Android · Local-first AI
Run GGUF and LiteRT-LM models 100% offline on your Android phone. No internet, no account, no data leaves your device. Optional cloud API fallback when you choose.
Get Snor
Download the APK and sideload it, or install from your file manager. Allow “install from unknown sources” when prompted.
Download APKWhat it does
Download and run GGUF models with GPU acceleration via Vulkan, or LiteRT-LM .litertlm models on CPU, GPU, or NPU. No internet required after download.
Any LiteRT-LM model can also serve your local network over an OpenAI-compatible /v1/chat/completions endpoint, with an optional ngrok or Cloudflare tunnel when you need reach from outside.
Seamlessly switch to OpenAI, Anthropic, Google Gemini, DeepSeek, OpenRouter, or NVIDIA NIM when you need more power. You choose where inference runs.
Send text and images in conversations. Vision works with local models (Qwen2-VL) and cloud providers alike.
Chats, tasks, and settings live in local storage via Hive. Nothing leaves your device unless you explicitly pick cloud mode.
On first launch, Snor reads your device’s RAM and sets optimal context size and token limits automatically.
A dedicated task view for structured AI-assisted workflows, alongside free-form chat.
How it works
Prompts start from a single input box and stream back into the same conversation.
llama.cpp with Vulkan, or LiteRT-LM, after a one-time download. No internet required.
OpenAI, Anthropic, Gemini, DeepSeek, OpenRouter, or NVIDIA NIM. Used only when you pick a provider.
Answers land in the same chat — local data stays on-device, cloud calls go only to the endpoint you chose.
Privacy
Local inference and the optional LAN server run on-device. Chats are stored locally, models are local, and nothing is sent anywhere without an explicit choice.
Even in cloud mode, API keys are kept in local storage and transmitted only to the provider’s endpoint — never to Snor, never to a third party.
FAQ
Yes, Snor runs GGUF and LiteRT-LM models completely on-device. No internet connection is required for local inference after the initial app and model download.
Snor supports GGUF models with Vulkan GPU acceleration. Tested models include Phi-3, TinyLlama, Gemma 2B, and various Llama 3 variants in Q4_K_M and Q5_K_M quantization levels.
Absolutely. Your chats, settings, and models remain on your device unless you explicitly choose to use cloud API fallback. Even in cloud mode, API keys are stored locally and transmitted only to the chosen provider.
Download the APK from snor.site and open it on any Android 8.0 or newer device, then allow installs from your file manager when prompted. Pick the build that matches your device ABI: ARM64, armeabi-v7a or x86_64.
Not unless you turn on cloud mode yourself. Local inference runs entirely on your device. In cloud mode, requests go directly from your device to the provider you selected, and your API key is stored locally.
Yes. Any LiteRT-LM model loaded in Snor can expose an OpenAI-compatible /v1/chat/completions endpoint on your local network, so other apps and tools on your Wi-Fi can use it. An ngrok or Cloudflare tunnel is optional if you need access from outside your network.
