@llm-utils/providers

LLM provider registry with comprehensive model metadata and capabilities. Easily discover and manage models across 14+ providers.

Installation

pnpm add @llm-utils/providers
# or
npm install @llm-utils/providers

Features

• 14+ Provider Support - OpenAI, Anthropic, Google, Mistral, Groq, DeepSeek, xAI, and more
• Model Metadata - Context windows, capabilities, pricing, and release dates
• Capability Filtering - Find models with vision, tool calling, streaming, etc.
• Credential Management - Auto-detection from environment variables
• Type-Safe - Full TypeScript support with comprehensive types

Usage

Quick Start

import {
  getModelById,
  getModelsWithCapability,
  ProviderRegistry
} from '@llm-utils/providers';

// Get a specific model
const gpt4o = getModelById('gpt-4o');
console.log(gpt4o?.contextWindow); // 128000
console.log(gpt4o?.capabilities.vision); // true

// Find all models with vision capability
const visionModels = getModelsWithCapability('vision');
console.log(visionModels.length); // Multiple models

// Create a registry from environment
const registry = new ProviderRegistry();
registry.registerFromEnv();

// Find models across registered providers
const streamingModels = registry.findModelsWithCapability('streaming');

Provider Registry

import { ProviderRegistry, createProviderInstance } from '@llm-utils/providers';

// Create a registry
const registry = new ProviderRegistry();

// Register with explicit credentials
registry.register('openai', { apiKey: 'sk-...' });
registry.register('anthropic', { apiKey: 'sk-ant-...' });

// Or register all from environment
registry.registerFromEnv();

// Get a provider instance
const openai = registry.get('openai');
if (openai?.isConfigured()) {
  const models = openai.listModels();
  const visionModels = openai.listModelsWithCapability('vision');
}

Model Discovery

import {
  allModels,
  getModelsByProvider,
  getModelsWithCapability
} from '@llm-utils/providers';

// Get all models
console.log(allModels.length);

// Get models by provider
const anthropicModels = getModelsByProvider('anthropic');
const openaiModels = getModelsByProvider('openai');

// Filter by capability
const toolCallingModels = getModelsWithCapability('toolCalling');
const jsonModeModels = getModelsWithCapability('jsonMode');
const embeddingModels = getModelsWithCapability('embeddings');

Provider Configuration

import { getProvider, getAllProviders, isProviderConfigured } from '@llm-utils/providers';

// Get provider config
const openai = getProvider('openai');
console.log(openai.baseUrl); // https://api.openai.com/v1
console.log(openai.requiredEnvVars); // ['OPENAI_API_KEY']

// List all providers
const providers = getAllProviders();
providers.forEach(p => console.log(`${p.name}: ${p.description}`));

// Check if configured from environment
if (isProviderConfigured('anthropic')) {
  console.log('Anthropic is ready to use');
}

Supported Providers

| Provider | ID | Models | |----------|-----|--------| | OpenAI | openai | GPT-4o, GPT-4, o1, DALL-E, Embeddings | | Anthropic | anthropic | Claude 3.5 Sonnet, Claude 3 Opus, Haiku | | Google | google | Gemini 2.0, Gemini 1.5 Pro/Flash | | Mistral | mistral | Mistral Large, Small, Codestral | | Groq | groq | Llama 3.3, Mixtral, Gemma | | DeepSeek | deepseek | DeepSeek V3, R1 | | xAI | xai | Grok 2, Grok Vision | | Azure OpenAI | azure-openai | OpenAI models on Azure | | AWS Bedrock | aws-bedrock | Claude, Mistral on AWS | | Ollama | ollama | Local models | | Together AI | together-ai | Open models | | Replicate | replicate | Various models | | Hugging Face | huggingface | Inference API | | Cohere | cohere | Command, Embed |

Model Capabilities

Each model includes capability flags:

interface ModelCapabilities {
  chat: boolean;           // Text generation
  vision: boolean;         // Image input
  toolCalling: boolean;    // Function calling
  jsonMode: boolean;       // JSON output mode
  structuredOutput: boolean; // Schema-based output
  streaming: boolean;      // Stream responses
  embeddings: boolean;     // Vector embeddings
  imageGeneration: boolean; // Image creation
  audio: boolean;          // Audio I/O
}

API Reference

Types

• ProviderId - Provider identifier union type
• ModelMetadata - Model information including capabilities and pricing
• ProviderConfig - Provider configuration including base URL and env vars
• ProviderCredentials - API keys and connection options
• ProviderInstance - Configured provider with methods

Functions

• getModelById(id) - Get model by ID
• getModelsByProvider(provider) - Get all models for a provider
• getModelsWithCapability(cap) - Filter models by capability
• getProvider(id) - Get provider configuration
• getAllProviders() - List all provider configs
• isProviderConfigured(id) - Check env var configuration
• createProviderInstance(id, creds?) - Create provider instance

Classes

• ProviderRegistry - Manage multiple provider instances

License

MIT