OGAM

Off Grid AI is an app that runs AI on your phone or Mac, offline, with the README stating that no data leaves the device. It combines text generation (Qwen 3, Llama 3.2, Gemma 3, Phi-4 and any GGUF model you bring), on-device Stable Diffusion image generation, vision models (SmolVLM, Qwen3-VL, Gemma 3n), on-device Whisper voice input, document analysis, and tool calling for web search, a calculator, date and time, device info and knowledge-base search. Projects can hold PDFs and text that are chunked and embedded on-device with a bundled MiniLM model and stored in SQLite.

It also supports GPU and NPU acceleration (OpenCL on Adreno, Metal on Apple Silicon, with the Hexagon NPU on Snapdragon), which the README itself marks experimental: only Q4_0 and Q8_0 quants are accelerated, other quants fall back to CPU, and some architectures produce garbled output. It can connect to OpenAI-compatible servers on your network such as Ollama and LM Studio, shows per-model memory use, and has a download manager. The README gives performance ranges, for example 15-30 tok/s on CPU on flagship phones, tested on Snapdragon 8 Gen 2/3 and Apple A17 Pro.

The app is on Google Play and the App Store, and you can build from source with `git clone` and `npm install`. The README also advertises an optional Pro tier (voice replies, personas, draft-then-approve actions, phone-Mac sync) priced at $69 once or $49 a year, so the free app and the paid features are separate. The code is MIT licensed and was last pushed on 2026-10-12. This review is based on the README and repository data; we have not run it ourselves.