure Layer"
        subgraph "AI Providers"
            OPENAI[OpenAI Provider<br/>GPT-5]
            ANTHROPIC[Anthropic Provider<br/>Claude Sonnet 4.5]
            GEMINI[Gemini Provider<br/>Gemini 2.5 Flash]
        end
        
        subgraph "Storage"
            JSON[JSON Repository<br/>File I/O]
        end
    end
    
    subgraph "📦 Domain Models"
        MODELS[Domain Models<br/>Translation • Terminology • Language]
    end
    
    CLI --> APP
    UI --> APP
    APP --> ENGINE
    ENGINE --> BATCH
    ENGINE --> REFLECT
    ENGINE --> TERM
    ENGINE --> INCR
    ENGINE --> FILTER
    ENGINE --> FORMAT
    ENGINE --> RTL
    
    BATCH --> OPENAI
    BATCH --> ANTHROPIC
    BATCH --> GEMINI
    REFLECT --> OPENAI
    REFLECT --> ANTHROPIC
    REFLECT --> GEMINI
    TERM --> OPENAI
    TERM --> ANTHROPIC
    TERM --> GEMINI
    
    TERM --> JSON
    INCR --> JSON
    
    ENGINE -.-> MODELS
    TERM -.-> MODELS
    BATCH -.-> MODELS
    
    style REFLECT fill:#ff6b6b,stroke:#c92a2a,stroke-width:3px,color:#fff
    style ENGINE fill:#4ecdc4,stroke:#087f5b,stroke-width:2px
    style CLI fill:#96f2d7,stroke:#087f5b
    style UI fill:#96f2d7,stroke:#087f5b

Module Responsibilities

Module	Responsibility	Key Features
CLI	Command-line interface	Argument parsing, help text, command execution
UI	Terminal presentation	Colored output, spinners, progress bars, tables
App	Application orchestration	Workflow coordination, error handling, result formatting
Translation Engine	Core translation logic	Batch management, workflow control, result assembly
Batch Processor	Concurrent processing	Parallel API calls, retry logic, rate limiting
Reflection Engine ⭐	Agentic quality control	LLM self-evaluation, improvement suggestions
Terminology Manager	Term management	Auto-detection, dictionary building, term translation
Incremental Translator	Delta processing	Diff analysis, selective translation, merge logic
Key Filter	Selective translation	Pattern matching, inclusion/exclusion rules
Format Protector	Format preservation	Placeholder detection, HTML/URL/Markdown protection
RTL Processor	RTL language support	Bidirectional markers, punctuation conversion
AI Providers	LLM integration	API abstraction, response parsing, error handling
JSON Repository	Data persistence	File I/O, JSON marshaling, validation

Translation Workflow

sequenceDiagram
    participant User
    participant CLI
    participant App
    participant Engine
    participant Term as Terminology<br/>Manager
    participant Batch as Batch<br/>Processor
    participant Reflect as Reflection<br/>Engine ⭐
    participant AI as AI Provider
    
    User->>CLI: jta translate source.json
    CLI->>App: Execute command
    
    rect rgb(240, 248, 255)
        Note over App,Engine: Phase 1: Preparation
        App->>Engine: Load & analyze JSON
        Engine->>Term: Detect/load terminology
        Term->>AI: Detect terms via LLM
        AI-->>Term: Return terms
        Engine->>Engine: Apply key filters
        Engine->>Engine: Create batches
    end
    
    rect rgb(255, 250, 240)
        Note over Batch,AI: Phase 2: Translation
        Engine->>Batch: Process batches (concurrent)
        loop For each batch
            Batch->>AI: Translate with terminology
            AI-->>Batch: Return translations
        end
    end
    
    rect rgb(255, 240, 245)
        Note over Reflect,AI: Phase 3: Agentic Reflection ⭐
        Engine->>Reflect: Review translations
        Reflect->>AI: Step 1: Evaluate quality
        AI-->>Reflect: Suggestions
        Reflect->>AI: Step 2: Apply improvements
        AI-->>Reflect: Improved translations
    end
    
    rect rgb(240, 255, 240)
        Note over Engine,App: Phase 4: Finalization
        Engine->>Engine: Process RTL if needed
        Engine->>Engine: Merge results
        Engine->>App: Return result
        App->>CLI: Format output
        CLI->>User: Display statistics
    end

Key Steps:

Load & Analyze: Load source JSON, detect changes (incremental mode)
Terminology: Auto-detect or load terminology dictionary
Filter: Apply key filters if specified
Batch: Split into batches for efficient processing
Translate: Send to AI provider with format instructions
Reflect ⭐: Two-step Agentic quality improvement (see below)
Process RTL: Apply bidirectional text handling if needed
Merge: Combine with unchanged translations
Save: Write final output with pretty formatting

🔄 Agentic Reflection Mechanism

Jta implements an agentic reflection system where the AI acts as both translator and quality reviewer. Instead of simple one-shot translation, the AI performs a complete quality improvement cycle:

Step 1: Initial Translation (1x API)

Source: "Welcome to {app_name}"
→ LLM Translation
→ Result: "欢迎使用 {app_name}"

Step 2: Quality Reflection (1x API)

The AI evaluates its own translation as an expert reviewer:

AI Reflection Task:
"Review the translation you just created. Analyze it across 4 dimensions:
(i) Accuracy: Are there any errors, mistranslations, or omissions?
(ii) Fluency: Does it sound natural? Any grammar or punctuation issues?
(iii) Style: Does it match the tone and cultural context appropriately?
(iv) Terminology: Are domain terms used consistently and correctly?

Provide specific, actionable suggestions for improvement."

→ AI Self-Critique:
"[welcome.message] The translation '欢迎使用 {app_name}' is accurate but 
could be more natural. Consider '欢迎来到' which conveys a warmer, more 
inviting tone that better matches the welcoming nature of 'Welcome to'."

Step 3: Self-Improvement (1x API)

The AI refines the translation based on its own expert feedback:

AI Improvement Task:
"Based on your expert analysis, improve the translation:
Original: Welcome to {app_name}
Initial Translation: 欢迎使用 {app_name}
Your Suggestion: Use '欢迎来到' for a warmer, more natural tone

Create the improved version while maintaining accuracy and format."

→ AI Improved Translation:
"[welcome.message] 欢迎来到 {app_name}"

Why Agentic Reflection Works

Key Advantages:

AI as Expert Reviewer: The same AI that translated understands the context, nuances, and challenges - making it uniquely qualified to critique and improve its own work
Beyond Static Rules: Instead of checking against predefined patterns, the AI dynamically identifies issues specific to each translation's context, tone, and cultural appropriateness
Contextual Improvements: The AI generates specific, actionable suggestions tailored to each piece of content rather than applying generic fixes
Iterative Quality: Each translation benefits from a complete review-and-refine cycle, catching subtle issues in fluency, tone, and cultural fit that single-pass translation might miss

Implementation Details:

Cost Structure: 3x API calls per batch (translate → reflect → improve)
Example: For 100 keys with batch-size 20: 15 total API calls (5 translate + 5 reflect + 5 improve)
Trade-off: 3x API cost in exchange for significantly higher translation quality
Optimization: Adjust --batch-size based on your needs (smaller batches = more reliable, larger = more efficient)
Model Impact: More capable models (GPT-5, Claude Sonnet 4.5, Gemini 2.5 Pro) produce better reflection insights and improvements

💡 Examples

Example 1: First-time Translation

$ jta en.json --to zh

📄 Loading source file...
✓ Source file loaded

📚 Loading terminology...
🔍 Detecting terminology...
✓ Detected 8 terms

🤖 Translating...
✓ Translation completed

💾 Saving translation...
✓ Saved to zh.json

📊 Translation Statistics
   Total items     100
   Success         100
   Failed          0
   Duration        45s
   API calls       15 (5 translate + 5 reflect + 5 improve)

Generated .jta-terminology.json:

{
  "source_language": "en",
  "preserve_terms": ["GitHub", "API", "OAuth"],
  "consistent_terms": {
    "en": ["repository", "commit", "pull request"]
  }
}

Example 2: Incremental Update

$ jta en.json --to zh

📄 Loading source file...
✓ Source file loaded

🔍 Analyzing changes...
   New: 5 keys
   Modified: 2 keys
   Unchanged: 93 keys

Continue? [Y/n] y

🤖 Translating...
✓ Translation completed

📊 Translation Statistics
   Total items     7
   Success         7
   Filtered        93 included, 0 excluded (of 100 total)
   Duration        3s
   API calls       1

Example 3: Key Filtering

# Translate only settings and user sections
$ jta en.json --to ja --keys "settings.**,user.**"

📊 Translation Statistics
   Filtered        45 included, 55 excluded (of 100 total)
   Total items     45
   Success         45

Example 4: Multi-language Batch

# Translate to multiple languages at once
$ jta en.json --to zh,ja,ko,es,fr -y

Processing: zh ━━━━━━━━━━━━━━━━━━━━ 100% (100/100) ✓
Processing: ja ━━━━━━━━━━━━━━━━━━━━ 100% (100/100) ✓
Processing: ko ━━━━━━━━━━━━━━━━━━━━ 100% (100/100) ✓
Processing: es ━━━━━━━━━━━━━━━━━━━━ 100% (100/100) ✓
Processing: fr ━━━━━━━━━━━━━━━━━━━━ 100% (100/100) ✓

✓ Successfully created 5 translation files

Example 5: CI/CD Integration

# .github/workflows/translate.yml
name: Auto-translate i18n files

on:
  push:
    paths:
      - 'locales/en.json'

jobs:
  translate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v3
      
      - name: Install Jta
        run: |
          # Option 1: Using Homebrew (Linux)
          brew tap hikanner/jta
          brew install jta
          
          # Option 2: Using Go
          # go install github.com/hikanner/jta/cmd/jta@latest
          
          # Option 3: Download binary
          # curl -L https://github.com/hikanner/jta/releases/latest/download/jta-linux-amd64 -o jta
          # chmod +x jta
          # sudo mv jta /usr/local/bin/
      
      - name: Translate
        env:
          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
        run: |
          jta locales/en.json --to zh,ja,ko -y
      
      - name: Commit translations
        run: |
          git config user.name "Translation Bot"
          git config user.email "[email protected]"
          git add locales/*.json
          git commit -m "chore: update translations" || exit 0
          git push

🛠 Configuration

Environment Variables

# AI Provider API Keys
export OPENAI_API_KEY=sk-...
export ANTHROPIC_API_KEY=sk-ant-...
export GEMINI_API_KEY=...

Command-line Options

Flags:
  --to string                  Target language(s), comma-separated (required for translation)
  --list-languages             List all supported languages and exit
  --provider string            AI provider (openai, anthropic, gemini) (default "openai")
  --model string               Model name (uses default if not specified)
  --api-key string             API key (or use environment variable)
  --source-lang string         Source language (auto-detected from filename if not specified)
  -o, --output string          Output file or directory
  --terminology-dir string     Terminology directory (default ".jta/")
  --skip-terminology           Skip term detection (use existing terminology)
  --no-terminology             Disable terminology management completely
  --redetect-terms             Re-detect terminology (use when source language changes)
  --incremental                Incremental translation (only translate new/modified content)
  --keys string                Only translate specified keys (glob patterns)
  --exclude-keys string        Exclude specified keys (glob patterns)
  --batch-size int             Batch size for translation (default 20)
  --concurrency int            Concurrency for batch processing (default 3)
  -y, --yes                    Non-interactive mode
  -v, --verbose                Verbose output

🔧 Troubleshooting

Common Issues

API Key Not Found

Error: OPENAI_API_KEY environment variable not set

Solution: Set the API key as an environment variable or pass it directly:

export OPENAI_API_KEY=sk-...
# Or
jta en.json --to zh --api-key sk-...

Translation Quality Issues

If translations are not meeting quality expectations:

Use a better model: Generally, newer/larger models provide better quality

# OpenAI
jta en.json --to zh --provider openai --model gpt-5

# Anthropic
jta en.json --to zh --provider anthropic --model claude-sonnet-4-5

# Gemini
jta en.json --to zh --provider gemini --model gemini-2.5-flash

Check terminology: Review and refine terminology files in .jta/

# Edit term definitions
vim .jta/terminology.json

# Edit translations
vim .jta/terminology.zh.json

Example terminology.json:

{
  "version": "1.0",
  "sourceLanguage": "en",
  "preserveTerms": ["YourBrand", "ProductName", "API"],
  "consistentTerms": ["important", "domain", "terms"]
}

Verify Agentic reflection is working: The two-step reflection (evaluate → improve) runs automatically. In verbose mode, you should see:

jta en.json --to zh --verbose

# Look for reflection output showing:
# - Step 2: Reflection (LLM evaluating quality)
# - Step 3: Improvement (LLM applying suggestions)
# - API calls: 3x per batch (translate + reflect + improve)

Format Elements Lost in Translation

The format protector should automatically preserve placeholders, but if you notice issues:

Check the format instructions in verbose mode
Verify your placeholders follow standard patterns: {var}, {{var}}, %s, %d
Report non-standard formats as issues

Rate Limit Errors

Error: Rate limit exceeded

Solution: Reduce concurrency and batch size:

jta en.json --to zh --concurrency 1 --batch-size 10

Large File Handling

For files with 1000+ keys:

# Process in smaller batches with lower concurrency
jta large.json --to zh --batch-size 10 --concurrency 2

# Or filter by sections
jta large.json --to zh --keys "section1.**"
jta large.json --to zh --keys "section2.**"

Performance Tips

Batch Size: Larger batches (20-50) are more efficient but use more tokens per request
Concurrency: Higher concurrency (3-5) speeds up translation but may hit rate limits
Incremental Mode: Always use incremental translation for updates (automatic)
Provider Selection: Choose based on your needs:
- Quality priority: Use latest/largest models from any provider
- Speed priority: Use faster models like GPT-3.5 Turbo or Gemini Flash
- Cost priority: Compare pricing across providers and choose smaller models
- Balance: GPT-4o, Claude 3.5 Sonnet, or Gemini 1.5 Pro offer good balance

Debug Mode

Enable verbose output to see detailed execution:

jta en.json --to zh --verbose

# You'll see:
# - Provider initialization
# - Batch processing details
# - Reflection engine decisions
# - API call statistics
# - Format validation reports

❓ FAQ

Q: How much does it cost to translate a typical i18n file?

A: For a 100-key file using OpenAI GPT-4o with Agentic reflection (3x API calls):

First translation: ~$0.15-0.30 (including reflection)
Incremental updates: ~$0.03-0.06 (only new/modified keys)
Without reflection (basic translate only): ~$0.05-0.10
Trade-off: 3x cost for significantly higher quality through AI self-evaluation and improvement

Q: Can I translate offline or use my own models?

A: Currently, Jta requires an internet connection and uses cloud AI providers (OpenAI, Anthropic, Google Gemini).

Q: Does Jta support variables inside translated strings?

A: Yes! All standard placeholder formats are automatically preserved:

{variable}, {{count}} (i18next, Vue I18n)
%s, %d, %(name)s (printf-style)
,  (HTML tags)

Q: How do I handle custom terminology?

A: Edit .jta-terminology.json manually:

{
  "source_language": "en",
  "preserve_terms": ["MyApp", "SpecialFeature"],
  "consistent_terms": {
    "en": ["user", "account", "settings"]
  }
}

Then run translation with --skip-terms to use your custom dictionary.

Q: Can I review translations before saving?

A: Currently, translations are saved automatically. For manual review:

Use --output to save to a separate file
Review and edit the output
Copy to your actual locale file when satisfied

Q: What languages are supported?

A: Jta currently supports 27 languages with full metadata:

European: English, Spanish, French, German, Italian, Portuguese, Russian, Dutch, Polish, Turkish
Asian: Chinese (Simplified/Traditional), Japanese, Korean, Thai, Vietnamese, Indonesian, Malay, Hindi, Bengali, Sinhala, Nepali, Burmese
Middle Eastern (RTL): Arabic, Persian, Hebrew, Urdu

To see the complete list with flags and native names:

jta --list-languages

Jta also supports any other language that your chosen AI provider supports - just use the standard language code (e.g., sv for Swedish, da for Danish).

Q: How is this different from other translation tools?

A: Jta uses an agentic reflection mechanism that goes beyond traditional translation:

AI Self-Review: The AI doesn't just translate - it critically evaluates its own work across accuracy, fluency, style, and terminology, then refines it based on its expert analysis
Dynamic Quality Control: Instead of static post-processing rules, the AI generates contextual, specific improvements for each piece of content
Intelligent Context: Automatically detects and maintains domain terminology, understands cultural nuances, and preserves technical formats
Incremental Intelligence: Translates only new or modified content, saving 80-90% on API costs for updates
Production-Ready: Built with Go for reliability, performance, and robust error handling

🤝 Contributing

Contributions are welcome! Please read our Contributing Guide for details.

Development Setup

# Clone the repository
git clone https://github.com/hikanner/jta.git
cd jta

# Install dependencies
go mod download

# Run tests
go test ./...

# Build
go build -o jta cmd/jta/main.go

# Run locally
./jta examples/en.json --to zh

📄 License

MIT License - see LICENSE for details.

🙏 Acknowledgments

Inspired by Andrew Ng's Translation Agent
Built with official AI provider SDKs:
Powered by:
- Cobra for CLI
- Lipgloss for beautiful terminal output
- Sonic for fast JSON parsing

📞 Support

🐛 Bug Reports: GitHub Issues
💬 Discussions: GitHub Discussions
📖 Documentation: Wiki
⭐ Star us: If you find Jta useful, give us a star on GitHub!

Made with ❤️ by the Jta team

Jta - Making i18n translation intelligent, reliable, and effortless.

Name		Name	Last commit message	Last commit date
Latest commit History 76 Commits
.github/workflows		.github/workflows
cmd/jta		cmd/jta
docs		docs
examples		examples
internal		internal
skills		skills
test/integration		test/integration
.gitignore		.gitignore
.goreleaser.yml		.goreleaser.yml
CHANGELOG.md		CHANGELOG.md
CONTRIBUTING.md		CONTRIBUTING.md
Dockerfile		Dockerfile
LICENSE		LICENSE
Makefile		Makefile
README.md		README.md
README.zh-CN.md		README.zh-CN.md
go.mod		go.mod
go.sum		go.sum

License

ckanner/jta

Folders and files

Latest commit

History

Repository files navigation