ToolNeuron is the first Android application to combine Chat AI (LLMs), Image Generation (Stable Diffusion), Voice AI (TTS/STT), and Custom Knowledge Injection (RAG) in a single, privacy-first package. Everything runs entirely on-device with zero internet dependency, or connect to cloud models for maximum flexibility.

Stop choosing between privacy and power. ToolNeuron gives you both.

Three Operating Modes:

• 🔒 Privacy Mode — Execute GGUF models (Llama 3, Mistral, Gemma), generate images with Stable Diffusion 1.5, use voice AI, and inject custom knowledge—all completely offline. Your data never leaves your phone.
• ⚡ Power Mode — Access 100+ premium cloud models (GPT-4, Claude 3.5, Gemini, DALL-E) via OpenRouter for complex tasks requiring maximum capability.
• 🔄 Hybrid Intelligence — Seamlessly switch between offline and cloud modes mid-conversation while preserving full context and conversation history.

🤖 Complete AI Suite Offline
The only mobile app that combines chat, image generation, voice, and knowledge injection—all running on-device without internet.

🎨 On-Device Image Generation
Run Stable Diffusion 1.5 (censored & uncensored) completely offline. Generate images on flights, in remote areas, anywhere.

🧠 RAG Data-Packs
Inject Wikipedia dumps, coding documentation, personal notes, or any custom knowledge directly into AI context—no model retraining required.

🔌 Extensible Plugin System
Add web search, content scraping, document analysis, and more. Build your own plugins for unlimited extensibility.

🎙️ Premium Offline Voice
11 professional TTS voices + Whisper STT—all running on-device with zero cloud dependencies and near-instant processing.

🌐 100+ Cloud Models
When you need maximum power, seamlessly access GPT-4, Claude, Gemini, and 100+ other models via OpenRouter integration.

Local Execution
Native support for GGUF model formats using llama.cpp. Run models like Llama 3, Mistral, Gemma, Phi, and more entirely on-device with optimized quantization for mobile hardware.

Cloud Orchestration
Unified API integration through OpenRouter provides instant access to 100+ state-of-the-art models without vendor lock-in or multiple subscriptions.

Intelligent Streaming
Real-time token generation with context-aware memory management ensures smooth performance whether running locally or in the cloud.

Stable Diffusion 1.5
Full SD 1.5 implementation running completely offline on your phone. Generate high-quality images in 30-90 seconds depending on your device.

Censored & Uncensored Options
Choose between SFW (censored) or uncensored models for artistic freedom and research applications.

Optimized for Mobile
Specially quantized and optimized to run on phones with 6GB+ RAM while maintaining image quality.

Dynamic Knowledge Injection
Mount custom datasets (JSON, text, markdown) to enhance AI responses with specialized knowledge without retraining models.

Use Cases:

• Inject Wikipedia dumps for educational queries
• Load coding documentation for development assistance
• Add personal notes or company data for context-aware responses
• Import research papers or domain-specific knowledge

Plugin Integration
Data-Packs work seamlessly with both local GGUF models and cloud models for maximum flexibility.

Text-to-Speech (TTS)
Powered by Sherpa-ONNX, includes 11 professional-grade voices (5 American Female, 2 American Male, 2 British Female, 2 British Male) running entirely on CPU/NPU with zero cloud dependencies.

Speech-to-Text (STT)
Offline Whisper-powered speech recognition for hands-free AI interaction. Perfect for driving, multitasking, or accessibility needs.

Zero Latency
All voice processing happens on-device with near-instantaneous synthesis and recognition.

Available Now:

• Web Search — Real-time information retrieval with search engine integration
• Web Scraper — Extract and inject content from any URL into conversation context
• DataHub — Mount and manage custom knowledge bases dynamically
• Document Viewer — Analyze and discuss PDF/text documents with AI

Coming Soon:

• Code execution environments
• Advanced image processing pipelines
• Multi-document analysis
• Custom plugin marketplace

• Conversation Persistence — Full chat history with efficient SQLite storage
• Dynamic Datasets — Attach custom knowledge without model retraining
• Context Preservation — Switch models mid-conversation without losing thread
• Export Options — Save conversations, code snippets, and generated images
• Multi-Session — Manage multiple conversation threads simultaneously

Chat Interface Multi-modal conversations	Model Hub 100+ models available	Code Canvas Syntax highlighting & export	Settings Complete customization

Visit ToolNeuron on APKPure for the latest stable release with automatic update notifications.

Download the latest release from GitHub Releases and install ToolNeuron-Beta-5.1.apk on Android 8.0+ devices.

# Clone repository
git clone https://github.com/Siddhesh2377/NeuroVerse.git
cd NeuroVerse

# Open in Android Studio (Ladybug or newer)
# Sync Gradle dependencies
./gradlew assembleDebug

# Install on connected device
./gradlew installDebug

1. Load a Chat Model (GGUF)

• Download a GGUF model from HuggingFace
  • Recommended: Llama-3-8B-Q4_K_M.gguf (4.5GB)
  • Budget: TinyLlama-1.1B-Q4_K_M.gguf (669MB)
• Navigate to Settings → Local Models → Import Model
• Select your downloaded GGUF file
• Wait for model to load, then start chatting offline!

2. Load Image Generation (Stable Diffusion)

• Download SD 1.5 model (censored or uncensored version)
• Navigate to Settings → Image Models → Import SD Model
• Select model file
• Generate images completely offline!

3. Enable Voice AI

• TTS voices are included by default (no download needed)
• For STT: Download Whisper model from Settings → Voice Models
• Enable voice input in chat interface

4. Create RAG Data-Packs

• Prepare your knowledge in JSON/text format
• Navigate to DataHub → Create New Pack
• Import your data files
• Attach to conversations for enhanced context

1. Visit OpenRouter.ai and create account
2. Generate an API key (free tier available)
3. In ToolNeuron: Settings → API Configuration
4. Enter your OpenRouter API key
5. Access 100+ models instantly (GPT-4, Claude, Gemini, etc.)

Simply switch between local and cloud models mid-conversation:

• Use offline LLM for privacy-sensitive queries
• Switch to GPT-4 for complex reasoning tasks
• Return to offline for continued privacy
• All context is preserved automatically!

• Operating System: Android 8.0+ (API 26)
• RAM: 4GB
• Storage: 2GB available space
• Use Case: Cloud models + basic TTS only

• Operating System: Android 10+
• RAM: 6GB+ (8GB preferred)
• Processor: Snapdragon 8 Gen 1 / Dimensity 8100 or equivalent
• Storage: 5GB+ available space
• NPU: Optional but improves performance significantly

• Operating System: Android 11+
• RAM: 8GB minimum (12GB preferred)
• Processor: Snapdragon 8 Gen 2 or equivalent flagship
• Storage: 8GB+ available space (for SD models)
• Generation Time: 30-90 seconds depending on device

• RAM: 12GB+
• Processor: Snapdragon 8 Gen 3 or equivalent
• Storage: 10GB+ free space
• Experience: Smooth chat + image generation + voice AI

• ✅ GGUF model support with llama.cpp
• ✅ 11 offline TTS voices via Sherpa-ONNX
• ✅ OpenRouter cloud integration (100+ models)
• ✅ Plugin system (Web Search, Scraper, DataHub)
• ✅ RAG Data-Packs for knowledge injection
• 🚧 Stable Diffusion 1.5 offline image generation
• 🚧 Offline Whisper STT integration

• Multi-voice TTS conversations (different voices for different characters)
• Advanced code export with syntax highlighting
• Desktop companion app (Windows/Linux sync)
• Enhanced plugin marketplace
• Vector database for long-term memory

• Multi-modal vision models (LLaVA, GPT-4V integration)
• TFLite and ONNX runtime support
• On-device video analysis
• Collaborative AI sessions
• Custom model fine-tuning tools

• Cross-platform synchronization (phone ↔ desktop)
• Community plugin marketplace
• Advanced RAG with semantic search
• Enterprise deployment options
• API for third-party integration

• Test prompts and APIs without cloud costs during development
• Run coding assistants offline on flights or with poor connectivity
• Inject documentation into RAG for context-aware code help
• Generate UI mockups and diagrams with SD
• Privacy-first development environment

• Zero data leaves your device in offline mode
• Verify privacy claims (open-source Apache 2.0)
• No tracking, no telemetry, no server logging
• Full control over your AI interactions
• Uncensored options for research and legitimate use

• Full AI capability on flights (no WiFi needed)
• Works in remote areas with no connectivity
• No roaming data costs for AI queries
• Generate travel content (images, itineraries) offline
• Voice translations without cloud latency

• Generate images for social media posts anywhere
• Brainstorm content ideas with AI offline
• Create variations and iterations without API limits
• No subscription costs eating into creator budgets
• Uncensored artistic freedom

• Free access to cutting-edge AI models
• Study AI without expensive subscriptions
• Load research papers into RAG for analysis
• Generate diagrams and visualizations
• Privacy for sensitive academic work

ToolNeuron implements modern Android development patterns with a hybrid native/Kotlin architecture:

Core Technologies:

• Language: Kotlin (UI/Logic) + C++ (Inference engines)
• UI Framework: Jetpack Compose (declarative, reactive UI)
• Local Inference: llama.cpp (GGUF models) + JNI bindings
• Image Generation: Stable Diffusion C++ implementation
• TTS Engine: Sherpa-ONNX (neural voices)
• STT Engine: Whisper via Sherpa-ONNX
• API Layer: Retrofit + OkHttp (cloud models)
• Database: Room (SQLite wrapper) for conversations
• Async Operations: Kotlin Coroutines + Flow
• Dependency Injection: Hilt/Dagger

Performance Optimizations:

• Quantized model support (Q4_K_M, Q5_K_S, etc.)
• Context caching for faster inference
• Memory-mapped model loading
• NPU acceleration where available
• Efficient token streaming
• Background processing with WorkManager

We welcome contributions from developers, researchers, AI enthusiasts, and privacy advocates!

1. Fork the repository
2. Create a feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes with descriptive messages
4. Push to your branch (git push origin feature/AmazingFeature)
5. Open a Pull Request with detailed description

High Priority:

• 🐛 Bug reports and fixes (especially device-specific issues)
• 📚 Documentation improvements and translations
• 🧪 Testing on various Android devices and chipsets
• 🔌 New plugin development
• 🎨 UI/UX enhancements

Medium Priority:

• 🌍 Internationalization (i18n) - help us support more languages
• ♿ Accessibility improvements
• 📊 Performance optimizations
• 🎓 Tutorial content and guides

Feature Requests:

• Check existing issues before creating new ones
• Provide clear use cases and examples
• Be patient - we're a small team!

• Follow Kotlin coding conventions
• Write meaningful commit messages
• Test on real devices when possible
• Document new features
• Respect user privacy in all contributions

Distributed under the Apache 2.0 License. See LICENSE for complete terms.

What this means:

• ✅ Commercial use - Use in commercial products
• ✅ Modification - Modify and create derivatives
• ✅ Distribution - Distribute freely
• ✅ Patent use - License includes patent rights
• ✅ Private use - Use privately without restrictions

Requirements:

• 📄 Include license and copyright notice
• 📝 Document any changes made
• 🔓 Make source available if distributing

"If I have seen further, it is by standing on the shoulders of giants." — Isaac Newton

ToolNeuron would not be possible without these exceptional open-source projects:

• llama.cpp by Georgi Gerganov — Efficient LLM inference in pure C/C++
• Sherpa-ONNX — Premium offline speech synthesis and recognition
• Stable Diffusion — Revolutionary text-to-image generation
• OpenRouter — Unified API gateway for 100+ AI models
• Jetpack Compose — Modern declarative UI for Android
• HuggingFace — Community and models that make AI accessible

Special thanks to the open-source AI community for making privacy-respecting AI possible.

• 💬 Discord Community — Real-time chat, support, and discussions
• 🐛 Issue Tracker — Report bugs or request features
• 💡 GitHub Discussions — Technical questions and ideas
• 📧 Email: Support — For private inquiries

• ⭐ Star this repository to show support and get updates
• 👀 Watch releases for new features and updates
• 🐦 Follow on Twitter @ToolNeuron — News and announcements
• 📱 APKPure — Automatic update notifications

Q: Will this drain my battery?
A: Local inference is power-intensive. For long sessions, keep your phone plugged in. Cloud mode uses minimal battery.

Q: How big are the model files?
A: GGUF models: 0.5GB-8GB depending on model size. SD 1.5: ~2GB. TTS/STT: 50-500MB.

Q: Can I use my own API keys?
A: Yes! BYOK (Bring Your Own Key) for OpenRouter. You control costs and usage.

Q: Is my data really private?
A: In offline mode, absolutely nothing leaves your device. Verify in our open-source code.

Q: Why not just use ChatGPT?
A: ToolNeuron gives you choice, privacy, offline capability, uncensored options, and zero subscriptions.

Q: Does it support iOS?
A: Not currently. Android only due to technical constraints of iOS.

Q: Can I monetize apps built with this?
A: Yes! Apache 2.0 license allows commercial use.