llm-utils

Shared LLM utility library for text-only LLM calls.

Features

• Unified API for OpenAI, Anthropic (Claude), and Google (Gemini)
• Auto-detection of provider from environment variables
• Simple, consistent interface across providers
• Error handling and retries
• JSON extraction utilities

Installation

npm install @arclabs561/llm-utils

Usage

Basic Usage

import { callLLM, detectProvider } from '@arclabs561/llm-utils';

// Auto-detect provider from environment
const provider = detectProvider();
if (provider) {
  const response = await callLLM('Your prompt here', provider.provider, provider.apiKey);
  console.log(response);
}

Tiered Model Selection

The library supports two tiers: simple (fast/cheap) and advanced (higher quality).

import { callLLM, detectProvider, MODEL_TIERS } from '@arclabs561/llm-utils';

const provider = detectProvider();

// Use simple models (default) - fast and cheap
const fastResponse = await callLLM('Prompt', provider.provider, provider.apiKey, {
  tier: 'simple' // gpt-4o-mini, claude-haiku, gemini-flash
});

// Use advanced models - higher quality, slower, more expensive
const qualityResponse = await callLLM('Prompt', provider.provider, provider.apiKey, {
  tier: 'advanced' // gpt-4o, claude-sonnet, gemini-pro
});

Using LLMClient Class

import { LLMClient } from '@arclabs561/llm-utils';

// Simple tier (default)
const client = new LLMClient({ tier: 'simple' });
const response = await client.complete('Your prompt');

// Advanced tier
const advancedClient = new LLMClient({ tier: 'advanced' });
const qualityResponse = await advancedClient.complete('Your prompt');

Environment Variables

The library auto-detects providers from these environment variables:

• GEMINI_API_KEY - Google Gemini
• OPENAI_API_KEY - OpenAI
• ANTHROPIC_API_KEY - Anthropic Claude
• VLM_PROVIDER - Explicit provider selection (gemini, openai, claude)
• API_KEY - Fallback (defaults to gemini)

License

MIT