Illustration of a shared AI assistant powering both Android and iOS devices, with connected user flows, synchronized prompts, and developer code samples bridging Swift and Kotlin.

Cross-Platform AI Agents: Building a Shared Gemini + Apple Intelligence Assistant

by

Developers are now building intelligent features for both iOS and Android — often using different AI platforms: Gemini AI on Android, and Apple Intelligence on iOS. So how do you build a shared assistant experience across both ecosystems?

This post guides you through building a cross-platform AI agent that behaves consistently — even when the underlying LLM frameworks are different. We’ll show design principles, API wrappers, shared prompt memory, and session persistence patterns.

📦 Goals of a Shared Assistant

  • Consistent prompt structure and tone across platforms
  • Shared memory/session history between devices
  • Uniform fallback behavior (offline mode, cloud execution)
  • Cross-platform UI/UX parity

🧱 Architecture Overview

The base model looks like this:


              [ Shared Assistant Intent Engine ]
                   /                    \\
      [ Gemini Prompt SDK ]         [ Apple Intelligence APIs ]
           (Kotlin + AICore)           (Swift + AIEditTask)
                   \\                    /
           [ Shared Prompt Memory Sync ]

Each platform handles local execution, but prompt intent and reply structure stay consistent.

🧠 Defining Shared Prompt Intents

Create a common schema:


{
  "intent": "TRAVEL_PLANNER",
  "data": {
    "destination": "Kerala",
    "duration": "3 days",
    "budget": "INR 10,000"
  }
}

Each platform converts this into its native format:

Apple Swift (AIEditTask)


let prompt = """
You are a travel assistant. Suggest a 3-day trip to Kerala under ₹10,000.
"""
let result = await AppleIntelligence.perform(AIEditTask(.generate, input: prompt))

Android Kotlin (Gemini)


val result = session.prompt("Suggest a 3-day trip to Kerala under ₹10,000.")

🔄 Synchronizing Memory & State

Use Firestore, Supabase, or Realm to store:

  • Session ID
  • User preferences
  • Prompt history
  • Previous assistant decisions

Send current state to both Apple and Android views for seamless cross-device experience.

🧩 Kotlin Multiplatform + Swift Interop

Use shared business logic for agents in Kotlin Multiplatform Mobile (KMM) to export common logic to iOS:


// KMM prompt formatter
fun formatTravelPrompt(data: TravelRequest): String {
    return "Plan a ${data.duration} trip to ${data.destination} under ${data.budget}"
}

🎨 UI Parity Tips

  • Use SwiftUI’s glass-like cards and Compose’s Material3 Blur for parity
  • Stick to rounded layouts, dynamic spacing, and minimum-scale text
  • Design chat bubbles with equal line spacing and vertical rhythm

🔍 Debugging and Logs

  • Gemini: Use Gemini Debug Console and PromptSession trace
  • Apple: Xcode AI Profiler + LiveContext logs

Normalize logs across both by writing JSON wrappers and pushing to Firebase or Sentry.

🔐 Privacy Considerations

  • Store session data locally with user opt-in for cloud sync
  • Mark cloud-offloaded prompts (on-device → server fallback)
  • Provide export history button with logs + summaries

✅ Summary

Building shared AI experiences across platforms isn’t about using the same LLM — it’s about building consistent UX, logic, and memory across SDKs.

🔗 Further Reading