VA21 OS - Secure AI-Powered Linux Operating System
Om Vinayaka 🙏 - The Remover of Obstacles
🛡️ A Complete Linux Operating System
VA21 OS is a full Linux operating system designed for researchers, writers, journalists, and security professionals. It's a real OS that you install on your computer - not an app, not a container.
🙏 Om Vinayaka AI at the CORE - Controls everything except Guardian AI (sandboxed security)
What is VA21 OS?
- Full Linux OS based on Debian GNU/Linux
- Om Vinayaka AI at the core - intelligent accessibility controlling all subsystems
- Zork-style text adventure interface for EVERY application
- Local AI via Ollama - Privacy-first, your data never leaves your device
- 1,600+ language support via Meta Omnilingual ASR
- Guardian AI security powered by IBM Granite 4.0 (sandboxed at kernel level)
- ~5 GB base installation (AI models downloaded during install)
Installation
Download the ISO from the Releases page and install like any Linux distribution:
| Edition | Description | Size |
|---|---|---|
| VA21 OS Debian | Full GNU toolkit, maximum compatibility | ~5 GB |
| VA21 OS Alpine | Lightweight, fast boot | ~2 GB |
# Write ISO to USB drive
sudo dd if=va21-os.iso of=/dev/sdX bs=4M status=progress
# Or use in VirtualBox/VMware
# Create VM → Use ISO as boot disk → Install
Quick Start (From Source)
git clone https://github.com/narasimhudumeetsworld/va21.git
cd va21/va21_system/linux_os
./scripts/build_iso.sh debian
# ISO will be in output/ directory
🏗️ Complete System Architecture
┌─────────────────────────────────────────────────────────────────────────┐
│ VA21 OS - Complete Architecture │
├─────────────────────────────────────────────────────────────────────────┤
│ 🛡️ KERNEL LEVEL - Guardian AI (Sandboxed Ollama - Port 11435) │
│ ├── IBM Granite 4.0 2B - Isolated security AI │
│ ├── Completely separate from user-facing AI │
│ ├── Cannot be influenced by user conversations │
│ ├── Think → Vet → Act methodology │
│ └── Independent threat analysis and blocking │
├─────────────────────────────────────────────────────────────────────────┤
│ 🙏 OM VINAYAKA AI - CORE CONTROLLER (User-facing Ollama - Port 11434) │
│ ├── CONTROLS ALL SUBSYSTEMS (except Guardian) │
│ ├── Automatic Zork UX for every app when installed │
│ ├── Automatic FARA profiles for voice control of ANY app │
│ ├── System-wide voice control for ALL applications │
│ ├── CLI tool wrappers (Gemini, Copilot, Codex, etc.) │
│ ├── LangChain + Obsidian mind maps knowledge base │
│ ├── Self-learning system that improves over time │
│ ├── Clarifying questions to understand user intent │
│ ├── Performance optimizer with model preloading │
│ ├── Feature discovery for new users │
│ └── FARA layer executes actions across entire OS │
├─────────────────────────────────────────────────────────────────────────┤
│ 🎤 Voice Intelligence Layer (Controlled by Om Vinayaka) │
│ ├── Meta Omnilingual ASR (1,600+ langs) - Apache 2.0 │
│ ├── Whisper/Solus AI (backup) - MIT │
│ ├── Rhasspy (wake words) - MIT │
│ ├── Piper TTS (fast) - MIT │
│ ├── Kokoro TTS (premium) - Apache 2.0 │
│ └── Hold Super Key = Push-to-Talk (Accessibility) │
├─────────────────────────────────────────────────────────────────────────┤
│ 🤖 Multi-Agent System (Controlled by Om Vinayaka) │
│ ├── Automatic role assignment (Coder, Reviewer, Planner, etc.) │
│ ├── Experience levels (Junior, Mid, Senior, Expert) │
│ ├── Context summaries for each agent │
│ ├── Microsoft AutoGen patterns - MIT │
│ └── LangChain orchestration - MIT │
├─────────────────────────────────────────────────────────────────────────┤
│ 🎮 Zork-Style Interface (Custom Created!) │
│ ├── Text adventure UI for EVERY application │
│ ├── Automatic interface generation on app install │
│ ├── Native tiling window manager │
│ ├── Obsidian-style Knowledge Vault │
│ └── Research Command Center │
├─────────────────────────────────────────────────────────────────────────┤
│ 💻 Sandboxed Terminals with Zork Accessibility │
│ ├── Gemini CLI, Codex, GitHub Copilot CLI support │
│ ├── Natural language to CLI command translation │
│ ├── Zork narrative wrapping for all output │
│ └── Voice control for terminal applications │
├─────────────────────────────────────────────────────────────────────────┤
│ 🧠 Self-Learning System │
│ ├── Learns common command patterns │
│ ├── Tracks user preferences │
│ ├── Monitors app usage patterns │
│ ├── Improves narratives over time │
│ └── Gets smarter with continued use! │
├─────────────────────────────────────────────────────────────────────────┤
│ 💾 Debian GNU/Linux Foundation │
│ ├── Full GNU toolkit │
│ ├── Debian package management │
│ ├── Flatpak integration │
│ └── ISO releases for real hardware install │
└─────────────────────────────────────────────────────────────────────────┘
✨ Features
🎮 Zork-Style Interface
VA21 OS features a unique text adventure interface inspired by classic Zork:
========================================
VA21 Research OS v1.0 (Vinayaka)
========================================
You awaken in the BOOT CHAMBER. The ancient Guardian AI
stirs to life, its amber eyes watching you carefully.
"Welcome, Researcher," the Guardian speaks.
> look
You are in the BOOT CHAMBER.
Exits: NORTH (Research Lab), EAST (Knowledge Vault)
> go north
You enter the RESEARCH LAB...
🎤 Voice Intelligence Layer (Meta Omnilingual ASR)
Complete multilingual voice processing with 1,600+ language support including 100+ Indian dialects:
| Component | Technology | License | Purpose |
|---|---|---|---|
| ASR (Primary) | Meta Omnilingual ASR ⭐⭐⭐⭐⭐ | Apache 2.0 | 1,600+ languages (Released Nov 2025) |
| ASR (Backup) | OpenAI Whisper | MIT | Offline backup ASR |
| Wake Word | Rhasspy | MIT | Custom wake word detection |
| TTS (Fast) | Piper TTS | MIT | Fast speech synthesis |
| TTS (Premium) | Kokoro TTS | Apache 2.0 | Premium natural voices |
| Local LLM | Ollama + IBM Granite | MIT + Apache 2.0 | Privacy-first local AI |
Indian Language Support: Hindi, Tamil, Telugu, Kannada, Bengali, Marathi, Gujarati, Malayalam, Punjabi, Odia, Assamese, and 100+ more dialects!
♿ Intelligent Accessibility with Om Vinayaka AI (Unique to VA21!)
VA21's accessibility goes far beyond traditional screen readers. Powered by the Om Vinayaka Accessibility Knowledge Base AI:
| Traditional Screen Readers | VA21 + Om Vinayaka AI |
|---|---|
| Reads keywords: "menu", "button" | Explains purpose: "This saves your work" |
| No context awareness | Understands your intent and current task |
| Just announces elements | Asks clarifying questions when needed |
| User must know commands | Natural conversation in any language |
| Single app support | Zork-style UX for EVERY app |
| Limited CLI support | Wraps CLI tools like Gemini, Copilot, Codex |
Om Vinayaka AI Features:
- Automatic Zork UX Generation: Every app gets a text adventure interface when installed
- System-Wide Voice Control: Control ANY application with voice, not just specific apps
- CLI Tool Wrapper: Gemini CLI, GitHub Copilot CLI, Codex, and more - all accessible via Zork interfaces
- Knowledge Base: LangChain + Obsidian mind maps store all app interfaces
- Clarifying Questions: AI asks for details when your intent is unclear
- Context-Aware Execution: Understands what app is active and what you want
How it works:
- Hold Super Key: Activates voice detection
- Speak naturally: "I want to search the internet" or "Save my document"
- Om Vinayaka AI understands: Asks clarifying questions if needed
- FARA Layer executes: Performs action in any application
- 1,600+ languages: Hindi, Tamil, Telugu, Spanish, French, and more
Example conversation:
User: "I want to find something on the internet"
VA21: "I can help you search. What would you like to look up?"
User: "Climate change research papers"
VA21: "Searching for climate change research papers. I'm using
privacy-respecting search so your query isn't tracked."
CLI Tool Accessibility Example:
User: "Ask Gemini about Python decorators"
VA21: "You stand before the GEMINI ORACLE, a shimmering portal of AI wisdom.
The oracle considers your question deeply...
[Gemini's response about Python decorators]
What else would you like to ask?"
🔒 Guardian AI Security (Sandboxed)
Powered by IBM Granite 4.0 via sandboxed Ollama in the kernel:
- Complete Isolation: Guardian AI runs separately from user-facing AI
- Think → Vet → Act methodology
- Real-time threat analysis
- ClamAV antivirus integration
- Air gap browser protection
- Cannot be influenced by user conversations
🔒 AI Privacy & Security System
Powered by IBM AI Privacy Toolkit (MIT) + LLM Guard (MIT):
- Data Anonymization: GDPR-compliant AI processing
- Prompt Injection Detection: Blocks malicious inputs
- Output Scanning: Filters harmful content
🤖 Multi-Agent System with Automatic Role Assignment (NEW!)
VA21 OS includes a powerful multi-agent system that works seamlessly with Om Vinayaka AI:
Automatic Agent Assignment: | Role | Experience | Responsibilities | |------|------------|------------------| | Orchestrator | Expert (10+ years) | Coordinates all agents, assigns tasks | | Planner | Senior (5-10 years) | Breaks down complex tasks, estimates effort | | Coder | Mid-Senior | Writes clean, efficient code | | Reviewer | Senior | Reviews code for quality and security | | Researcher | Mid | Researches topics, gathers information | | Debugger | Senior | Identifies and fixes bugs |
How it works:
- You describe what you need - "Build me a web scraper for news articles"
- Om Vinayaka AI understands - Asks clarifying questions if needed
- Planner breaks it down - Creates actionable task list
- Agents are auto-assigned - Based on task type and complexity
- Work is executed - Each agent focuses on their specialty
- Results are unified - Om Vinayaka presents the final output
Local AI with Ollama (Privacy-First):
# VA21 uses local Ollama by default - your data never leaves your device!
# Models supported:
- granite3.3:2b # General tasks (IBM Granite)
- granite-code:3b # Code generation
- granite-guardian:2b # Security (sandboxed)
# Or use external APIs (optional):
export OPENAI_API_KEY="your-key" # OpenAI
export ANTHROPIC_API_KEY="your-key" # Anthropic Claude
export GROQ_API_KEY="your-key" # Groq (fast)
Agent Cooperation Architecture:
┌─────────────────────────────────────────────────────────────────┐
│ 🙏 OM VINAYAKA AI (Hub) │
│ Understands intent, routes to agents, learns │
├─────────────────────────────────────────────────────────────────┤
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │Orchestr. │ │ Planner │ │ Coder │ │ Reviewer │ │
│ │ (Expert) │ │ (Senior) │ │ (Mid) │ │ (Senior) │ │
│ └──────────┘ └──────────┘ └──────────┘ └──────────┘ │
│ │ │ │ │ │
│ └───────────┴────────────┴────────────┘ │
│ LOCAL OLLAMA (Privacy) │
│ or External APIs (Optional) │
└─────────────────────────────────────────────────────────────────┘
🧠 Dynamic AI Quantization System
- Adaptive Performance: AI models automatically quantize to 4-bit, 5-bit, or 8-bit based on available system memory
- Memory-Aware Loading: Intelligent model loading and unloading
- Quality Preservation: Optimal quality-to-memory tradeoffs
📚 Tiered Memory System with Obsidian Brain Maps (NEW!)
- Three-Tier Architecture:
- Working Memory (current context)
- Short-Term Memory (session)
- Long-Term Memory (persistent)
- Separate Knowledge Bases: Each AI (Guardian, Helper, Agent) has its own memory space
- Interconnected Context: Tiered tags and context-aware retrieval
- Anti-Hallucination Engine: Validates all AI memory retrieval
🎨 Futuristic Interface & Easter Eggs
- Halo/Cortana Theme: Type "cortana call the masterchief" to activate a stunning holographic interface inspired by the Halo universe
- Dark/Light Theme Toggle: Switch between themes with Ctrl+Shift+T
- Command Palette: Quick access to all commands with Ctrl+K
- Minimal Toggle Option: Easy theme switching from settings
💾 Advanced Backup with Version History
- Automatic Periodic Backups: Never lose your work again
- Pre-Reset Safety Backups: Easily restore during system reset
- Version History Timeline: Browse and restore from any point in time
- Helper AI Integration: Ask the AI to restore from backups naturally
- Compression & Auto-Cleanup: Efficient storage management
📦 App Center with Flathub & Debian Integration
- One-Click App Installation: Search, preview, and install apps easily
- Flatpak Support: Access thousands of sandboxed applications via Flathub
- Debian Packages: Full access to the vast Debian repository
- Spotlight-Like Search: Find and install apps right from the command palette
🔌 FARA App Compatibility Layer (Alpha)
- Microsoft FARA-Inspired Technology: Intelligent UI automation for legacy apps
- Screenshot-Based Analysis: Visual understanding of application interfaces
- Legacy GTK2/Qt4 Support: Seamless integration with older applications
- Wine Compatibility: Run Windows applications with FARA assistance
- Context-Aware Automation: AI-driven action planning for complex workflows
🤖 Intelligent Helper AI
- Backup Knowledge: AI knows your version history (sanitized for security)
- Natural Language Restoration: Just ask "restore from yesterday"
- System Awareness: Get status updates and recommendations
- Task Automation: Request automated tasks through natural conversation
- Easter Egg Activation: Fun interactions unlock special features
🏗️ VA21 OS - Complete Architecture
┌─────────────────────────────────────────────────────────────────────────┐
│ VA21 OS - Complete Architecture │
├─────────────────────────────────────────────────────────────────────────┤
│ 🛡️ KERNEL LEVEL - Guardian AI (Sandboxed Ollama - Port 11435) │
│ ├── IBM Granite 4.0 2B - Isolated security AI │
│ ├── Completely separate from user-facing AI │
│ ├── Cannot be influenced by user conversations │
│ ├── Think → Vet → Act methodology │
│ └── Independent threat analysis and blocking │
├─────────────────────────────────────────────────────────────────────────┤
│ 🙏 OM VINAYAKA ACCESSIBILITY AI (User-facing Ollama - Port 11434) │
│ ├── Automatic Zork UX for every app when installed │
│ ├── System-wide voice control for ALL applications │
│ ├── CLI tool wrappers (Gemini, Copilot, Codex, etc.) │
│ ├── LangChain + Obsidian mind maps knowledge base │
│ ├── Self-learning system that improves over time │
│ ├── Clarifying questions to understand user intent │
│ └── FARA layer executes actions across entire OS │
├─────────────────────────────────────────────────────────────────────────┤
│ 🎤 Voice Intelligence Layer │
│ ├── Meta Omnilingual ASR (1,600+ langs) - Apache │
│ ├── Whisper/Solus AI (backup) - MIT │
│ ├── Rhasspy (wake words) - MIT │
│ ├── Piper TTS (fast) - MIT │
│ ├── Kokoro TTS (premium) - Apache 2.0 │
│ └── Hold Super Key = Push-to-Talk (Accessibility) │
├─────────────────────────────────────────────────────────────────────────┤
│ 🤖 Multi-Agent System │
│ ├── Microsoft AutoGen - MIT │
│ ├── Agent Zero patterns - MIT │
│ ├── OpenCode patterns (Build/Plan) - MIT │
│ └── LangChain orchestration - MIT │
├─────────────────────────────────────────────────────────────────────────┤
│ 🎮 Zork-Style Interface (Custom Created!) │
│ ├── Text adventure UI for EVERY application │
│ ├── Automatic interface generation on app install │
│ ├── Native tiling window manager │
│ ├── Obsidian-style Knowledge Vault │
│ └── Research Command Center │
├─────────────────────────────────────────────────────────────────────────┤
│ 💻 Sandboxed Terminals with Zork Accessibility │
│ ├── Gemini CLI, Codex, GitHub Copilot CLI support │
│ ├── Natural language to CLI command translation │
│ ├── Zork narrative wrapping for all output │
│ └── Voice control for terminal applications │
├─────────────────────────────────────────────────────────────────────────┤
│ 🧠 Self-Learning System │
│ ├── Learns common command patterns │
│ ├── Tracks user preferences │
│ ├── Monitors app usage patterns │
│ ├── Improves narratives over time │
│ └── Gets smarter with continued use! │
├─────────────────────────────────────────────────────────────────────────┤
│ 💾 Debian GNU/Linux Foundation │
│ ├── Full GNU toolkit │
│ ├── Debian package management │
│ ├── Flatpak integration │
│ └── ISO releases for real hardware install │
└─────────────────────────────────────────────────────────────────────────┘
🔒 Core Security Architecture
Triple-AI System (Om Vinayaka at Core)
┌─────────────────────────────────────────────────────────────────────────┐
│ SECURITY ARCHITECTURE │
├─────────────────────────────────────────────────────────────────────────┤
│ │
│ 🛡️ GUARDIAN AI (Kernel Level - Port 11435) [ISOLATED] │
│ ├── IBM Granite Guardian 2B model │
│ ├── Think → Vet → Act methodology │
│ ├── CANNOT be influenced by user conversations │
│ └── Independent threat analysis │
│ │
│ ════════════════════ ISOLATION BARRIER ════════════════════ │
│ │
│ 🙏 OM VINAYAKA AI (Core Controller - Port 11434) │
│ ├── Controls ALL user-facing subsystems │
│ ├── Intelligent accessibility for 1,600+ languages │
│ ├── Self-learning system │
│ ├── Anti-Hallucination Engine │
│ │ │
│ │ ┌─────────┬─────────┬─────────┬─────────┬─────────┐ │
│ └───┤ Agents │ Research│ Writing │ Coding │ System │ │
│ └─────────┴─────────┴─────────┴─────────┴─────────┘ │
│ │
└─────────────────────────────────────────────────────────────────────────┘
- Guardian AI (Security Core): Sandboxed at kernel level - ISOLATED from Om Vinayaka
- Om Vinayaka AI (Core Controller): Controls all user-facing systems with intelligent accessibility
- Multi-Agent System: Task automation controlled BY Om Vinayaka AI
- Air Gap Protection: Complete isolation between Guardian and user-facing AI
- Principle of Least Privilege: No direct website interactions or form submissions
Security Features
- ✅ Real-time Threat Analysis: Every input analyzed by Guardian before processing
- ✅ Self-Analysis & Healing: Autonomous code security scanning
- ✅ 5-Day Quarantine Protocol: Safe integration of external intelligence
- ✅ Localhost-Only Operation: No external network exposure
- ✅ Pattern-Based Detection: Advanced malware and injection detection
- ✅ Anti-Hallucination Engine: Om Vinayaka prevents AI from making up information
- ✅ Om Vinayaka at Core: All user interactions flow through intelligent accessibility
🔬 Research Command Center
Multiple Sandboxed Terminals
- Tiling Window Management: Run multiple terminal sessions with quad, triple, or six-pane layouts
- Isolation Levels: Minimal, Standard, and Strict sandbox levels
- Session Logging: All terminal activity logged to the knowledge vault
Obsidian-Style Knowledge Vault
- Wiki-Style Links: Create interconnected notes with
[[link]]syntax - Knowledge Graph: Visual representation of your research connections
- LLM Memory Integration: Persistent memory for AI context
- Brain Maps: Visual memory maps for each AI component
Sensitive Information Protection
- Automatic Redaction: Detect and redact API keys, passwords, tokens
- Category-Based Filtering: Control what gets redacted
🧠 Context-Aware Summary Engine
Prevents AI hallucinations and context overflow:
- Automatic Summarization: When AI context gets too large
- Full Preservation: Complete knowledge stored in Obsidian vault
- Smart Compression: Prioritizes critical context over background info
- Hallucination Prevention: AI never loses important context
💻 System Requirements
Minimum Requirements
| Component | Requirement |
|---|---|
| RAM | 4 GB |
| CPU | 2 cores |
| Storage | 20 GB |
| Note | AI features will be limited |
Recommended Requirements
| Component | Requirement |
|---|---|
| RAM | 8 GB |
| CPU | 4 cores |
| Storage | 40 GB |
| Note | Full AI experience |
Optimal Requirements
| Component | Requirement |
|---|---|
| RAM | 16 GB |
| CPU | 8 cores |
| Storage | 80 GB |
| GPU | NVIDIA with CUDA (optional) |
| Note | Best performance with all features |
RAM Breakdown (Full Installation)
| Component | RAM (MB) |
|---|---|
| Base OS (Debian) | 512 |
| Desktop Environment | 256 |
| Guardian AI (Sandboxed Ollama) | 2048 |
| User-facing Ollama | 2048 |
| Om Vinayaka AI | 256 |
| Voice Recognition | 512 |
| Text-to-Speech | 256 |
| Summary Engine | 64 |
| Total Calculated | ~6 GB |
🚀 Quick Installation
From ISO (Recommended)
Download the VA21 OS ISO from the Releases page and install like any Linux distribution:
# Write ISO to USB drive
sudo dd if=va21-os.iso of=/dev/sdX bs=4M status=progress
# Or use in VirtualBox/VMware
# Create VM → Use ISO as boot disk → Install
Build from Source
git clone https://github.com/narasimhudumeetsworld/va21.git
cd va21/va21_system/linux_os
./scripts/build_iso.sh debian
# ISO will be in output/ directory
📋 Additional System Requirements
Disk Space Requirements
- VA21 OS Base Installation: ~5 GB (includes Zork UI, Guardian AI, all features)
- With All AI Models: ~10 GB (full model suite downloaded during install)
Note: Guardian AI model is always included. Other AI models are downloaded on-demand during first use. See docs/MODEL_REQUIREMENTS.md for detailed breakdown.
Minimum Requirements (7GB RAM)
- Debian-based Linux (Debian 12+, Ubuntu 22.04+, Linux Mint 21+)
- Python 3.8+
- 7GB RAM (for standard usage with AI features)
- 5GB disk space for VA21 OS base installation
- Flatpak (optional, for App Center)
Recommended Requirements (10GB RAM)
- 8-10GB RAM (for heavy multitasking with all AI models)
- 10GB disk space for full installation with all AI models
- Modern CPU (Intel Core i5/AMD Ryzen 5 or better)
- SSD storage for faster model loading
Memory Optimization
VA21 uses dynamic context-aware AI activation to minimize RAM usage:
- 🧠 Lazy Loading - Models loaded only when needed
- 📦 Dynamic Quantization - 4-bit, 5-bit, or 8-bit based on available RAM
- 🔄 Context-Aware Unloading - Automatic memory management
- 💾 Memory Mapping - Efficient large file handling
| Usage Scenario | RAM Required | Quantization | Description |
|---|---|---|---|
| Minimal | ~3GB | 4-bit | Basic browsing, text editing |
| Standard | ~5GB | 5-bit | Multiple apps, AI chat |
| Heavy Multitasking | ~7GB | 8-bit | Many apps, FARA compatibility |
| Full Development | ~9GB | 8-bit/FP16 | All AI features, IDE, Docker |
⌨️ Keyboard Shortcuts
| Shortcut | Action |
|---|---|
Ctrl+K |
Open Command Palette |
Ctrl+B |
Toggle Side Panel |
Ctrl+Shift+T |
Toggle Theme |
Ctrl+Shift+S |
Create Backup |
Ctrl+Shift+R |
Open Restore Dialog |
🎮 Easter Eggs
- Type "cortana call the masterchief" in the chat to unlock the Halo interface theme!
- More secrets to discover...
🏗️ Technical Stack
VA21 OS is built with privacy-first, local-first technologies:
Core System
| Component | Technology | License |
|---|---|---|
| OS Base | Debian GNU/Linux | DFSG |
| Window Manager | Custom Tiling WM | Apache 2.0 |
| Shell | Zork-Style Interface | Apache 2.0 |
| Package Manager | Flatpak + APT | Various |
🙏 Om Vinayaka AI Stack (Central Intelligence)
| Component | Technology | License | Purpose |
|---|---|---|---|
| Local LLM | Ollama | MIT | Privacy-first local AI |
| AI Models | IBM Granite | Apache 2.0 | General + Code + Security |
| Orchestration | LangChain | MIT | AI workflow management |
| Knowledge Base | Obsidian-style Vault | - | Mind maps + Wiki links |
| FARA Layer | Microsoft FARA Integration | MIT | Universal app control |
🎤 Voice Intelligence Layer
| Component | Technology | License | Purpose |
|---|---|---|---|
| ASR (Primary) | Meta Omnilingual ASR ⭐⭐⭐⭐⭐ | Apache 2.0 | 1,600+ languages |
| ASR (Backup) | OpenAI Whisper | MIT | Offline backup |
| Wake Word | Rhasspy | MIT | "Hey VA21" detection |
| TTS (Fast) | Piper TTS | MIT | Fast synthesis |
| TTS (Premium) | Kokoro TTS | Apache 2.0 | Natural voices |
🤖 Multi-Agent System
| Component | Technology | License | Purpose |
|---|---|---|---|
| Framework | Microsoft AutoGen ⭐⭐⭐⭐⭐ | MIT | Agent conversation |
| Patterns | Agent Zero | MIT | Cooperation patterns |
| Roles | OpenCode | MIT | Build/Plan agents |
| Deployment | LMDeploy | Apache 2.0 | Efficient inference |
🔒 Security & Privacy
| Component | Technology | License | Purpose |
|---|---|---|---|
| Guardian AI | Sandboxed Ollama (port 11435) | Proprietary | Kernel-level security |
| LLM Security | LLM Guard | MIT | Prompt injection defense |
| Privacy | IBM AI Privacy Toolkit | MIT | GDPR compliance |
| Antivirus | ClamAV | GPL | Malware protection |
🏢 Key Technology Partners
| Partner | Contributions |
|---|---|
| IBM | Granite AI models, AI Privacy Toolkit |
| Microsoft | ONNX Runtime, FARA, AutoGen, Phi models |
| Meta AI | Omnilingual ASR (1,600+ languages), LLaMA, PyTorch |
| Hugging Face | Model hosting, Transformers library |
| Ollama | Local LLM deployment |
🙏 Acknowledgments
VA21 OS is built on the shoulders of giants:
Voice Intelligence Layer
- Meta Omnilingual ASR - [Apache 2.0] ⭐⭐⭐⭐⭐ 1,600+ languages including 100+ Indian dialects (Released November 2025)
- OpenAI Whisper - github.com/openai/whisper (MIT) - Offline backup ASR
- Rhasspy - github.com/rhasspy/rhasspy (MIT) - Wake word detection
- Piper TTS - github.com/rhasspy/piper (MIT) - Fast TTS synthesis
- Kokoro TTS - github.com/remsky/Kokoro-FastAPI (Apache 2.0) - Premium voices
Security & Privacy
- IBM AI Privacy Toolkit - github.com/IBM/ai-privacy-toolkit (MIT) - GDPR compliance
- LLM Guard - github.com/protectai/llm-guard (MIT) - LLM security
- LMDeploy - github.com/InternLM/lmdeploy (Apache 2.0) - Efficient LLM deployment
Multi-Agent Automation
- Microsoft AutoGen - github.com/microsoft/autogen (MIT) ⭐⭐⭐⭐⭐ Multi-agent conversation framework (Backed by Microsoft Research)
- Agent Zero - github.com/agent0ai/agent-zero (MIT) Multi-agent cooperation patterns
- OpenCode - github.com/sst/opencode (MIT) Role-based agents (Build/Plan)
LLM & AI Processing
- LangChain - github.com/langchain-ai/langchain (MIT) - AI orchestration
- IBM Granite - huggingface.co/ibm-granite (Apache 2.0)
- Microsoft - ONNX Runtime, FARA, AutoGen, Phi models (MIT)
- Meta AI - Omnilingual ASR (Apache 2.0), LLaMA, PyTorch
- Hugging Face - Democratizing AI model access
- Ollama - Simple local LLM deployment (MIT)
Open Source Foundation
- Debian Project - Rock-solid OS foundation
- Linux Foundation - Linux kernel stewardship
- Linux Community - Every contributor who makes open source possible
- Flathub - Revolutionizing Linux application distribution
📄 License
This project is licensed under a modified Apache License 2.0 with proprietary components.
- Permissive (Apache 2.0): Most features with acknowledgment
- Proprietary: Guardian AI, Think>Vet>Act, Sandbox Testing, Anti-Hallucination Engine
See LICENSE for full terms and acknowledgments.
Copyright (c) 2024-2025 Prayaga Vaibhav. All rights reserved.
🛡️ VA21 OS - Secure by Design, Intelligent by Nature 🐧
Log in or sign up for Devpost to join the conversation.