Documentation

How Local Mind chooses models and engines, and what each catalogue entry has actually been shown to do.

Pages

• Model catalogue — what a catalogue entry records, how downloads are verified, and what the unevaluated badge means.
• Inference engines — how an engine is selected, and why a tuned build is offered under its own name.
• Nemotron 3.5 Lightning — a 30B mixture-of-experts model with mamba blocks. In development
• DeepSeek V4.1-Flash weights — reading the community quantisations of a 748B model. In development

How to read availability on these pages

Every page says plainly whether something is in a released build, in a TestFlight build, or only on a development branch. These are three different things and the pages do not blur them. A feature described as in development is not available to download, and no date is promised for it.

Likewise, no page quotes a speed figure without naming the model and quantisation, the machine, the storage the weights were read from, the engine and the context length, and saying how it was measured. Where a figure in the app is modelled rather than measured, the page says so.