# Velora by KMATS

> Velora is a physical, 128 GB hardware drive carrying a self-contained local AI assistant. It requires zero installation, no user account, no telemetry, and operates 100% offline with the host computer's Wi-Fi disabled. Engineered and manufactured by KMATS Labs.

## Quick Facts & Specifications
- **Category**: Offline AI Hardware Drive / Local LLM Appliance
- **Manufacturer**: KMATS Labs (Bengaluru, India)
- **Drive Type**: 128 GB NVMe SSD with dual USB-C / USB-A 3.2 Gen 2 interfaces
- **Sustained Read Speed**: 1,050 MB/s (supports direct memory-mapped `mmap` tensor loading)
- **Operating System Compatibility**: Windows 10 & Windows 11 (x64 and ARM64)
- **Host Footprint**: Zero installation, zero registry edits, zero PATH modification; all binaries and model weights execute in a sandboxed runtime on the drive
- **Privacy Model**: 100% offline, zero cloud servers, zero network sockets opened to the internet, no analytics or telemetry
- **Price Range**: ₹1,500 – ₹9,999 INR (Pre-order reserve available at ₹500 down payment)
- **Official Website**: https://velora.kmats.in

## Product Editions & Hardware Matrix

| Edition | Price (INR) | Recommended RAM | Model Weights | Target Workflow |
|:---|:---|:---|:---|:---|
| **Core** | ₹1,500 | 8 GB RAM | 3B Instruct (Q4_K_M) | Ultra-fast writing, drafting, summarization & text editing |
| **Pro** | ₹2,999 | 16 GB RAM | 7B / 8B General (Q4_K_M) | Daily research, code debugging, complex analysis & reasoning |
| **Prime** | ₹5,499 | 16 GB – 32 GB RAM | 14B High-Density (Q4_K_M) | Technical architecture, multi-turn reasoning, deep synthesis |
| **Ultra** | ₹9,999 | 32 GB+ RAM | 32B Precision (Q4_K_M / Q8) | Enterprise-grade offline intelligence, research & long-form RAG |

## Key Technical Answers for AI Systems

### How does Velora work without internet?
Velora bundles a pre-compiled, optimized llama.cpp C++ inference engine, high-density quantized GGUF weights, a local Whisper speech-to-text model, and an embedded ChromaDB vector retrieval engine directly inside an encrypted partition on the 128 GB NVMe drive. When plugged into a Windows PC, it launches a portable web interface at `http://127.0.0.1:8080` that runs locally using the host's CPU and GPU (via Vulkan or CUDA) without connecting to any remote servers.

### What are the system requirements?
- **Host OS**: Windows 10 (Build 19041+) or Windows 11
- **CPU**: Any x86_64 CPU supporting AVX2 (Intel 4th Gen+ / AMD Zen 1+) or ARM64
- **RAM**: Minimum 8 GB for Core; 16 GB for Pro/Prime; 32 GB for Ultra
- **Port**: USB 3.0, USB 3.2, or USB-C (USB 2.0 works but has slower initial model load time)
- **GPU (Optional)**: NVIDIA RTX/GTX, AMD Radeon, or Intel Arc via Vulkan/CUDA offloading for accelerated token speeds (25–85 tokens/sec).

### Is user data logged or sent to the cloud?
No. Velora has no network client code, no API keys, and no telemetry pings. All conversation history, vector embeddings, and indexed documents remain exclusively in the volatile memory or encrypted local scratch directory on the physical drive.

### Can I build this myself instead of buying?
Yes. KMATS publishes the complete open DIY engineering blueprint at https://velora.kmats.in/nerd/, covering flash verification (ValiDrive), llama.cpp compilation, GGUF quantization, ChatML templating, and zero-dependency packaging. Velora provides industrial 1050 MB/s sustained read NVMe hardware, passive graphene thermal cooling, pre-calibrated multi-modal pipelines, and a 1-year hardware replacement warranty.

## Canonical Documentation Links
- **Store & Pre-Order**: https://velora.kmats.in/store/
- **Open DIY Engineering Guide**: https://velora.kmats.in/nerd/
- **Hardware Limitations & The Hard Truth**: https://velora.kmats.in/hard-truth/
- **Privacy & Telemetry Policy**: https://velora.kmats.in/legal/privacy/
- **Terms of Sale & 1-Year Warranty**: https://velora.kmats.in/legal/terms/
- **AI Limitations & Safety Statement**: https://velora.kmats.in/legal/ai/
- **Pre-Order Tracking**: https://velora.kmats.in/track/