SwiftUI Hooks

SwiftyAI includes three observable state helpers for SwiftUI apps:

AIChat for multi-turn chat with streaming assistant messages.
AICompletion for a single prompt and a single generated output.
SwiftyChat for async chat sends managed by your own task.

They are state objects, not UI kits. You still build the view.

Helper	State shape	Best for
`AIChat`	`messages`, `input`, `isLoading`, `error`	Support chat, copilots, conversational search, streaming assistants
`AICompletion`	`input` / `output`, `isLoading`, `error`	Rewrite boxes, title generators, one-shot summaries
`SwiftyChat`	`messages`, `isStreaming`, `error`	Async chat flows where the caller awaits `send(_:)`
Direct model init	Model instance passed at construction	Previews, tests, dependency injection
Model string init	`"provider/model"` plus `AI.configure`	App-wide provider setup

Chat State

import SwiftUI
import SwiftyAI
 
struct ChatView: View {
    @State private var chat = AIChat(
        model: "openai/gpt-4o-mini",
        systemPrompt: "Be concise.",
        maxMessages: 20
    )
 
    var body: some View {
        VStack {
            List(chat.messages) { message in
                Text(message.content)
            }
 
            HStack {
                TextField("Message", text: $chat.input)
                Button("Send") {
                    chat.send()
                }
                .disabled(chat.isLoading || chat.input.isEmpty)
            }
        }
    }
}

AIChat appends the user message immediately, streams into an assistant message, exposes isLoading, stores error, and supports stop() and reset().

Completion State

struct CompletionView: View {
    @State private var completion = AICompletion(
        model: "openai/gpt-4o-mini",
        options: GenerationOptions(temperature: 0.4)
    )
 
    var body: some View {
        VStack {
            TextEditor(text: $completion.input)
 
            Button("Generate") {
                completion.send()
            }
            .disabled(completion.isLoading)
 
            Text(completion.output)
        }
    }
}

AICompletion also exposes prompt and completion aliases for input and output.

Direct Models

Both helpers also accept direct model instances:

let model = OpenAICompatibleProvider(
    baseURL: "https://api.openai.com/v1",
    apiKey: openAIKey,
    model: "gpt-4o-mini"
)
 
let chat = AIChat(model: model, systemPrompt: "Answer as support.")

Use direct models when previews or tests should not depend on global AI.configure state.

Related docs

Read AIChat, AICompletion, and SwiftyChat for dedicated examples.