Update SDK documentation for Kotlin Multiplatform, iOS v0.9.2, and add AI agent guides

[iamstuffed]

Summary

Comprehensive documentation updates for LEAP Android SDK (v0.9.7) and iOS SDK (v0.9.2), including new AI agent usage guides for both platforms.

Android SDK Updates (v0.9.7)

Core SDK Changes

• ✅ Updated to Kotlin Multiplatform architecture
• ✅ Distinguished platform-specific downloaders:
  • LeapModelDownloader for Android (WorkManager, notifications, background downloads)
  • LeapDownloader for cross-platform (iOS, macOS, JVM)
• ✅ Added comprehensive Android permission setup (notifications, foreground service)
• ✅ Updated all examples to use ViewModel architecture with proper lifecycle management
• ✅ Replaced deprecated Gson with kotlinx.serialization

Audio Format Requirements

• ✅ Detailed audio input specifications based on inference engine:
  • WAV format only (no MP3, AAC, OGG)
  • 16 kHz sample rate recommended (auto-resampled if different)
  • Mono channel required (stereo rejected)
  • Supports Float32, Int16, Int24, Int32 PCM encodings
• ✅ Audio output documentation (24 kHz from generation models)
• ✅ Audio recording examples for Android and iOS

New: Android AI Agent Usage Guide

• ✅ Complete reference covering installation, loading, and management
• ✅ Model loading patterns (automatic, download without loading, cross-platform)
• ✅ Model download management (status, size, removal, cancellation)
• ✅ Core classes documentation (ModelRunner, Conversation, ChatMessage, etc.)
• ✅ Generation patterns with ViewModel examples
• ✅ TOML version catalog examples
• ✅ Storage validation and error handling

iOS SDK Updates (v0.9.2)

Version Update

• ✅ Updated quick start guide to v0.9.2
• ✅ Updated Swift Package Manager and CocoaPods references

New: iOS AI Agent Usage Guide

• ✅ Complete reference matching Android guide structure
• ✅ Installation via SPM and CocoaPods
• ✅ Model loading methods (automatic download, separate download, custom manifest, local bundle)
• ✅ Core classes documentation (ModelRunner, Conversation, etc.)
• ✅ Generation patterns with ObservableObject ViewModel
• ✅ Model download management
• ✅ Multimodal input/output examples
• ✅ Function calling and structured output
• ✅ Conversation persistence with Codable

Updated Examples

• ✅ Updated cloud-ai-comparison.mdx to use current Leap.load() API
• ✅ Fixed enum cases (.User → .user)
• ✅ Updated to manifest-based model loading

Files Modified

Android

• leap/edge-sdk/android/android-quick-start-guide.mdx
• leap/edge-sdk/android/ai-agent-usage-guide.mdx (new)
• leap/edge-sdk/android/conversation-generation.mdx
• leap/edge-sdk/android/messages-content.mdx
• leap/edge-sdk/android/model-loading.mdx
• leap/edge-sdk/android/utilities.mdx
• leap/edge-sdk/android/cloud-ai-comparison.mdx

iOS

• leap/edge-sdk/ios/ios-quick-start-guide.mdx
• leap/edge-sdk/ios/ai-agent-usage-guide.mdx (new)
• leap/edge-sdk/ios/messages-content.mdx
• leap/edge-sdk/ios/cloud-ai-comparison.mdx

Navigation

• docs.json (added AI agent guides to navigation)

Breaking Changes

None - all changes are documentation updates only.

Testing

• ✅ Verified Mintlify dev server runs without errors
• ✅ All code examples match current SDK APIs (v0.9.7 Android, v0.9.2 iOS)
• ✅ Verified against LeapSDK-Examples and apollo-android projects
• ✅ Audio format requirements verified against inference engine source

Notes

• AI agent guides provide comprehensive SDK reference for AI assistants
• Both platform guides follow consistent structure for easier maintenance
• All examples use current best practices (ViewModel, proper lifecycle, kotlinx.serialization)
• Plugin versions verified against example projects

[iamstuffed]

⏺ Yes! This AI agent usage guide is extremely helpful for using the SDK. Here's why:

  What Makes It Effective                                                                                                                                                       
   
  ✅ Complete Information in One Place                                                                                                                                          ─
                                                                                                       
  - No need to search multiple pages or external resources
  - All core classes, methods, and patterns documented
  - Clear architecture diagram shows the flow

  ✅ Copy-Paste Ready Code Examples

  Every pattern has complete, working code:
  // I can literally copy this and adapt it
  class ChatViewModel(application: Application) : AndroidViewModel(application) {
      private val downloader = LeapModelDownloader(application)
      // ... complete working example
  }

  ✅ Critical Information Clearly Marked

  // CRITICAL: Use runBlocking(Dispatchers.IO), not viewModelScope.launch
  This prevents common mistakes without trial and error.

  ✅ Multiple Approaches Documented

  - Method 1: Automatic (most common)
  - Method 2: Separate download/load (specific use cases listed)
  - Method 3: Cross-platform

  I know when to use each approach.

  ✅ Imports Section

  import ai.liquid.leap.downloader.LeapModelDownloader
  import ai.liquid.leap.message.ChatMessage
  No guessing package names!

  ✅ Audio Requirements Are Crystal Clear

  Format: WAV (RIFF) - No MP3/AAC/OGG
  Sample Rate: 16 kHz recommended
  Channels: Mono (1 channel) REQUIRED - stereo rejected
  I won't waste time trying unsupported formats.

  ✅ Complete ViewModel Example

  Shows the entire lifecycle - not just fragments:
  - Model loading
  - Generation with cancellation
  - Proper cleanup in onCleared()
  - StateFlow for UI updates

  What I Could Do Without Looking Anything Up

  With just this guide, I can:
  1. ✅ Set up dependencies (Gradle or TOML)
  2. ✅ Add required permissions
  3. ✅ Load models (3 different ways)
  4. ✅ Generate text responses
  5. ✅ Handle multimodal input (vision, audio)
  6. ✅ Implement function calling
  7. ✅ Manage model downloads
  8. ✅ Handle errors properly
  9. ✅ Structure code with ViewModel
  10. ✅ Debug common issues

  Minor Suggestion

  Add one more section at the very top:
  ## Quick Start (30 seconds)
  1. Add dependencies: `implementation(libs.leap.sdk)`
  2. Add permissions to AndroidManifest.xml
  3. Copy the Complete ViewModel Example
  4. Call `loadModel()` then `sendMessage()`

  For users who just want to get running immediately.

  Bottom line: This guide gives me everything I need to implement LEAP SDK features confidently without external lookups. It's comprehensive yet concise - exactly what an AI
  agent needs.