AI summary
On-device AI server that loads GGUF or LiteRT models and exposes an OpenAI-compatible API over a local network. The latest update adds support for newer architectures, including LFM3, and image, audio, video, and text generation and analysis; the app description also notes these inputs are unsupported.
Generated by AI. May contain inaccuracies.
About this app
SaMu Lab turns your Android phone into a personal AI server that runs entirely on your device with no cloud no subscription and no data leaving your hands. You load a GGUF or LiteRT model from your phone storage and the app boots it into memory then exposes an OpenAI compatible API on port 8080 over your local network. Any OpenAI compatible client like PocketPal PrivateLM Cherry Studio or plain curl can connect with the URL and API key the app shows and start chatting with your model. Everything runs offline so it works on a flight in a village or anywhere without signal. Every conversation stays on your phone and never touches a third party server. There are no token bills no monthly fees and no usage limits. You can name each model create as many API keys as you want set your own system prompt change the port toggle LAN access and tune the context window from inside the app. The engine is llama.cpp compiled natively through JNI so there is no subprocess overhead and it auto detects big and little CPU cores to tune thread counts for MediaTek Snapdragon and Exynos chips. Flash attention and quantized KV cache are enabled by default so first token latency stays low and streaming tokens arrive live. Supported formats include GGUF for Gemma Llama Qwen Phi DeepSeek Mistral and hundreds more and LiteRT for Google models with more formats planned. The server can host multiple client devices at once each with its own API key so three or four phones can share one local model. Endpoints include v1 models v1 chat completions with streaming and non streaming and v1 completions so any standard client works without modification. SaMu Lab needs Android 8 or newer on an ARM64 device with at least 4 GB RAM for small models and 8 GB or more for 7B models. It does not do cloud inference does not include a chat interface does not support image or audio input and does not use GPU acceleration on Mali chips. It is built for students developers privacy focused users travelers and anyone who wants a free private always available AI that lives on their own phone. Cloud AI stays in the cloud your AI stays with you.
What's new
Now Supports latest architecture models like LFM3 and others Supports image , audio, video, text generation and also analysis
About this version
- Version
- 0.1.0 (1)
- Size
- 36.33 MB
- Requires Android
- 8.0
- Min SDK
- 26
- Architecture
- arm64-v8a
- Downloads
- 52
- Updated
- Nov 15, 2026
- Package
- com.neurasamu.build
Similar apps
Ratings & reviews
- 50
- 40
- 30
- 20
- 10