Complete RAG Flutter implementation (full state) (#419)

RAG Flutter SDK.

->there are a bunch of UI/UX issues with this like button not loading, the round spinny download thing not being proper etc, but the rag pipeline works.

-> also onnx and rag when built together were giving duplicate symbol errors as rag requires onnx, so a future task that should be done soon is to include onnx in core as well, and perhaps add a conditional?

Author: VyasGuru

examples/flutter/RunAnywhereAI/lib/app/runanywhere_ai_app.dart

@@ -10,8 +10,8 @@ import 'package:runanywhere_ai/core/design_system/app_spacing.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/public/extensions/rag_module.dart';
 import 'package:runanywhere_llamacpp/runanywhere_llamacpp.dart';
-import 'package:runanywhere_onnx/runanywhere_onnx.dart';
 
 /// RunAnywhereAIApp (mirroring iOS RunAnywhereAIApp.swift)
 ///
@@ -140,11 +140,8 @@ class _RunAnywhereAIAppState extends State<RunAnywhereAIApp> {
   Future<void> _registerModulesAndModels() async {
     debugPrint('📦 Registering modules with their models...');
 
-    // LlamaCPP module with LLM models
-    // Using explicit IDs ensures models are recognized after download across app restarts
+    // --- LLAMACPP MODULE ---
     await LlamaCpp.register();
-
-    // Yield after heavy backend registration
     await Future<void>.delayed(Duration.zero);
 
     LlamaCpp.addModel(
@@ -190,8 +187,6 @@ class _RunAnywhereAIAppState extends State<RunAnywhereAIApp> {
       memoryRequirement: 400000000,
     );
 
-    // Tool Calling Optimized Models
-    // LFM2-1.2B-Tool - Designed for concise and precise tool calling (Liquid AI)
     LlamaCpp.addModel(
       id: 'lfm2-1.2b-tool-q4_k_m',
       name: 'LiquidAI LFM2 1.2B Tool Q4_K_M',
@@ -206,16 +201,10 @@ class _RunAnywhereAIAppState extends State<RunAnywhereAIApp> {
           'https://huggingface.co/LiquidAI/LFM2-1.2B-Tool-GGUF/resolve/main/LFM2-1.2B-Tool-Q8_0.gguf',
       memoryRequirement: 1400000000,
     );
-    debugPrint('✅ LlamaCPP module registered with LLM models (including tool-calling optimized models)');
-
-    // Yield between module registrations
+    debugPrint('✅ LlamaCPP module registered');
     await Future<void>.delayed(Duration.zero);
 
-    // Register VLM (Vision Language) models
-    // VLM models require 2 files: main model + mmproj (vision projector)
-    // Bundled as tar.gz archives for easy download/extraction
-
-    // SmolVLM 500M - Ultra-lightweight VLM for mobile (~500MB total)
+    // --- VLM MODULE ---
     RunAnywhere.registerModel(
       id: 'smolvlm-500m-instruct-q8_0',
       name: 'SmolVLM 500M Instruct',
@@ -229,57 +218,81 @@ class _RunAnywhereAIAppState extends State<RunAnywhereAIApp> {
       memoryRequirement: 600000000,
     );
     debugPrint('✅ VLM models registered');
-
-    // Yield between module registrations
-    await Future<void>.delayed(Duration.zero);
-
-    // Diffusion (image generation) is not registered here. CoreML diffusion is supported
-    // only in the Swift SDK and Swift example app; Flutter/RN do not register diffusion.
-
-    // ONNX module with STT and TTS models
-    // Using tar.gz format hosted on RunanywhereAI/sherpa-onnx for fast native extraction
-    // Using explicit IDs ensures models are recognized after download across app restarts
-    await Onnx.register();
-
-    // Yield after heavy backend registration
     await Future<void>.delayed(Duration.zero);
 
+    // --- ONNX MODULE (STT/TTS via Core SDK) ---
     // STT Models (Sherpa-ONNX Whisper)
-    Onnx.addModel(
+    RunAnywhere.registerModel(
       id: 'sherpa-onnx-whisper-tiny.en',
       name: 'Sherpa Whisper Tiny (ONNX)',
-      url:
-          'https://github.com/RunanywhereAI/sherpa-onnx/releases/download/runanywhere-models-v1/sherpa-onnx-whisper-tiny.en.tar.gz',
+      url: Uri.parse('https://github.com/RunanywhereAI/sherpa-onnx/releases/download/runanywhere-models-v1/sherpa-onnx-whisper-tiny.en.tar.gz'),
+      framework: InferenceFramework.onnx,
       modality: ModelCategory.speechRecognition,
       memoryRequirement: 75000000,
     );
-    Onnx.addModel(
+
+    RunAnywhere.registerModel(
       id: 'sherpa-onnx-whisper-small.en',
       name: 'Sherpa Whisper Small (ONNX)',
-      url:
-          'https://github.com/RunanywhereAI/sherpa-onnx/releases/download/runanywhere-models-v1/sherpa-onnx-whisper-small.en.tar.gz',
+      url: Uri.parse('https://github.com/RunanywhereAI/sherpa-onnx/releases/download/runanywhere-models-v1/sherpa-onnx-whisper-small.en.tar.gz'),
+      framework: InferenceFramework.onnx,
       modality: ModelCategory.speechRecognition,
       memoryRequirement: 250000000,
     );
 
     // TTS Models (Piper VITS)
-    Onnx.addModel(
+    RunAnywhere.registerModel(
       id: 'vits-piper-en_US-lessac-medium',
       name: 'Piper TTS (US English - Medium)',
-      url:
-          'https://github.com/RunanywhereAI/sherpa-onnx/releases/download/runanywhere-models-v1/vits-piper-en_US-lessac-medium.tar.gz',
+      url: Uri.parse('https://github.com/RunanywhereAI/sherpa-onnx/releases/download/runanywhere-models-v1/vits-piper-en_US-lessac-medium.tar.gz'),
+      framework: InferenceFramework.onnx,
       modality: ModelCategory.speechSynthesis,
       memoryRequirement: 65000000,
     );
-    Onnx.addModel(
+
+    RunAnywhere.registerModel(
       id: 'vits-piper-en_GB-alba-medium',
       name: 'Piper TTS (British English)',
-      url:
-          'https://github.com/RunanywhereAI/sherpa-onnx/releases/download/runanywhere-models-v1/vits-piper-en_GB-alba-medium.tar.gz',
+      url: Uri.parse('https://github.com/RunanywhereAI/sherpa-onnx/releases/download/runanywhere-models-v1/vits-piper-en_GB-alba-medium.tar.gz'),
+      framework: InferenceFramework.onnx,
       modality: ModelCategory.speechSynthesis,
       memoryRequirement: 65000000,
     );
-    debugPrint('✅ ONNX module registered with STT/TTS models');
+    debugPrint('✅ STT/TTS models registered via Core SDK');
+    await Future<void>.delayed(Duration.zero);
+
+    // --- RAG EMBEDDINGS ---
+    RunAnywhere.registerMultiFileModel(
+      id: 'all-minilm-l6-v2',
+      name: 'All MiniLM L6 v2 (Embedding)',
+      files: [
+        ModelFileDescriptor(
+          relativePath: 'model.onnx',
+          destinationPath: 'model.onnx',
+          url: Uri.parse(
+              'https://huggingface.co/Xenova/all-MiniLM-L6-v2/resolve/main/onnx/model.onnx'),
+        ),
+        ModelFileDescriptor(
+          relativePath: 'vocab.txt',
+          destinationPath: 'vocab.txt',
+          url: Uri.parse(
+              'https://huggingface.co/Xenova/all-MiniLM-L6-v2/resolve/main/vocab.txt'),
+        ),
+      ],
+      framework: InferenceFramework.onnx,
+      modality: ModelCategory.embedding,
+      memoryRequirement: 25500000,
+    );
+    debugPrint('✅ ONNX Embedding models registered');
+    await Future<void>.delayed(Duration.zero);
+
+    // --- RAG BACKEND ---
+    try {
+      await RAGModule.register();
+      debugPrint('✅ RAG backend registered');
+    } catch (e) {
+      debugPrint('⚠️ RAG backend not available (RAG features disabled): $e');
+    }
 
     debugPrint('🎉 All modules and models registered');
   }

examples/flutter/RunAnywhereAI/lib/features/chat/chat_interface_view.dart

@@ -16,6 +16,7 @@ import 'package:runanywhere_ai/features/models/model_selection_sheet.dart';
 import 'package:runanywhere_ai/features/models/model_status_components.dart';
 import 'package:runanywhere_ai/features/models/model_types.dart';
 import 'package:runanywhere_ai/features/settings/tool_settings_view_model.dart';
+import 'package:runanywhere_ai/features/rag/rag_demo_view.dart';
 import 'package:runanywhere_ai/features/structured_output/structured_output_view.dart';
 import 'package:shared_preferences/shared_preferences.dart';
 
@@ -454,6 +455,17 @@ class _ChatInterfaceViewState extends State<ChatInterfaceView> {
       appBar: AppBar(
         title: const Text('Chat'),
       actions: [
+          IconButton(
+            icon: const Icon(Icons.article_outlined),
+            onPressed: () {
+              Navigator.of(context).push<void>(
+                MaterialPageRoute<void>(
+                  builder: (context) => const RagDemoView(),
+                ),
+              );
+            },
+            tooltip: 'Document Q&A',
+          ),
           IconButton(
             icon: const Icon(Icons.data_object),
             onPressed: () {
@@ -947,4 +959,4 @@ class _MessageBubbleState extends State<_MessageBubble> {
       ),
     );
   }
-}
+}

examples/flutter/RunAnywhereAI/lib/features/models/model_list_view_model.dart

@@ -106,6 +106,8 @@ class ModelListViewModel extends ChangeNotifier {
         return ModelCategory.imageGeneration;
       case sdk.ModelCategory.audio:
         return ModelCategory.audio;
+      case sdk.ModelCategory.embedding:
+        return ModelCategory.embedding;
     }
   }
 

examples/flutter/RunAnywhereAI/lib/features/models/model_selection_sheet.dart

@@ -507,8 +507,15 @@ class _ModelSelectionSheetState extends State<ModelSelectionSheet> {
     });
 
     try {
-      // Update view model selection state
-      await _viewModel.selectModel(model);
+      // RAG contexts record the selection only — do NOT pre-load into memory.
+      // The RAG pipeline loads models on demand when the document is ingested.
+      final isRagContext = widget.context == ModelSelectionContext.ragEmbedding ||
+          widget.context == ModelSelectionContext.ragLLM;
+
+      if (!isRagContext) {
+        // Update view model selection state (loads the model into memory)
+        await _viewModel.selectModel(model);
+      }
 
       // Call the callback - this is where the actual model loading happens
       // The callback knows the correct context and how to load the model

examples/flutter/RunAnywhereAI/lib/features/models/model_status_components.dart

@@ -251,6 +251,10 @@ class ModelRequiredOverlay extends StatelessWidget {
         return Icons.mic;
       case ModelSelectionContext.vlm:
         return Icons.center_focus_strong;
+      case ModelSelectionContext.ragEmbedding:
+        return Icons.data_object;
+      case ModelSelectionContext.ragLLM:
+        return Icons.question_answer_outlined;
     }
   }
 
@@ -266,6 +270,10 @@ class ModelRequiredOverlay extends StatelessWidget {
         return 'Voice Assistant';
       case ModelSelectionContext.vlm:
         return 'Vision Language Model';
+      case ModelSelectionContext.ragEmbedding:
+        return 'Document RAG';
+      case ModelSelectionContext.ragLLM:
+        return 'Document RAG';
     }
   }
 
@@ -281,6 +289,10 @@ class ModelRequiredOverlay extends StatelessWidget {
         return 'Voice assistant requires multiple models. Let\'s set them up together.';
       case ModelSelectionContext.vlm:
         return 'Select a vision-language model to analyze images. Point your camera or pick a photo to get AI descriptions.';
+      case ModelSelectionContext.ragEmbedding:
+        return 'Select an embedding model to encode document chunks for retrieval.';
+      case ModelSelectionContext.ragLLM:
+        return 'Select a language model to generate answers from retrieved document context.';
     }
   }
 }

examples/flutter/RunAnywhereAI/lib/features/models/model_types.dart

@@ -124,7 +124,9 @@ enum ModelSelectionContext {
   stt,
   tts,
   voice,
-  vlm;
+  vlm,
+  ragEmbedding,
+  ragLLM;
 
   String get title {
     switch (this) {
@@ -138,6 +140,10 @@ enum ModelSelectionContext {
         return 'Select Model';
       case ModelSelectionContext.vlm:
         return 'Select VLM Model';
+      case ModelSelectionContext.ragEmbedding:
+        return 'Select Embedding Model';
+      case ModelSelectionContext.ragLLM:
+        return 'Select LLM Model';
     }
   }
 
@@ -158,6 +164,10 @@ enum ModelSelectionContext {
         };
       case ModelSelectionContext.vlm:
         return {ModelCategory.vision, ModelCategory.multimodal};
+      case ModelSelectionContext.ragEmbedding:
+        return {ModelCategory.embedding};
+      case ModelSelectionContext.ragLLM:
+        return {ModelCategory.language};
     }
   }
 }

examples/flutter/RunAnywhereAI/lib/features/rag/document_service.dart

@@ -0,0 +1,79 @@
+// Document Service
+//
+// Utility for extracting plain text from PDF and JSON files.
+// Used to prepare document content for RAG ingestion.
+
+import 'dart:convert';
+import 'dart:io';
+
+import 'package:syncfusion_flutter_pdf/pdf.dart';
+
+// MARK: - DocumentService
+
+/// Extracts plain text from PDF and JSON files.
+///
+/// Supports:
+/// - PDF — All page text concatenated with newlines (via syncfusion_flutter_pdf).
+/// - JSON — Raw UTF-8 file content returned as-is.
+class DocumentService {
+  // Private constructor — static-only class.
+  DocumentService._();
+
+  /// Extract plain text from a file at [filePath].
+  ///
+  /// Determines the document type from the file extension.
+  ///
+  /// Throws [UnsupportedError] for unsupported file extensions.
+  /// Throws [Exception] if the file cannot be read or parsed.
+  static Future<String> extractText(String filePath) async {
+    final extension = filePath.split('.').last.toLowerCase();
+
+    switch (extension) {
+      case 'pdf':
+        return _extractPDFText(filePath);
+      case 'json':
+        return _extractJSONText(filePath);
+      default:
+        throw UnsupportedError(
+          'Unsupported document format: .$extension. Only PDF and JSON files are supported.',
+        );
+    }
+  }
+
+  // MARK: - Private Helpers
+
+  static Future<String> _extractPDFText(String filePath) async {
+    final file = File(filePath);
+    final bytes = await file.readAsBytes();
+
+    final document = PdfDocument(inputBytes: bytes);
+    final extractor = PdfTextExtractor(document);
+
+    final buffer = StringBuffer();
+    final pageCount = document.pages.count;
+
+    for (var i = 0; i < pageCount; i++) {
+      final pageText = extractor.extractText(startPageIndex: i, endPageIndex: i);
+      if (pageText.trim().isNotEmpty) {
+        if (buffer.isNotEmpty) buffer.write('\n');
+        buffer.write(pageText);
+      }
+    }
+
+    document.dispose();
+
+    final result = buffer.toString();
+    if (result.isEmpty) {
+      throw Exception(
+        'Failed to extract text from PDF. The file may be corrupted or image-only.',
+      );
+    }
+
+    return result;
+  }
+
+  static Future<String> _extractJSONText(String filePath) async {
+    final file = File(filePath);
+    return file.readAsString(encoding: utf8);
+  }
+}