A lightweight, open-source AI assistant that runs locally on phones, Raspberry Pi, and low-RAM devices — no cloud, no subscriptions, no beefy GPU required.
Models compressed to Q4_K_M format via llama.cpp — 75% size reduction with minimal quality loss.
No GPU required. Optimized SIMD/NEON instructions squeeze maximum speed out of any ARM or x86 chip.
JNI bridge to llama.cpp. No Termux. No hacks. A real APK that works out of the box.
No analytics, no crash reports sent out. Your data, conversations, and models stay on your device.
OpenAI-compatible local API. Plug NanoMind into any app or script already using the OpenAI SDK.
Not locked to one model. Use Mistral, Phi-3, Gemma, or any community-quantized model from HuggingFace.
Ship a working Android app and Linux CLI binary. Basic chat, model download, offline inference.
One-tap model download optimized per device class. No manual GGUF hunting on HuggingFace.
Web search, RAG on local files, voice input. Community-built extensions via an open plugin API.
Expand to iOS TestFlight and Windows ARM. Every major platform covered.