Skip to Content

AI Functions

Elegant, type-safe AI functions that deliver predictable results

Overview

AI Functions is a TypeScript library that provides strongly-typed interfaces to AI capabilities, transforming unpredictable AI outputs into reliable, deterministic components for your applications. It enables you to create elegant, type-safe interfaces to AI capabilities that seamlessly integrate with your existing codebase.

Features

  • Dynamic AI Functions: Use any function name with the ai proxy
  • Template Literals: Generate content using template literals
  • No-Schema Support: Generate unstructured text responses without schema validation
  • Schema Support: Validate AI responses with Zod schemas and descriptions
  • List Generation: Generate lists of items with the list function
  • Model Configuration: Override model, temperature, and other settings
  • Environment Awareness: Automatically uses AI_GATEWAY environment variable when available
  • Multiple Invocation Patterns: Choose the syntax that fits your coding style
  • Type-Safe Interfaces: Get predictable, structured outputs that match your schemas
  • Elegant API Design: Clean, intuitive interfaces that prioritize developer experience

Installation

npm install ai-functions

Usage

Basic Usage

import { ai } from 'ai-functions'
 
// Generate text using template literals
const text = await ai`Write a short story about a robot`
 
// Generate a list of items
const items = await list`List 5 programming languages`
 
// Generate a list with async iterator support
// The list function automatically detects when it's used in a for-await-of loop
for await (const language of list`List 10 programming languages`) {
  console.log(`Processing language: ${language}`)
}

Creating Custom Functions with AI

The AI function provides an elegant way to define custom AI functions with type-safe schemas:

import { AI } from 'ai-functions'
 
// Define custom AI functions with type-safe schemas
export const ai = AI({
  // Book Proposal Generator
  createBookProposal: {
    title: 'proposed title of the book',
    subtitle: 'proposed subtitle of the book',
    author: 'name of the author',
    targetAudience: ['primary audience segments for the book'],
    marketAnalysis: 'analysis of the current market for this type of book',
    competitiveBooks: ['list of similar books in the market'],
    uniqueSellingPoints: ['what makes this book different and valuable'],
    keyTakeaways: ['main insights readers will gain'],
    marketingPotential: 'assessment of marketing opportunities',
    coverDescription: 'visual description of potential cover design',
    estimatedWordCount: 'approximate word count for the entire book',
    estimatedTimeToComplete: 'timeline for completing the manuscript',
    summary: 'one paragraph summary of the book concept',
  },
})

Dynamic Function Calls

import { ai } from 'ai-functions'
 
// Call any function name with parameters
const categories = await ai.categorizeProduct({
  name: 'Product name',
  description: 'Product description',
})
 
// Parameters are used to generate a schema for validation
const blogPost = await ai.writeBlogPost({
  title: 'Blog post title',
  keywords: 'comma, separated, keywords',
  tone: 'professional',
})

Schema Support

import { ai } from 'ai-functions'
import { z } from 'zod'
 
// Define a schema for validation
const schema = z.object({
  title: z.string(),
  content: z.string(),
  tags: z.array(z.string()),
})
 
// Generate content with schema validation
const post = await ai`Write a blog post about AI`({
  schema,
  temperature: 0.7,
})

Configuration

import { ai } from 'ai-functions'
 
// Override model, temperature, and other settings
const text = await ai`Write a poem about the ocean`({
  model: 'gpt-4o',
  temperature: 0.9,
  maxTokens: 500,
})

License

MIT

Business-as-Code