Licences

Med-Mind runs on an open-weights model. This page is where its licence and notice live, as that licence requires.

Two ways Med-Mind runs, and neither sends your writing anywhere

On your phone	What it costs you
Built into iOS — Apple Foundation Models, on iPhone 15 Pro and newer with Apple Intelligence switched on	Nothing. The model is already on the phone, so there is no download at all.
Med-Mind Companion M1 — our own pack, for every other phone	A one-time 932 MB download, and then it works offline permanently.

The app checks which applies and tells you before anything downloads.

Med-Mind Companion M1

The pack the app downloads is a redistribution of Qwen3-1.7B, published by the Qwen team at Alibaba Cloud, in the 4-bit dynamically quantised LiteRT-LM build produced by the litert-community project. The weights are unmodified; only the filename differs.

Field	Value
Upstream model	Qwen/Qwen3-1.7B
Build	litert-community/Qwen3-1.7B (`Qwen3-1.7B_dynamic_wi4b32_afp32.litertlm`)
Copyright	Copyright 2025 Alibaba Cloud
Licence	Apache License, Version 2.0 — full text, notice
Size	977,184,032 bytes
SHA-256	`2eeffef7b51bc3e1225ea69fe7aa5f417397934b56a5b6c20cc068d6fd2c918b`

The Apache License, Version 2.0 permits redistribution under a different name provided the licence and this notice accompany the distribution. That is why the model is called Med-Mind Companion S1 and why this page exists.

Verifying your download

The app checks this hash itself before it will load the model, and deletes the file rather than using it if the hash does not match. You can check the same thing:

curl -sL https://medmind.cstsolution.com/models/medmind-companion-m1/medmind-companion-m1.litertlm \
  | shasum -a 256

The built-in option

Where the phone supports it, Med-Mind uses the model built into iOS through Apple's Foundation Models framework and downloads nothing at all. That model is Apple's, is covered by your agreement with Apple, and like everything else here it runs on the device.

Models we deliberately did not ship

Both of these are good models. Neither was chosen, for licensing reasons rather than quality ones, and it is worth being open about why.

Model	Why not
Gemma 3 1B	The Gemma Terms of Use require every distribution to carry a “Built with Gemma” notice and to pass the full agreement to each recipient. Usable, but it puts Google's name on the core of the product.
Llama 3.2 1B	The Llama Community License requires any derivative model's name to begin with “Llama”, so it could not be called Med-Mind Companion at all.