AI chat

LaunchSwift ships an AI chat UI (AIChatView) backed by AIState, with SSE streaming from backend routes.

Files

iOS

• ios/SwiftLaunch/Features/AI/AIState.swift
• ios/SwiftLaunch/Features/AI/Models/AIModels.swift
• ios/SwiftLaunch/Features/AI/Services/AIService.swift
• ios/SwiftLaunch/Features/AI/Views/AIChatView.swift
• ios/SwiftLaunch/Features/AI/AIChatAccessPolicy.swift

Backend

• backend/src/routes/chat.ts
• backend/src/routes/ai-proxy.ts
• backend/src/services/ai-proxy.ts
• backend/src/services/chat-store.ts

AIModel enum (current source)

enum AIModel: String, CaseIterable, Sendable {
    case gpt4 = "gpt-5-mini"
    case claude = "claude-sonnet-4-6"
    case gemini = "gemini-3-flash"
    case workersAI = "@cf/meta/llama-4-scout-17b-16e-instruct"
}

AIState defaults to .workersAI.

AIState API

@MainActor
@Observable
final class AIState {
    var conversations: [ConversationSummary] = []
    var messages: [ChatMessage] = []
    var selectedConversationID: String?
    var selectedModel: AIModel = .workersAI
    var isLoading = false
    var isStreaming = false
    var error: AIError?

    func loadConversations() async
    func loadConversation(id: String) async
    func sendMessage(content: String)
    func retryLastAction()
    func cancelStream()
}

Streaming protocol used by /api/chat routes

chat.ts emits SSE events:

Event	Data
conversation	{ id, title }
token	{ text }
done	{ conversationId }
error	{ message }

AIState consumes these events and updates the assistant message as tokens arrive.

Backend routes and behavior

backend/src/index.ts mounts:

• app.route("/api/chat", chatRoutes)
• app.route("/api/ai", aiProxyRoutes)

Effective endpoints

Method	Path	Auth required	Rate limited
POST	/api/chat	Yes (protectedRoute)	No
POST	/api/chat/:id/message	Yes	No
GET	/api/chat/history	Yes	No
GET	/api/chat/:id	Yes	No
DELETE	/api/chat/:id	Yes	No
POST	/api/ai/chat	Yes	Yes (createRateLimit)

Request payloads

/api/chat and /api/chat/:id/message

{
  "provider": "workers-ai",
  "model": "@cf/meta/llama-4-scout-17b-16e-instruct",
  "message": "Explain async/await in Swift",
  "systemPrompt": null
}

/api/ai/chat

Uses unified chat payload with messages array:

{
  "provider": "openai",
  "model": "gpt-5-mini",
  "stream": true,
  "messages": [
    { "role": "user", "content": "Hello" }
  ]
}

Provider configuration

Configured in backend/src/services/ai-proxy.ts:

• openai → OPENAI_API_KEY
• anthropic → ANTHROPIC_API_KEY
• gemini → GEMINI_API_KEY
• workers-ai → Cloudflare AI binding (AI) via WorkersAIProvider

Persistence

Conversation and message persistence for /api/chat* is handled by chat-store.ts:

• uses D1 when available
• falls back to in-memory store otherwise (e.g. test/non-D1 environments)