SLM360 LensNano

869KB of AI That Replaces Your Cloud

A complete on-device voice + vision AI engine designed for smart glasses. Handles NLU, voice command classification, entity extraction, and OCR. 1.67M parameters in 869KB, sub-10ms inference on CPU, 100% offline. Built on the 360 Engine in pure Rust with zero garbage collection.

Specifications

Architecture

 1  Stage 1 (Always On, 17MB, ~5mW): Microphone > 360 VAD (2MB) > 360 Wake (15MB)
 2  Stage 2 (On Demand, 43MB): 360 STT (42MB) > SLM360 NLU (869KB) > Action
 3  NLU: 192-dim, 4 layers, 6 GQA heads, SwiGLU + RoPE + RMSNorm
 4  Vision: 360 OCR for reading signs, menus, prescriptions, labels
 5  Framework: 360 Engine (Rust, zero GC), GPU training via 360 Tensor (Metal/CUDA)

Features

• 15 voice intents: navigate, call, message, capture, music, fitness, weather, alarms, shopping, translate, notifications, focus mode, identify, read text, assistant
• Entity extraction for contacts, locations, times, durations, languages, frame types
• 360 OCR Vision: on-device neural OCR for real-world text recognition
• Hindi + English with native Hinglish code-mixing support
• 99 languages supported via STT engine
• 100% on-device. Zero data leaves the glasses. Works offline and in airplane mode
• OEM-ready with configurable wake word per brand
• Rust SDK, ARM-native, ONNX export for ARM, x86, WASM, RISC-V

Deployment Targets

• >ARM-native for smart glasses hardware
• >ONNX export for ARM, x86, WASM, RISC-V
• >OEM SDK integration with configurable wake words