@llmtr/cli

AI-powered translation CLI — translate text & files using your favourite LLM.

Translate text, files, or watch for changes and auto-retranslate. Supports multiple providers and multiple target languages in one run. Comes with an interactive TUI for real-time progress and a config wizard.

Install

npm install -g @llmtr/cli

Or use without installing:

npx @llmtr/cli translate "Hello" --lang zh-CN

Quick Start

# Interactive config wizard
llmtr config

# Translate text
llmtr translate "Good morning" --lang zh-CN,ja,fr

# Translate a file
llmtr translate --file README.md --lang zh-CN --output ./i18n

# Watch mode — re-translate whenever the file changes
llmtr watch content.md --lang zh-CN,de

# Short alias
llmtr t "Hello" --lang es

Commands

llmtr version (aliases: llmtr v, llmtr --version)

Print the current version.

llmtr version
llmtr v
llmtr --version

llmtr config

Interactive TUI wizard to create or edit a config file.

Options:
  -g, --global    Edit global config at ~/.llmtr.config.json
  -o, --output    Config file path (overrides default)

The wizard shows a live config card at the top that updates as you fill in each field.

llmtr translate (alias: llmtr t)

One-shot translation of text or a file.

Arguments:
  [text]            Text to translate

Options:
  -f, --file        Input file to translate
  -l, --lang        Target languages, comma-separated  (e.g. "zh-CN,ja,fr")
  -o, --output      Output directory for translated files
  -p, --provider    AI provider: openai | anthropic | google | mistral | deepseek
  -m, --model       Model name (uses provider default when omitted)
      --api-key     API key (overrides env var and config file)
      --prompt      One-time user prompt appended to the request
      --stdout      Print result to stdout (single language only, pipeable)
      --concurrency Max parallel translations (default: 3)

Examples:

# Translate inline text
llmtr t "Hello, world!" --lang zh-CN,ja

# Translate a file into two languages
llmtr translate --file docs/guide.md --lang zh-CN,ko --output docs/i18n

# Pipe output to another command
llmtr t "Hello" --lang fr --stdout | pbcopy

# Use a one-time prompt
llmtr t --file CHANGELOG.md --lang zh-CN --prompt "Use simplified technical language"

llmtr watch

Watch a file and re-translate automatically whenever it changes.

Arguments:
  <file>            File to watch (required)

Options:
  -l, --lang        Target languages, comma-separated
  -o, --output      Output directory for translated files
  -p, --provider    AI provider
  -m, --model       Model name
      --api-key     API key
      --prompt      One-time user prompt
      --debounce    Debounce delay in ms (default: 500)

Example:

llmtr watch src/content.md --lang zh-CN,de --output src/i18n

Configuration

llmtr searches for a config file starting from the current directory and walking up to ~. The first file found wins.

Config file names (searched in order):

• llmtr.config.json
• .llmtr.config.json
• .llmtrrc.json / .llmtrrc
• llmtr.config.js / .mjs / .ts
• package.json → "llmtr" key

Global config: ~/.llmtr.config.json

Config Schema

{
  "provider": "openai",
  "model": "gpt-4o",
  "apiKey": "sk-...",
  "sourceLanguage": "en",
  "targetLanguages": ["zh-CN", "ja", "fr"],
  "systemPrompt": "Translate naturally and preserve formatting.",
  "output": {
    "directory": "./i18n",
    "fileNamePattern": "{name}.{lang}{ext}"
  },
  "watch": {
    "debounce": 500
  }
}

fileNamePattern tokens: {name} (basename without ext), {lang}, {ext} (with dot).

API Key

API keys are resolved in this priority order:

1. --api-key CLI flag
2. apiKey in config file
3. Environment variable (see table below)

| Provider | Environment Variable | |---|---| | openai | OPENAI_API_KEY | | anthropic | ANTHROPIC_API_KEY | | google | GOOGLE_API_KEY | | mistral | MISTRAL_API_KEY | | deepseek | DEEPSEEK_API_KEY |

package.json Integration

{
  "llmtr": {
    "provider": "openai",
    "targetLanguages": ["zh-CN"]
  }
}

Proxy

llmtr automatically reads proxy settings from environment variables and applies them to all network requests (including LLM API calls).

Supported variables (checked in priority order):

HTTPS_PROXY=http://host:port       # HTTP/HTTPS proxy
https_proxy=http://host:port
ALL_PROXY=socks5://host:port       # SOCKS5 proxy (e.g. Clash)
all_proxy=socks5://host:port
HTTP_PROXY=http://host:port
http_proxy=http://host:port

To clear a proxy, use unset — not export VAR= (an empty string is silently ignored):

# ✓ correct
unset HTTPS_PROXY HTTP_PROXY ALL_PROXY https_proxy http_proxy all_proxy

# ✗ wrong — leaves the variable set to "", which some tools mishandle
export HTTPS_PROXY=

When a proxy is active, the TUI displays a subtle ↳ via socks5 127.0.0.1:7897 indicator below the header.

License

MIT © Moozon