Note: Click on the logo to download the latest version of the app apk file.

AI Voice Genie Logo

Enterprise-Grade, Unified Multi-Model AI Assistant Engine for Mobile

Flutter 3.38.2 Dart 3.10.0 Android & iOS Feature-First MVVM Screen Protected Open Source Firebase Integration Provider Hive Storage AI Models

AI Voice Genie is a high-performance, production-grade Flutter mobile application engineered as a unified hub for leading artificial intelligence providers. Rather than forcing users to manage fragmented subscriptions across separate apps, AI Voice Genie consolidates OpenAI, Google Gemini, and Anthropic Claude into a single, cohesive, local-first mobile interface.

Designed around a Bring Your Own Key (BYOK) paradigm, the application grants users complete ownership over their AI provider accounts and model preferences. Users can seamlessly switch between OpenAI GPT-4o & GPT-Image-1, Google Gemini-2.5-Flash && Gemini-2.5-Flash-Image, and Anthropic Claude 3.5 Sonnet for text generation, AI image generation, multimodal vision analysis, and PDF document parsing.

Built upon a Local-First Architecture powered by zero-latency Hive local databases, chat history is read and rendered instantaneously without network delays. A durable background outbox (ChatOutboxTask) automatically queues mutations offline and reconciles them with Cloud Firestore when connectivity returns. The application also integrates real-time continuous Speech-to-Text (STT) dictation with custom grammar formatting, offline Text-to-Speech (TTS) response playback, detailed model token cost tracking, custom monthly budget limits, anti-screenshot API key screen protection, and native scoped media downloading (Pictures/AI Voice Genie on Android, Photos on iOS).

Experience a meticulously crafted UI supporting fully responsive Light and Dark themes.

| Voice Assistant Functionality |

AI Voice Genie normalizes communications with multiple AI vendor APIs into a single request/response engine (AiOrchestrator). Below is the exact breakdown of supported models and their capabilities:

AI Voice Genie implements a smart, adaptive preferences architecture managed by AiPreferencesProvider and backed by ProviderRegistry:

• Capability Matrix Filtering: When a user selects a model, the UI dynamically adapts. For instance, selecting Claude 3.5 Sonnet automatically hides image generation controls (preventing invalid API calls), while selecting Gemini exposes aspect ratio selectors (1:1, 16:9, 4:3, 3:4) and resolution quality tiers.
• Custom System Instructions: Users can configure global system persona prompts, max response tokens (up to 4,096 tokens), and vision detail quality (Low, High, Auto).
• Sanitized Request Building: Before any API call is initiated, ProviderRegistry.sanitizePreferences() strips unsupported parameters from the request payload, ensuring zero HTTP 400 bad request errors due to model capability mismatch.

For complete transparency and financial control, AI Voice Genie features a dedicated Usage Tracking System (UsageProvider & UsageRepositoryImpl):

• Token Metric Breakdown: Real-time tracking of Prompt Tokens, Completion Tokens, and Total Tokens consumed across every AI model call.
• Estimated Cost Calculation: Automatically computes exact USD spending ($) per interaction based on UsagePricingTable rates.
• Monthly Budget Limits: Users can define a custom monthly budget (e.g. 10ドル.00/month). The UI displays a visual progress bar and triggers warning banners when consumption approaches 80% or 100% of the allocated budget.
• Historical Telemetry Sync: Usage data is stored locally in Hive for instant rendering and synced to AIVoiceGenie/UsersUsage/{uid} in Cloud Firestore.

Enterprise-grade security controls protect user data and sensitive credentials at every layer:

• 🛡️ Anti-Screenshot & Anti-Screen Recording Protection (no_screenshot): The API Key Setup and Management screens (KeySetupScreen) invoke native OS security policies (FLAG_SECURE on Android and secure window buffering on iOS) in initState(). This completely blocks screenshots, screen recordings, and background task switcher previews from capturing sensitive API keys.
• 🔑 Encrypted On-Device Storage (flutter_secure_storage & encrypt): API keys are encrypted at rest using AES-256 (encrypt: ^5.0.3) and stored inside OS-level secure storage (Android Keystore / iOS Keychain).
• 🎭 Masked Key Display: Keys are truncated and masked on screen (sk-a...789) via StringExtension.maskedApiKey to prevent visual shoulder surfing.
• ☁️ Restricted Firestore Rules: Remote backup of API keys is stored in AIVoiceGenie/UsersAPIKeys/{uid} protected by granular Firestore Security Rules that enforce strict user-only read/write access.
• 🔐 OAuth2 Security & Nonce Verification: Google and Apple OAuth sign-in flows generate cryptographic SHA-256 nonces (crypto: ^3.0.7) to protect against replay attacks.
• 🗑️ Hard Account Destruction: Account deletion executes a total wipe: hard-deletes remote Firestore profile documents and API key records, wipes all local Hive database boxes, resets OAuth tokens, and deletes the Firebase Auth user account.
• 🙈 Zero-Trust Git Secrets Policy: Environment files (.env), Firebase config files (google-services.json, GoogleService-Info.plist), key properties, and keystores are strictly excluded from version control via .gitignore.

• 🤖 Multi-Model BYOK Engine: Native integrations with OpenAI (gpt-4o, gpt-image-1), Gemini (gemini-2.5-flash), and Claude (claude-3-5-sonnet).
• ⚡ Local-First Instant UI: Hive key-value storage engine ensures instantaneous screen rendering with zero network delay.
• 🔄 Durable Offline Outbox Sync: Actions performed offline are stored as ChatOutboxTask items and processed asynchronously by ChatSyncService when online.
• 📜 Cursor-Based Pagination: Memory-efficient infinite scrolling loads 15 conversations per page in history and 30 initial / 20 scroll-up messages in chat detail.
• 🎙️ Speech-to-Text (STT) & Auto-TTS: Continuous microphone dictation with automatic grammar formatting, plus intelligent Text-to-Speech (TTS) auto-playback for AI responses based on user preferences.
• 🧹 Prompt Sanitization Engine: Automatically intercepts and strips conversational filler and greetings from prompts to maximize model attention and generate highly relevant conversation titles.
• 🛡️ Safety & Moderation Layer: Deep integration with Gemini and OpenAI moderation endpoints to gracefully catch and localize safety violation blocks.
• 📄 Multimodal Vision & PDF Analysis: Client-side PDF text extraction (syncfusion_flutter_pdf) and vision analysis for multi-image prompts.
• 💾 Native Scoped Gallery Saving: Scoped MethodChannel (ImageSavePlugin.kt for Android MediaStore API 29+ & ImageSavePlugin.swift for iOS PhotoKit) saves AI generated images directly to Pictures/AI Voice Genie or Photos without intrusive permissions.
• 📊 Usage Telemetry & Spending Limits: Real-time token consumption metrics, estimated USD cost calculations, and monthly budget alert limits.
• 🌐 Multilingual & Adaptive Themes: Light/Dark/System theme toggle and real-time English/Hindi localization.

AI Voice Genie is built using Feature-First MVVM (Model-View-ViewModel) with Provider for reactive state management.

graph TD
 subgraph Presentation_Layer ["Presentation Layer (UI & Views)"]
 UI_Screens["Screens & Views<br/>(LoginScreen, ChatScreen, ChatDetailScreen,<br/>ProfileView, ConversationHistoryScreen, KeySetupScreen, UsageScreen, LegalScreen)"]
 UI_Widgets["Widgets & Components<br/>(MessageBubble, CustomConversationCard, EmptyHistoryView,<br/>ChatInputBar, ProfileAvatar, AiPreferencesPanel)"]
 end
 subgraph ViewModel_Layer ["ViewModel / Provider Layer (State Management)"]
 AuthProvider["AuthProvider (Session & Auth State)"]
 ChatProvider["ChatProvider (Messages, History, Pagination)"]
 ProfileViewModel["ProfileViewModel (Profile & Account Actions)"]
 AiPreferencesProvider["AiPreferencesProvider (Capability Matrix & Prefs)"]
 VoiceSpeechProvider["VoiceSpeechProvider (STT Dictation & TTS Playback)"]
 UsageProvider["UsageProvider (Token Metrics & Budget Limits)"]
 end
 subgraph Domain_Layer ["Domain Layer (Business Entities & Contracts)"]
 Entities["UserModel, ConversationModel, MessageModel, ChatAttachment, ChatOutboxTask"]
 Contracts["AuthRepository, ChatRepository, ApiKeyRepository, UsageRepository"]
 end
 subgraph Data_Layer ["Data Layer (Repositories & Stores)"]
 AuthRepo["AuthRepositoryImpl"]
 ChatRepo["ChatRepositoryImpl"]
 LocalStore["LocalChatStore (Hive)"]
 OutboxStore["ChatOutboxStore & ChatSyncService"]
 UsageRepo["UsageRepositoryImpl"]
 end
 subgraph Infrastructure_Layer ["Infrastructure Layer"]
 Firebase["Firebase Auth & Cloud Firestore"]
 AI_Engine["AiOrchestrator (OpenAI, Gemini, Claude Adapters)"]
 NativeChannels["MethodChannel ImageSavePlugin (Android / iOS)"]
 ScreenSecurity["NoScreenshot Plugin (FLAG_SECURE)"]
 end
 UI_Screens -->|Binds to| ViewModel_Layer
 UI_Widgets -->|Binds to| ViewModel_Layer
 ViewModel_Layer -->|Uses Entities| Domain_Layer
 ViewModel_Layer -->|Invokes Repositories| Data_Layer
 Data_Layer -->|Implements Contracts| Contracts
 Data_Layer -->|Local Storage| LocalStore
 Data_Layer -->|Remote Sync| Firebase
 Data_Layer -->|AI Execution| AI_Engine
 Data_Layer -->|Gallery Storage| NativeChannels
 UI_Screens -->|Enforces Security| ScreenSecurity

For technical leads and engineering reviewers, AI Voice Genie implements several advanced patterns to guarantee high performance, resilience, and security on mobile devices:

Traditional local apps often load entire database tables into memory, causing UI jank as data grows. AI Voice Genie utilizes a cursor-based pagination algorithm against the local Hive database. By sorting keys lexicographically (e.g., timestamp_conversationId), the app skips directly to the last rendered node and fetches only a discrete chunk (15 conversations or 20 messages). This ensures the app's RAM footprint remains strictly $O(1)$ and the UI maintains a perfect 60-120fps, even if the user has 10,000+ saved messages.

To handle the chaotic nature of mobile networks, the app relies on the Outbox Pattern. When a user sends a message offline:

1. The UI optimistically updates instantly from the local Hive store.
2. A ChatOutboxTask (containing the mutation payload) is durably serialized to a dedicated Hive outbox.
3. A background ChatSyncService, listening to connectivity_plus, observes network restoration.
4. The service drains the outbox queue, executing Firestore writes with Exponential Backoff to handle transient cloud rate limits, ensuring zero data loss during tunnel/subway network drops.

Raw Speech-to-Text (STT) streams are often unformatted and difficult to read. As the native speech_to_text engine streams raw words, AI Voice Genie applies a real-time regex-based Grammar Enhancement Pipeline. It automatically capitalizes proper nouns (e.g., "openai" → "OpenAI"), detects interrogative clauses to append question marks, and filters out introductory fluff (e.g., "Hey Genie") to ensure AI models receive highly structured, token-efficient prompt strings.

Mobile environments are inherently hostile. AI Voice Genie adopts a zero-trust posture for user API keys:

• Hardware-Backed Encryption: Keys are encrypted via AES-GCM and stored in the Android Keystore / iOS Secure Enclave (flutter_secure_storage).
• Memory Obfuscation: Keys are masked (sk-a...789) when rendered in memory for the UI.
• OS-Level Screen Protection: The no_screenshot plugin dynamically applies Android's FLAG_SECURE window layout parameter and iOS secure text buffering upon entering the KeySetupScreen. This legally prevents the OS, screen recorders, and malicious background apps from scraping the screen buffer.
• Sanitization Pipeline: Before any HTTP request fires, the ProviderRegistry executes a strict capability check, stripping unsupported fields (like image aspect ratios for text-only models) to guarantee zero 400 Bad Request exceptions.

To ensure zero impact on UI frame rates, the app implements a "fire-and-forget" telemetry pattern using Firebase Analytics and Firebase Crashlytics.

• Crashlytics: Automatically intercepts unhandled Dart exceptions, native crashes, and out-of-memory (OOM) errors, pushing stack traces to the cloud without locking the main thread.
• Analytics: Tracks critical user journeys (e.g., ai_request_success, image_downloaded, subscription_limit_reached) asynchronously, allowing engineering teams to monitor model usage and feature adoption in real-time.

Here is the exact list of packages declared in pubspec.yaml with their defined versions and operational roles:

AI Voice Genie provides complete transparency and user support through dedicated legal and assistance interfaces:

• Legal Section (LegalScreen): Features a segmented tab bar toggling between the Privacy Policy (assets/legal/privacy_policy.md) and Terms of Service (assets/legal/terms_of_service.md). It details data handling practices, zero data sale commitments, and AI model usage policies.
• About App (AboutScreen): Displays app metadata, current version (package_info_plus), copyright information, and developer credits.
• Support Panel (SupportPanel): Embedded in the Profile screen, providing direct access to a mailto: customer support email link via url_launcher allowing users to contact the app developer directly.

• Flutter SDK: 3.38.2
• Dart SDK: 3.10.0
• Xcode 15+ (for iOS builds) or Android Studio Jellyfish+ (for Android builds)
• A Firebase project with Auth (Google & Apple) and Cloud Firestore enabled.

1. Clone the repository:

   git clone https://github.com/YourUsername/voice_genie.git
   cd voice_genie

2. Install exact package dependencies:

   flutter pub get

3. Firebase Setup:

   • Add google-services.json to android/app/.
   • Add GoogleService-Info.plist to ios/Runner/.

4. Launch the App:

   flutter run

We welcome bug reports and feature requests from both end-users and internal team members! Please use our GitHub repository's Issue Tracker to submit your feedback.

• GitHub Issues: Please raise all issues directly in the repository's issue tracker.
• Bug Reports: Include your device model, OS version, steps to reproduce, and attach any relevant screenshots or screen recordings. Internal team members should also attach Firebase Crashlytics log IDs if applicable.
• Feature Requests: Outline the proposed feature, the target AI model (OpenAI, Gemini, Claude), and your use case.

Always branch off the development branch using descriptive naming conventions:

• feature/your-feature-name (e.g., feature/claude-vision-support)
• fix/issue-description (e.g., fix/hive-pagination-crash)
• hotfix/critical-bug (for production emergencies)

1. Ensure your local branch is up to date with development.
2. Run code formatting: dart format lib/
3. Run static analysis: flutter analyze (Must pass with 0 issues).
4. Submit the PR and request review from at least one senior engineer or technical lead.

Copyright © 2026 Arpit Aswal.

This repository is provided for portfolio and evaluation purposes only.

Commercial use, redistribution, modification, or reproduction without written permission is prohibited.