vinpac/pipr

PIPR - Prepare Input to Prompt and Resolve

PIPR is a library that helps you generate conversational AI prompts with ease. Pipr can be used to generate natural language responses to specific prompts by calling the OpenAI GPT-3 API. Pipr is designed to give better DX on writing prompt functions

Installation

Install pipr using npm or yarn:

npm install @pipr/core

yarn add @pipr/core

Usage

Here is an example usage of Pipr:

const getAge = createPipr()
  .input(
    z.object({
      name: z.string(),
      age: z.number(),
    })
  )
  .prepare(({ age }) => {
    return {
      ...input,
      age: age + 1,
    };
  })
  .prompt({
    system: 'You will remember everythin I say',
    user: ({ name }) => `The age of ${input.name} is:`,
  })
  .history(({ input }) => [
    {
      user: `John is ${input.age - 1} years old`,
      assistant: 'Nice to meet you, John!',
    },
    { user: `Alice is ${input.age + 1} years.`, assistant: 'Hello Alice!' },
  ])
  .resolve(ctx => {
    return ctx.blocks[0].content;
  });

const age = await getAge({ name: 'Alice', age: 44 });
console.log(age); // 46

.input

The .input method sets the schema for the input data. The schema should be defined using the zod library. The method returns a prompter instance that can be used to set the prompt configuration.

pipr.input(z.object({ name: z.string(), age: z.number() }));

.prepare

The .prepare method sets a preparer function that will be called before the prompt is generated. The preparer function takes the raw input data as a parameter and returns a prepared input data that will be used to generate the prompt. This method can be used to fetch async data needed to add to the prompt.

pipr.input(schema).prepare(async rawInput => {
  return {
    name: rawInput.name.toUpperCase(),
    age: rawInput.age * 2,
  };
});

.prompt

The .prompt method sets the prompt configuration. The configuration is an object with user and system properties that represent the user's input and the AI's response, respectively. The properties can be set to a string or a function that returns a string.

pipr.input(schema).prompt({
  user: 'What is your name?',
  system: "You're best greater",
});

.history

The .history method sets a function that will be called to generate a history for the prompt. The function takes a promptify and input as a parameter and should return an array of prompt examples. Prompt examples are objects with a user and an assistant property that represent the user's input and the AI's response, respectively.

pipr
  .input(schema)
  .prompt({
    system: 'Hi there! What can I do for you today?',
    user: ({ name }) => `My name is ${name}. What is your name?`,
  })
  .history(async ({ promptify, prepare }) => {
    const prepared = await prepare({
      name: 'John',
      age: 30,
    });

    return [
      {
        user: promptify(prepared).user, // My name is John. What is your name?
        assistant: 'Hello John! My name is ChatGPT. How can I help you today?',
      },
    ];
  });

.resolve

The .resolve method is called after the request is sent to the Open AI API. It takes the OpenAI API responded and resolves the value to return.