For developers
Build on FAMTEC
FAMTEC Local AI shows how a private, on-device AI can run on a phone with no account and no cloud. The part that makes that work, the FAMTEC Runtime, is available to developers who want the same thing in their own apps.
What you get
- A model manager that picks a model to suit the device, downloads it, and checks it against a pinned checksum.
- One shared model file for several apps on a device, so users download it once.
- Native modules for fast integrity checks and for reading text from photos and PDFs on the device.
- Safety and search logic: an on-device crisis and distress check that runs before the model, and optional, privacy-conscious web search.
How the licence works (summary)
- Free for personal, educational and research use, and for any app that earns up to US$1 million in a year. It is measured per app, over the last 12 months, and your other apps and businesses do not count.
- Commercial licence only above that. No fee is payable until FAMTEC has published standard pricing and given you 90 days' notice, and terms for versions you have already released stay stable.
- You can build and ship your own commercial apps on it, including AI assistants, modify it, and publish your own app's source with the licence notices kept. You keep ownership of your own code.
- Show “Powered by FAMTEC” with a link on your credits screen. You may not resell the Runtime itself as an SDK, or use it to train models.
- The AI models themselves are third-party and Apache-2.0. We do not charge for them. We license our code.
This is source-available, not open source. This summary is not the licence; the full draft is in the repository licence file, and a lawyer-reviewed version will be published here before general release.
Get in touch
Email rob@fineartmedia.tech with what you are building and roughly when it will ship.
Third-party notices
- Models: Qwen3 0.6B, 1.7B, 8B, 14B (Qwen team, Alibaba Cloud) and Qwen3-4B-Instruct-2507 GGUF (Unsloth), all Apache-2.0, downloaded from Hugging Face.
- Engine: llama.cpp and llama.rn, MIT.
- Framework: React Native, Expo and their libraries, mostly MIT, with BSD, ISC and Apache-2.0 components. No GPL, AGPL or LGPL code is included.
- Search sources, only when you switch web search on: DuckDuckGo, Wikipedia, and optionally Brave Search with your own key.
Product names belong to their owners and are used only to identify them.