Yet another node API client library for interacting with AI providers using a common interface.

I wanted my own so I knew how it worked

npm run ask mistral 'In brief, how many AGIs will it take to change a lightbulb?'
...
Using API: mistral
Model: default
Prompt: In brief, how many AGIs will it take to change a lightbulb?
Using mistral key from: .env file

...it's uncertain how many would be needed to change a lightbulb...

npm run ask groq 'What is your name, and which model are you?'
...
Using API: groq
Model: default
Prompt: What is your name, and which model are you?

I'm an artificial intelligence model known as Llama. Llama stands for "Large Language Model Meta AI."

import { OpenAI, Claude, KeyManager } from 'hyperdata-clients';

Working for me against :

• Ollama (local)
• Mistral (free & speedy, needs API key)
• Groq (fast inference, needs API key)
• OpenAI (requires $s and API key)

Various other clients sketched out, will likely need tweaking.

Only tested on recent Ubuntu.

There's a very basic CLI for checking the thing (see below), also runnable hardcoded examples eg.

node examples/MistralMinimal.js
node examples/GroqMinimal.js

import { createClient, createEmbeddingClient } from 'hyperdata-clients';

// For API clients
const openAIClient = await createClient('openai', { apiKey: 'your-api-key' });

// For embedding clients
const nomicClient = await createEmbeddingClient('nomic', { apiKey: 'your-nomic-key' });
const ollamaEmbedding = await createEmbeddingClient('ollama');

// For MCP clients (untested)
const mcpClient = await createClient('mcp', { /* mcp config */ });

• Support for multiple AI providers
• Dedicated embedding model support
• Environment-based configuration
• Secure API key management
• Consistent interface across providers
• Type definitions included
• Extensive test coverage

• Ollama
• OpenAI
• Claude
• Mistral
• Groq
• Perplexity
• HuggingFace

• Nomic (via Atlas API)
• Ollama (local embeddings)

Prequisites : recent node

git clone https://github.com/danja/hyperdata-clients.git
cd hyperdata-clients
npm install

Really minimal for testing purposes :

# Basic usage
npm run ask [provider] [options] "your prompt"

# or more directly
node examples/minimal.js [provider] [options] "your prompt"

# Mistral
npm run ask mistral --model 'open-codestral-mamba' 'tell me about yourself'

# Groq (fast inference)
npm run ask groq --model 'llama-3.1-8b-instant' 'explain quantum computing'

# Example with Ollama running locally, it'll default model to qwen2:1.5b
npm run ask ollama 'how are you?'

# requires an API key
node examples/minimal.js openai 'what are you?'

The library provides dedicated support for text embeddings through specialized embedding clients:

import { createEmbeddingClient } from 'hyperdata-clients';

// Using Nomic Atlas API (requires NOMIC_API_KEY)
const nomicClient = await createEmbeddingClient('nomic');
const embeddings = await nomicClient.embed([
    'The quick brown fox jumps over the lazy dog',
    'Artificial intelligence is transforming technology'
]);

// Using local Ollama (requires Ollama running with nomic-embed-text-v1.5)
const ollamaClient = await createEmbeddingClient('ollama');
const singleEmbedding = await ollamaClient.embedSingle('Hello world');

// Example output: embeddings are arrays of numbers
console.log(`Generated ${embeddings.length} embeddings`);
console.log(`Each embedding has ${embeddings[0].length} dimensions`);

# Run the embedding demo
node examples/embedding.js

The embedding clients support:

• Batch processing: Embed multiple texts at once
• Single text convenience: embedSingle() method for individual texts
• Model selection: Custom model via options
• Error handling: Consistent error handling across providers

![docs/images/structure.png]

Comprehensive API documentation is available in the docs directory. To generate or update the documentation:

# Install dependencies (if not already installed)
npm install

# Generate documentation
npm run docs

# The documentation will be available in docs/jsdoc/index.html

The documentation includes:

• API reference for all components
• Getting started guide
• Code examples
• Configuration options

The project includes an extensive test suite to ensure reliability and compatibility across different providers. To run the tests:

# Run all tests
npm test

# Run tests with coverage report
npm run coverage

# Run tests in watch mode during development
npm run test:ui

Test coverage reports are generated in the coverage directory after running the coverage command. This includes:

• Line coverage
• Function coverage
• Branch coverage

Contributions are welcome! Please ensure all tests pass and add new tests for any new features or bug fixes.

MIT License