@gringow/gringow

AI-powered translation library for template strings using multiple LLM providers via multi-llm-ts. Gringow provides smart caching, deterministic content hashing, and seamless integration for modern JavaScript/TypeScript projects.

Features

• 🌍 Multi-language translation using LLM providers
• 🏷️ Tagged template literal syntax for inline translations
• 💾 Smart caching with content-based hashing
• 🔌 Multiple LLM providers (OpenAI default, Anthropic, Gemini, Groq, Mistral, Ollama, and more)
• ⚙️ Config-driven behavior through gringow.config.*
• 📦 Optimized builds via Vite/Next.js plugins

Installation

npm install @gringow/gringow
# or
pnpm add @gringow/gringow
# or
yarn add @gringow/gringow

Quick Start

import Gringow from '@gringow/gringow'

// Simple string translation
const { cacheId, translation } = await Gringow`Hello world`
console.log(translation.translation)
// => { "pt-BR": "Olá mundo", "fr-CA": "Bonjour le monde" }

// Template string with dynamic values
const name = 'John'
const { translation: result } = await Gringow`Welcome, ${name}!`
console.log(result.translation)
// => { "pt-BR": "Bem-vindo, John!", "fr-CA": "Bienvenue, John!" }

Usage Examples

Basic Translation

import Gringow from '@gringow/gringow'

// Translate a static string
const { translation } = await Gringow`Hello world`
console.log(translation.translation['pt-BR']) // "Olá mundo"

Dynamic Content

// Template literals are flattened before hashing - dynamic values preserved
const user = { name: 'Alice', role: 'admin' }
const { translation } = await Gringow`User ${user.name} has ${user.role} privileges`

console.log(translation.translation['pt-BR'])
// => "Usuário Alice tem privilégios de admin"

Cache Management

import Gringow, { rebuild, checkCache } from '@gringow/gringow'

const { cacheId } = await Gringow`Hello world`

// Check if translation exists in cache
const cached = await checkCache(cacheId)
console.log(cached) // GringowCacheItem or undefined

// Force re-translation by rebuilding cache entry
const refreshed = await rebuild(cacheId)
console.log(refreshed.translation.translation)

Browser Usage

For client-side interpolation without Node.js dependencies:

import { interpolateTemplate } from '@gringow/gringow/browser'

const template = 'Hello ${name}, you have ${count} messages'
const result = interpolateTemplate(template, ['Alice', 5])
console.log(result) // "Hello Alice, you have 5 messages"

Configuration

Gringow uses cosmiconfig to load configuration from your project root. Supported formats: gringow.config.js, gringow.config.mjs, gringow.config.json, or a gringow field in package.json.

Default Configuration

// gringow.config.js
export default {
  debug: false,
  log: { level: 'info' },
  cache: {
    dir: './gringow',
    file: 'gringow.json',
    prefix: 'grw',
    write: true,
  },
  sourceLanguage: 'en-US',
  defaultLanguage: 'pt-BR',
  languages: ['pt-BR', 'fr-CA'],
  llm: {
    provider: 'openai',
    apiKey: process.env.OPENAI_API_KEY,
    model: 'gpt-4o-mini',
    temperature: 1.0,
    sleep: 0,
  },
  globby: './src/**/*.{js,jsx,ts,tsx}',
}

Example: Using Gemini

// gringow.config.js
export default {
  llm: {
    provider: 'gemini',
    apiKey: process.env.GEMINI_API_KEY,
    model: 'gemini-2.0-flash',
    prePromptAppend: 'Please respond in a friendly tone.',
  },
  languages: ['es-ES', 'de-DE', 'ja-JP'],
}

Example: Using Anthropic

// gringow.config.js
export default {
  llm: {
    provider: 'anthropic',
    apiKey: process.env.ANTHROPIC_API_KEY,
    model: 'claude-3-5-sonnet-20241022',
    temperature: 0.7,
  },
}

Prompt Customization

Fine-tune LLM behavior with prompt options:

export default {
  llm: {
    provider: 'openai',
    apiKey: process.env.OPENAI_API_KEY,
    prePrompt: 'Translate the following text:', // Default prompt prefix
    postPrompt: 'Ensure translations are natural.', // Additional context
    prePromptAppend: 'Use formal language.', // Appended to pre-prompt
    postPromptPrepend: 'Maintain original formatting.', // Prepended to post-prompt
    prePromptReplace: 'Custom instruction here', // Replaces entire pre-prompt
  },
}

LLM Providers

OpenAI (Default)

Uses OpenAI through multi-llm-ts. Requires API key.

// gringow.config.js
export default {
  llm: {
    provider: 'openai',
    apiKey: process.env.OPENAI_API_KEY,
    model: 'gpt-4o-mini',
  },
}

Pros: Broad model availability, simple default setup
Cons: API costs, requires internet

Anthropic

Uses Claude models via Anthropic API. Requires API key.

// gringow.config.js
export default {
  llm: {
    provider: 'anthropic',
    apiKey: process.env.ANTHROPIC_API_KEY,
    model: 'claude-3-5-sonnet-20241022',
  },
}

Pros: High-quality translations, fast
Cons: API costs, requires internet

Gemini

Uses Google's Gemini models. Requires API key.

// gringow.config.js
export default {
  llm: {
    provider: 'gemini',
    apiKey: process.env.GEMINI_API_KEY,
    model: 'gemini-2.0-flash',
  },
}

Pros: Fast, cost-effective
Cons: API costs, requires internet

API Reference

Core Functions

Gringow(strings, ...values): Promise<GringowResponse>

Main translation function. Accepts tagged template literals or plain strings.

const result = await Gringow`Hello ${name}`

Returns: Promise<GringowResponse> with { cacheId, translation }

rebuild(cacheId: string): Promise<GringowResponse>

Re-translates a cached entry using the current LLM configuration.

const refreshed = await rebuild('grw_abc123')

Cache Helpers

import {
  checkCache,      // Check if cacheId exists
  getCache,        // Get entire cache object
  getOrCreateCache,// Get cache or create empty one
  writeCacheToDisk,// Manually write cache entry
  deleteCacheItem, // Remove cache entry
  clearCache,      // Clear all cache entries
} from '@gringow/gringow'

Utilities

import {
  createCacheId,         // Generate cache ID from string
  flatTemplateString,    // Flatten template literal to string
  awaitedConfig,         // Access resolved configuration
  isLocalConfigFound,    // Check if local config exists
  getConfigSource,       // Get config file path
} from '@gringow/gringow'

import { interpolateTemplate } from '@gringow/gringow/browser'

Type Definitions

type GringowTranslation = Record<string, string>

type GringowCacheItem = {
  original: string
  translation: GringowTranslation
  append?: string
  prepend?: string
  rebuild?: boolean
}

type GringowResponse = {
  cacheId: string
  translation: GringowCacheItem
}

How It Works

1. Input Processing: Template strings are flattened to plain text (preserving dynamic values)
2. Cache ID Generation: Content-based hash created using createCacheId()
3. Cache Lookup: Check ./gringow/gringow.json for existing translation
4. LLM Query: If not cached, request translation from configured provider
5. Response Parsing: Parse provider response to GringowCacheItem format
6. Cache Storage: Write to disk when cache.write is true

Example cache structure:

{
  "grw_a1b2c3": {
    "original": "Hello world",
    "translation": {
      "pt-BR": "Olá mundo",
      "fr-CA": "Bonjour le monde"
    }
  }
}

Development

# Install dependencies
pnpm install

# Build the package
pnpm run build

# Run tests
pnpm run test

# Watch mode
pnpm run test:watch

Related Packages

Gringow provides a complete ecosystem for AI-powered translation:

• @gringow/gringow - Core translation library
• @gringow/gringow-vite - Vite plugin for build-time extraction
• @gringow/gringow-react - React hooks and components
• @gringow/gringow-shadow - Web Component layer
• @gringow/gringow-cli - CLI tool
• @gringow/gringow-nextjs - Next.js plugin (experimental)

Related Resources

• Documentation
• GitHub Repository
• Examples

License

MIT © Renato Gaspar

Questions or Issues? Open an issue on GitHub