Completion Library

Completion lets you:

• Monitor and debug your LLM prompts in production.
• Track the usage and cost of your LLM models.

Key Features

• Monitor Prompts: Track the prompts sent to your LLM models in real-time. We intentionally separate prompt arguments from the prompt template to ensure that you can monitor the prompts, automatically detect version changes, and debug issues on historical data.
• Cost Tracking: Track the usage and cost of your LLM models. We provide a detailed breakdown of the cost of each prompt, including the cost of the model, the cost of the prompt, and the total cost.
• Version Control: Manage the versions of your prompts. We automatically detect changes in the prompt template and arguments, allowing you to track the evolution of your prompts over time.

Installation

Install the Completion Node library using npm or yarn:

npm install @completionhq/completion-node

# or

yarn add @completionhq/completion-node

Usage

Below are examples demonstrating how to use Completion in your Node.js application.

Completion is LLM-agnostic, meaning it can be used with any LLM model, including those from OpenAI, Anthropic, Google, Mistral etc.

Basic Logging

import { Completion } from '@completionhq/completion-node';

const completion = new Completion({
  apiKey: 'abc123',
});

const messagesTemplate = [
  {
    role: 'User',
    message: 'Hello, how are you {name}?',
  },
  {
    role: 'Assistant',
    message: 'I am good, thank you. How can I help you today?',
  },
];

const promptArguments = {
  name: 'Alice',
};

const output = "Is there anything I can help you with today?";

await completion.log({
  templateName: 'greeting',
  messagesTemplate,
  promptArguments,
  output,
  model: 'gpt-3.5-turbo',
});

Log with compiled prompt

const compiledMessages = [
  {
    role: 'User',
    message: 'Hello, how are you Alice?',
  },
  {
    role: 'Assistant',
    message: 'I am good, thank you. How can I help you today?',
  },
];
await completion.log({
  templateName: 'greeting',
  compiledMessages,
  output,
  model: 'gpt-3.5-turbo',
});

Use compiled prompt to log

const compiledPrompt = "User: Hello, how are you Alice?\nAssistant: I am good, thank you. How can I help you today?";

await completion.log({
  templateName: 'greeting',
  compiledPrompt,
  output,
  model: 'gpt-3.5-turbo',
});