wtf is offline-ai-assistant?

Offline AI Assistant is a step by step guide, written mostly as documentation rather than a finished app, for turning an Android phone into a private AI assistant that runs entirely on the device with no internet connection needed after setup. It walks a reader through installing Termux, a terminal app for Android, then compiling llama.cpp, a project that runs AI language models locally, and finally downloading two specific models: Qwen2.5-Coder for coding help and SmolVLM for understanding images. The documentation is organized as fifteen ordered steps inside a docs folder, starting from an introduction and supported device list, moving through installing dependencies and compiling llama.cpp, downloading and testing each model, and building small helper scripts called ask.py and config.py that tie everything together into a usable assistant, ending with a troubleshooting section. Two custom commands, coder and vision, are created along the way to make the two models easy to invoke from the terminal. Because the models run locally, the project lists real hardware requirements. The recommended setup is Android 12 or newer with a recent high end Snapdragon chip, 12 gigabytes of RAM or more, and at least 15 gigabytes of free storage, while the stated minimum is Android 11 with a Snapdragon 8 Plus Gen 1 chip and 8 gigabytes of RAM. An internet connection is only needed during the initial installation and model downloads, after that, all inference happens on the device and no data is sent anywhere. The project is aimed at Android users comfortable following a technical guide who want a coding assistant and basic image understanding without depending on cloud AI services, rather than at people looking for a polished one tap app. It is released under the MIT License and is described as an actively maintained, open source project welcoming contributions and bug reports.