Navigating the Library

This page describes the Swift AI SDK structure. It differs from the original TypeScript SDK which includes web-specific UI libraries.

The Swift AI SDK is a comprehensive toolkit for building AI applications in Swift across all Apple platforms and server-side Swift. This page explains the structure and available components.

What’s Included

The Swift AI SDK ports the AI SDK Core from the TypeScript AI SDK, providing:

• AI SDK Core (SwiftAISDK): Unified API for generating text, structured objects, and tool calls with LLMs
• Provider System: Pre-built integrations for OpenAI, Anthropic, Google, Groq, and OpenAI-compatible providers
• Foundation Types (AISDKProvider): Language model interfaces, provider protocols, and shared types
• Utilities (AISDKProviderUtils): HTTP clients, JSON handling, schema validation, retry logic
• Zod Adapter (AISDKZodAdapter): Zod-like schema DSL for structured data validation

What’s Not Included

The Swift AI SDK focuses on core AI functionality and does not include web-specific libraries:

• AI SDK UI (React/Svelte/Vue hooks like useChat, useCompletion) - JavaScript/TypeScript only
• AI SDK RSC (React Server Components utilities) - Next.js specific

If you’re building hybrid applications with Swift backends and JavaScript frontends, you can use the TypeScript AI SDK UI with Swift AI SDK Core on the server.

Package Structure

The Swift AI SDK is organized into focused packages:

Package	Purpose	Dependencies
AISDKProvider	Foundation types and protocols (LanguageModelV3, etc.)	None
AISDKProviderUtils	HTTP, JSON, schema, validation, retry utilities	AISDKProvider
SwiftAISDK	Core API (generateText, streamText, generateObject)	Provider, ProviderUtils
OpenAIProvider	OpenAI integration (GPT-4, GPT-3.5, embeddings, etc.)	Provider, ProviderUtils
AnthropicProvider	Anthropic Claude integration	Provider, ProviderUtils
GoogleProvider	Google Gemini integration	Provider, ProviderUtils
GroqProvider	Groq integration	Provider, ProviderUtils
OpenAICompatibleProvider	OpenAI-compatible API integration	Provider, ProviderUtils
AISDKZodAdapter	Zod-like schema DSL for structured data	AISDKProvider
EventSourceParser	Server-Sent Events (SSE) parsing (internal utility)	None

Platform Compatibility

Swift AI SDK works across all Apple platforms and server-side Swift:

Platform	Support
iOS / iPadOS	✓
macOS	✓
watchOS	✓
tvOS	✓
Server-side Swift	✓
Linux (Swift on Linux)	✓

Minimum versions:

• macOS 13+
• iOS 16+
• watchOS 9+
• tvOS 16+

Core Features

Text Generation

import SwiftAISDK
import OpenAIProvider

let result = try await generateText(
  model: openai("gpt-4o"),
  prompt: "Write a short story about a cat."
)
print(result.text)

Streaming

let stream = try await streamText(
  model: openai("gpt-4o"),
  prompt: "Count to 10."
)

for try await textPart in stream.textStream {
  print(textPart, terminator: "")
}

Structured Data

import AISDKZodAdapter

let schema = z.object([
  "name": z.string(),
  "age": z.number()
])

let result = try await generateObject(
  model: openai("gpt-4o"),
  schema: schema,
  prompt: "Generate a person profile."
)
print(result.object)

Tool Calling

let result = try await generateText(
  model: openai("gpt-4o"),
  tools: [
    "getWeather": tool(
      description: "Get weather for a location",
      parameters: z.object(["location": z.string()])
    ) { args in
      return "Sunny, 72°F"
    }
  ],
  prompt: "What's the weather in San Francisco?"
)

Embeddings

let result = try await embed(
  model: openai.embedding("text-embedding-3-small"),
  value: "Hello, world!"
)
print(result.embedding) // [0.123, -0.456, ...]

Getting Started

Choose a quickstart based on your environment:

• iOS & macOS Apps - Building native apps with SwiftUI or UIKit
• Server (Vapor) - Building server-side Swift APIs
• CLI - Building command-line tools

Integration with Web Frontends

If you’re building a web application with a Swift backend:

1. Backend (Swift): Use Swift AI SDK Core for LLM calls, embeddings, and structured data generation
2. Frontend (JavaScript/TypeScript): Use AI SDK UI for React/Vue/Svelte chat interfaces

The two SDKs are designed to work together seamlessly through HTTP APIs.

Documentation Structure

• Getting Started: Platform-specific quickstarts and setup guides
• AI SDK Core: Core functions (generateText, streamText, generateObject, tools, embeddings)
• Agents: Building autonomous AI agents with workflow patterns
• Zod Adapter: Zod-like schema DSL for structured data validation
• Foundations: Understanding providers, models, prompts, tools, and streaming

Next Steps

Ready to start building? Pick a quickstart guide:

• iOS & macOS Quickstart
• Server (Vapor) Quickstart
• CLI Quickstart

Or explore the AI SDK Core overview to learn about core concepts.