L10nMonster Anthropic Helper

This package provides integration between L10nMonster and Anthropic's Claude models via both direct API and Google Vertex AI.

Installation

npm install @l10nmonster/helpers-anthropic

Usage

Direct API Configuration

import { AnthropicAgent } from '@l10nmonster/helpers-anthropic';

const agent = new AnthropicAgent({
    id: 'claude-translator',
    model: 'claude-3-5-sonnet-latest',
    quality: 80,
    temperature: 0.1,
    maxTokens: 4096,
    apiKey: 'your-anthropic-api-key'  // Direct API access
});

Vertex AI Configuration

import { AnthropicAgent } from '@l10nmonster/helpers-anthropic';

const agent = new AnthropicAgent({
    id: 'claude-translator',
    model: 'claude-3-5-sonnet@20241022',
    quality: 80,
    temperature: 0.1,
    maxTokens: 4096,
    vertexProject: 'your-gcp-project-id',
    vertexLocation: 'us-central1'
});

Authentication

The AnthropicAgent supports two authentication methods:

Direct API

1. API Key: Obtain an API key from Anthropic Console
2. Set apiKey: Pass your API key in the configuration

Vertex AI

1. Service Account: Set up a service account with Vertex AI permissions
2. gcloud CLI: Run gcloud auth login and gcloud config set project YOUR_PROJECT_ID
3. Environment Variables: Set GOOGLE_APPLICATION_CREDENTIALS to point to your service account key file

Retry Behavior

The AnthropicAgent uses the native retry mechanism built into both the Anthropic SDK and Anthropic Vertex SDK:

• Automatic Retries: Both SDKs automatically handle retries with exponential backoff
• Configurable: Set maxRetries in the constructor to control retry attempts (passed directly to both SDKs)
• Smart Error Handling: Automatically retries on network errors, rate limits, and temporary service issues
• No Manual Implementation: Unlike other providers, this uses the SDK's native retry logic for better reliability
• Consistent Behavior: The same maxRetries value is used whether you're using direct API or Vertex AI

Configuration Options

• model (required): The Claude model to use
• quality (required): Quality score for translations (0-100)
• apiKey: Your Anthropic API key (for direct API access)
• temperature: Controls randomness (0.0-1.0, default: 0.1)
• maxTokens: Maximum output tokens (default: 4096)
• maxRetries: Maximum number of retries (passed to Anthropic SDK)
• vertexProject: GCP project ID (for Vertex AI, auto-detected if not provided)
• vertexLocation: GCP region (for Vertex AI, default: 'global')
• persona: Custom translator persona (optional)
• customSchema: Custom response schema (optional)

Testing

Run the test suite using Node.js built-in testing:

# From the package directory
npm test

# From the workspace root
npm test --workspace=helpers-anthropic

The test suite includes:

• Unit tests for all major functionality
• Integration tests with the LLMTranslationProvider inheritance
• Error handling and edge case scenarios
• Mock testing without external API calls

Requirements

• Node.js >= 22.11.0
• Google Cloud Project with Vertex AI API enabled
• Proper authentication setup for Google Cloud

License

MIT