prompting v0.2.0

prompting

A batteries-included, model-agnostic prompt engineering library for Node.js and TypeScript.

Build dynamic, reusable prompts that output structured data that's ready to use in your application or API. Compatible with all text-based generative language models such as OpenAI GPT.

Features

• Intuitive, flexible Prompt builder
• Reusable prompt templates with variables
• Validated output in JSON or CSV
• Model-agnostic, extensible generation API
• Serializable to database and files

Installation

To install prompting, use npm:

npm install prompting

Examples

Simple text prompt

import {Prompt} from 'prompting';

const prompt = Prompt().text('What is your favorite animal?')

console.log(prompt.toString(); // 'What is your favorite animal?'

Using template variables and default values

import {Prompt} from 'prompting';

const prompt = Prompt()
  .text('What is your favorite {{topic}}?')
  .defaults({topic: 'animal'});

prompt.toString(); // 'What is your favorite animal?'
prompt.vars({topic: 'color'}).toString(); // 'What is your favorite color?'

Generating prompt responses

The library also contains a flexible Generator class for generating responses to a Prompt. For convenience, the Generator.prompt() method creates a new prompt that is bound to the Generator instance and can be invoked by calling generate().

Here's an example using the OpenAIGenerator:

import {OpenAIGenerator} from 'prompting';

const gpt = new OpenAIGenerator({apiKey: 'my_api_key'});

const prompt = gpt.prompt().text('What is your favorite {{topic}}?');

const result = await prompt.generate({topic: 'color'});

The generate method returns a Promise that resolves to the model's response for the prompt.

Structured JSON data with validation

To output a structured object and validate the result automatically, construct your prompt using the schema method. The Prompt class leverages the power of JSON Schema and the battle-tested validation library ajv to validate the response.

const prompt = Prompt()
  .text('List {{num}} books by the author {{author}}.')
  .defaults({num: 3})
  .schema({
    type: 'array',
    items: {
      type: 'object',
      properties: {
        title: {type: 'string'},
        year: {type: 'string'},
      },
      required: ['title', 'year'],
    },
  });

const result = await prompt.generate({author: 'George Orwell'});

The generate method returns a Promise that resolves to the model's response if it matches the schema, or rejects with a validation error if the model's response doesn't match the schema.

TypeScript Support

The library supports strongly typed prompts, arguments, and return types when used with TypeScript. The Prompt class supports generics to specify the expected arguments and return type.

Here's an example:

import {Prompt} from 'prompting';

type BookVars = {author: string};
type Book = {title: string, year: string};

const prompt = Prompt<BookVars, Book>()
  .text('What is the most popular book by {{author}}?')
  .schema({
    type: 'object',
    properties: {
      title: {type: 'string'},
      year: {type: 'string'},
    },
    required: ['title', 'year'],
  });

const result: Book = await prompt.generate({author: 'George Orwell'});

In this example, the generate method takes an argument of type BookVars and returns a Promise that resolves to a Book object, or rejects with a validation error if the model fails to generate a valid response.

Prompt API

PromptOptions

Contributing

Contributions to prompting are welcome! To contribute, please fork the repository and make your changes, then submit a pull request.