LocalMathy

LocalMathy solves math problems entirely on your device. Nothing you type or photograph ever leaves your phone. With no account, no cloud, and no networking once the model is set up.

It runs VibeThinker-3B, a compact reasoning model, through Google's LiteRT on-device runtime. Ask a question and the app streams the model's step-by-step reasoning into a collapsible pane, then renders a clean final answer with full Markdown and LaTeX math.

Features

* Fully offline and private, all inference runs locally on the GPU (with a CPU fallback)

* Step-by-step reasoning you can expand or hide

* Answers rendered with proper Markdown and LaTeX math typesetting

* Tap a boxed answer to copy it

* Solve from a photo, snap or pick a picture of a problem, crop it, and an optional on-device vision model reads it into an editable question

* Optional local history of your questions and answers

* Light and dark themes

Getting the model

On first launch you download the reasoning model (about 1.9 GB) from Hugging Face, or import a .litertlm file you already have. It is stored on-device and only needs to be set up once. Photo solving uses a separate optional vision model you can add later.