# LLMs.txt
# Standardized Machine-Readable Site Manifest
# Specification: https://llmstxt.org

> Lapis is a native on-device AI application for Apple Silicon (iOS, iPadOS).
> It runs large language models, reasoning, and vision-language models locally using Apple MLX and Metal GPU acceleration.
> 100% offline, air-gapped, zero telemetry, completely free.

- [Main Website](https://lapis.lvrpiz.com/): Official homepage with interactive MLX telemetry and feature overview.
- [Full Documentation](https://lapis.lvrpiz.com/llms-full.txt): Comprehensive technical specifications, architecture, and FAQ for LLM reasoning engines.
- [App Store](https://apps.apple.com/us/app/ai-offline-llm-chat-lapis/id6754216427): Official download link on the Apple App Store.
- [FAQ](https://lapis.lvrpiz.com/faq): Technical and operational answers regarding memory management and supported devices.
- [Privacy Policy](https://lapis.lvrpiz.com/privacy-policy): Complete commitment to on-device sandboxing and zero data collection.
- [Press Kit](https://lapis.lvrpiz.com/press-kit): Official brand assets and developer contact.

## Core Capabilities
- **Engine**: Apple MLX + Metal Performance Shaders (MPS).
- **Vision / OCR**: On-device document parsing with SmolVLM and Qwen2-VL.
- **Voice Mode**: Real-time voice interaction with Voice Activity Detection (VAD) and WebGPU liquid shader rendering.
- **Model Hub**: Hugging Face community MLX weights support (Qwen2.5, DeepSeek-R1-Distill, Llama-3.2).