feat(flutter): trim generated output and cleanup SDK

Author: sanchitmonga22

docs/BUILD_ORGANIZATION.md

@@ -80,15 +80,13 @@ App startup runs a multi-phase sequence in `initState` via `addPostFrameCallback
 1. **Eager .so loading** (Android only) — `DynamicLibrary.open()` on 6 `.so` files to preload before any SDK call
 2. **SDK init** — reads API key / base URL from secure storage (`KeychainHelper`); calls `RunAnywhere.initialize(...)` with or without credentials
 3. **Module registration** — guarded by a static `_modulesRegistered` flag to survive hot-reload. Registers: LlamaCpp (9 GGUF models), Genie NPU (Android/Snapdragon only, chip-conditional models), VLM (SmolVLM 500M), Sherpa STT/TTS (Whisper + Piper models), RAG embeddings (MiniLM), ONNX backend, RAG backend
-4. **Model refresh** — `ModelManager.shared.refresh()` notifies all listeners
 
 ### State Management
 
-Three patterns coexist:
+Two patterns coexist:
 
-1. **Provider** (app-wide) — only `ModelManager.shared` is in the `MultiProvider` tree at the root
-2. **Singleton ChangeNotifier + ListenableBuilder** (feature-level) — `ModelListViewModel.shared`, `ToolSettingsViewModel.shared`, `ConversationStore.shared`, `DeviceInfoService.shared` are accessed directly, not through Provider
-3. **Local setState** (per-screen ephemeral state) — recording flags, streaming text buffers, error messages
+1. **Singleton ChangeNotifier + ListenableBuilder** (feature-level) — `ModelListViewModel.shared`, `ToolSettingsViewModel.shared`, `ConversationStore.shared`, `DeviceInfoService.shared` are accessed directly
+2. **Local setState** (per-screen UI state) — recording flags, streaming text buffers, error messages, voice setup state
 
 ### Navigation
 
@@ -135,6 +133,7 @@ All AI calls go through `RunAnywhere`:
 - **RAG**: `DocumentService` uses `syncfusion_flutter_pdf` for PDF text extraction. The RAG model selection flow does NOT pre-load models into memory — it only passes paths to `RAGConfiguration`.
 - **Tools**: three demo tools registered (`get_weather`, `calculate`, `get_current_time`). Weather tool uses Open-Meteo free API via `package:http`.
 - **Structured Output**: uses `LLMGenerationOptions(jsonSchema:)` with predefined schemas.
+- **Remaining SDK-owned cleanup**: the large model catalog still lives in app startup because moving it safely requires shared SDK/package ownership beyond this example-app lane.
 
 ## Build Configuration Gotchas
 

docs/CPP_PROTO_OWNERSHIP.md

@@ -118,7 +118,7 @@ This sample app demonstrates the full power of the RunAnywhere Flutter SDK:
 | **Speech-to-Text** | Voice transcription with batch & live modes | `RunAnywhere.stt.transcribe()` |
 | **Text-to-Speech** | Neural voice synthesis with Piper TTS | `RunAnywhere.tts.synthesize()` |
 | **Voice Assistant** | Full STT to LLM to TTS pipeline with auto-detection | `RunAnywhere.voice` |
-| **Model Management** | Download, load, and manage multiple AI models | `ModelManager` |
+| **Model Management** | Download, load, and manage multiple AI models | `RunAnywhere.models` / `RunAnywhere.downloads` |
 | **Storage Management** | View storage usage and delete models | `RunAnywhere.downloads.getStorageInfo()` |
 | **Offline Support** | All features work without internet | On-device inference |
 
@@ -138,8 +138,8 @@ The app follows Flutter best practices with a clean architecture pattern:
 ├───────┼────────────┼────────────┼────────────┼─────────────┼────────┤
 │       ▼            ▼            ▼            ▼             ▼        │
 │  ┌──────────────────────────────────────────────────────────────┐   │
-│  │                   Provider State Management                   │   │
-│  │                   (ModelManager, Services)                    │   │
+│  │                 Feature ViewModels + UI State                 │   │
+│  │           (SDK facades, Services, ListenableBuilder)          │   │
 │  └──────────────────────────────────────────────────────────────┘   │
 ├─────────────────────────────────────────────────────────────────────┤
 │                                                                     │
@@ -161,9 +161,9 @@ The app follows Flutter best practices with a clean architecture pattern:
 
 ### Key Architecture Decisions
 
-- **Provider Pattern** — `ChangeNotifier` + `Provider` for state management
+- **Feature-Local State** — Screens keep ephemeral UI state local and call SDK facades directly
 - **Feature-First Structure** — Each feature is self-contained with its own views and logic
-- **Shared Core Services** — `ModelManager`, `AudioRecordingService`, `AudioPlayerService`
+- **Shared Core Services** — `AudioRecordingService`, `AudioPlayerService`, persistence and device helpers
 - **Design System** — Consistent `AppColors`, `AppTypography`, `AppSpacing` tokens
 - **SDK Integration** — Direct SDK calls with async/await and Stream support
 
@@ -190,7 +190,6 @@ RunAnywhereAI/
 │   │   │   └── app_types.dart         # Shared type definitions
 │   │   │
 │   │   ├── services/
-│   │   │   ├── model_manager.dart     # SDK model management wrapper
 │   │   │   ├── audio_recording_service.dart  # Microphone capture
 │   │   │   ├── audio_player_service.dart     # TTS playback
 │   │   │   ├── permission_service.dart       # Permission handling

examples/flutter/RunAnywhereAI/CLAUDE.md

@@ -2,18 +2,16 @@ import 'dart:async';
 import 'dart:io' show Platform;
 
 import 'package:flutter/material.dart';
-import 'package:provider/provider.dart';
 import 'package:runanywhere/runanywhere.dart';
 import 'package:runanywhere_ai/app/content_view.dart';
 import 'package:runanywhere_ai/core/design_system/app_colors.dart';
-import 'package:runanywhere_ai/core/services/model_manager.dart';
 import 'package:runanywhere_ai/core/utilities/constants.dart';
 import 'package:runanywhere_ai/core/utilities/keychain_helper.dart';
 import 'package:runanywhere_genie/runanywhere_genie.dart';
 import 'package:runanywhere_llamacpp/runanywhere_llamacpp.dart';
 import 'package:runanywhere_onnx/runanywhere_onnx.dart';
 
-/// RunAnywhereAIApp (mirroring iOS RunAnywhereAIApp.swift)
+/// RunAnywhereAIApp
 ///
 /// Main application entry point with SDK initialization.
 class RunAnywhereAIApp extends StatefulWidget {
@@ -87,8 +85,6 @@ class _RunAnywhereAIAppState extends State<RunAnywhereAIApp> {
       debugPrint(
           '🔧 Environment: ${RunAnywhere.environment?.description ?? "Unknown"}');
 
-      await ModelManager.shared.refresh();
-
       debugPrint(
           '💡 Models registered, user can now download and select models');
     } catch (e) {
@@ -123,8 +119,7 @@ class _RunAnywhereAIAppState extends State<RunAnywhereAIApp> {
       // ── Step 1: Download ─────────────────────────────────────────
       debugPrint('▶ Step 1: downloading $testModelId …');
       final sw = Stopwatch()..start();
-      final progressStream =
-          RunAnywhere.downloads.start(testModelId);
+      final progressStream = RunAnywhere.downloads.start(testModelId);
 
       double lastPct = 0;
       await for (final p in progressStream) {
@@ -144,8 +139,6 @@ class _RunAnywhereAIAppState extends State<RunAnywhereAIApp> {
 
       // Let the download adapter finalize (path update, registry sync)
       await Future<void>.delayed(const Duration(milliseconds: 500));
-      await ModelManager.shared.refresh();
-
       // Re-run discovery so the SDK sees the downloaded file
       final models = await RunAnywhere.models.available();
       final dlModel = models.where((m) => m.id == testModelId).firstOrNull;
@@ -170,8 +163,7 @@ class _RunAnywhereAIAppState extends State<RunAnywhereAIApp> {
         temperature: 0.7,
         systemPrompt: 'You are a helpful assistant.',
       );
-      final result =
-          await RunAnywhere.llm.generate('Hello!', genOpts);
+      final result = await RunAnywhere.llm.generate('Hello!', genOpts);
       sw.stop();
       debugPrint(
           '✅ generate() => "${result.text.substring(0, result.text.length.clamp(0, 120))}"');
@@ -188,8 +180,7 @@ class _RunAnywhereAIAppState extends State<RunAnywhereAIApp> {
         temperature: 0.7,
         systemPrompt: 'You are a helpful assistant.',
       );
-      final stream = RunAnywhere.llm
-          .generateStream('What is 2+2?', streamOpts);
+      final stream = RunAnywhere.llm.generateStream('What is 2+2?', streamOpts);
       final buf = StringBuffer();
       int tokenCount = 0;
       await for (final event in stream) {
@@ -222,7 +213,7 @@ class _RunAnywhereAIAppState extends State<RunAnywhereAIApp> {
     }
   }
 
-  /// True once we've registered modules + models exactly once. Without
+  /// True once we've registered modules + models once. Without
   /// this guard, hot-reload (or any second call) re-runs the entire
   /// LLM catalog registration block, which is wasteful (B-FL-3-002).
   static bool _modulesRegistered = false;
@@ -248,7 +239,6 @@ class _RunAnywhereAIAppState extends State<RunAnywhereAIApp> {
 
   /// Register modules with their associated models
   /// Each module explicitly owns its models - the framework is determined by the module
-  /// Matches iOS registerModulesAndModels pattern exactly
   Future<void> _registerModulesAndModels() async {
     if (_modulesRegistered) {
       debugPrint('📦 Modules already registered — skipping (B-FL-3-002)');
@@ -505,7 +495,8 @@ class _RunAnywhereAIAppState extends State<RunAnywhereAIApp> {
       modality: ModelCategory.MODEL_CATEGORY_VOICE_ACTIVITY_DETECTION,
       memoryRequirement: 5000000,
     );
-    debugPrint('✅ STT/TTS/VAD models registered via Core SDK (incl. system-tts)');
+    debugPrint(
+        '✅ STT/TTS/VAD models registered via Core SDK (incl. system-tts)');
     await Future<void>.delayed(Duration.zero);
 
     // --- RAG EMBEDDINGS ---
@@ -555,42 +546,37 @@ class _RunAnywhereAIAppState extends State<RunAnywhereAIApp> {
 
   @override
   Widget build(BuildContext context) {
-    return MultiProvider(
-      providers: [
-        ChangeNotifierProvider.value(value: ModelManager.shared),
-      ],
-      child: MaterialApp(
-        scaffoldMessengerKey: _messengerKey,
-        title: 'RunAnywhere AI',
-        debugShowCheckedModeBanner: false,
-        theme: ThemeData(
-          colorScheme: ColorScheme.fromSeed(
-            seedColor: AppColors.primaryBlue,
-            brightness: Brightness.light,
-          ),
-          useMaterial3: true,
-          appBarTheme: const AppBarTheme(
-            centerTitle: true,
-            elevation: 0,
-          ),
-          navigationBarTheme: NavigationBarThemeData(
-            indicatorColor: AppColors.primaryBlue.withValues(alpha: 0.2),
-          ),
+    return MaterialApp(
+      scaffoldMessengerKey: _messengerKey,
+      title: 'RunAnywhere AI',
+      debugShowCheckedModeBanner: false,
+      theme: ThemeData(
+        colorScheme: ColorScheme.fromSeed(
+          seedColor: AppColors.primaryBlue,
+          brightness: Brightness.light,
+        ),
+        useMaterial3: true,
+        appBarTheme: const AppBarTheme(
+          centerTitle: true,
+          elevation: 0,
+        ),
+        navigationBarTheme: NavigationBarThemeData(
+          indicatorColor: AppColors.primaryBlue.withValues(alpha: 0.2),
+        ),
+      ),
+      darkTheme: ThemeData(
+        colorScheme: ColorScheme.fromSeed(
+          seedColor: AppColors.primaryBlue,
+          brightness: Brightness.dark,
         ),
-        darkTheme: ThemeData(
-          colorScheme: ColorScheme.fromSeed(
-            seedColor: AppColors.primaryBlue,
-            brightness: Brightness.dark,
-          ),
-          useMaterial3: true,
-          appBarTheme: const AppBarTheme(
-            centerTitle: true,
-            elevation: 0,
-          ),
+        useMaterial3: true,
+        appBarTheme: const AppBarTheme(
+          centerTitle: true,
+          elevation: 0,
         ),
-        themeMode: ThemeMode.system,
-        home: _buildHome(),
       ),
+      themeMode: ThemeMode.system,
+      home: _buildHome(),
     );
   }