Apiary SDK

Apiary is an open-source LLM Gateway for managing requests across leading providers. It routes all LLM requests through a single API endpoint, simplifying access to multiple models with a unified API, centralized management of features like routing, caching, and guardrails, and built-in observability.

Use the Apiary TypeScript SDK in your applications to easily send client requests to multiple LLM providers.

Supported Providers

• Anthropic
• Gemini
• OpenAI

Setup

Prerequisites

To send LLM requests using the Apiary SDK, you must have a running Apiary instance set up in your AWS account. Follow the setup instructions in our installation guide to easily deploy Apiary's infrastructure using our CLI tool.

The deployment process will provide an AiGatewayRestApiEndpoint URL and API key that you can use to send LLM requests with the Apiary SDK.

Installation

To install the Apiary SDK, run the following command:

npm install @apiary-gateway/sdk

Usage

Import the Apiary client and initialize an instance with your configuration details.

Call the chatCompletion method with a prompt (required). A threadID may also be provided to include conversation context with the prompt. Optionally, specify the LLM provider and model to override your configuration details for these fields.

import Apiary from '@apiary-gateway/sdk';

// Initialize the Apiary client
const client = new Apiary({
  apiKey: '<YOUR_API_KEY>',
  baseUrl: 'https://<YOUR_GATEWAY_URL>.com/dev',
  // Optionally include a userId to partition cache responses based on userId
  userId: '<userId>',
  // Optionally include a metadata object with your custom routing configuration groups
  metadata: {
    'user-type': 'pro-users'
  },
});

const main = async () => {
  const completion = await client.chatCompletion(
    // Example prompt
    'Do all bees make honey?',
    // Optionally include a `threadID` to manage conversation context 
    '1745267450008',
    // Optionally specify the provider and model to override your configuration defaults
    'openai',
    'gpt-4',
  );

  // Log the returned LLM response
  console.log(completion.response.text);

  // Log the threadID for the returned LLM response
  console.log(completion.threadID);
}

main();