DocsFAQ
FAQ
Short answers to common questions.
Does it work offline?
Yes, once a model is on the phone. Chat, reasoning and the tools of the phone need no connection: what you type is processed on the device and is not sent anywhere to generate a reply. Only downloading a model and the Internet tools need the network.
Where does my data go?
Nowhere by default. Models, chats, memory and settings stay on the phone, and there is no account and no telemetry. A request leaves only to download a model (Hugging Face) or, if you turned them on, for the Internet tools. The details are on the privacy page.
Why is it slow on my phone?
The model runs on your phone's own processor, and a phone is far from a server. On the phone we measured, Qwen3 1.7B replies at about 5.5 tokens/s, and the engine benchmark goes from 3.9 tokens/s for a 3B model to 0.2 for a 7B one on CPU. tensai marks models of 3.5 billion parameters or more as Tight fit, a rule of thumb. Try a smaller model, check the backend, lower the context. See Performance.
Is there an iPhone version?
Not yet. tensai is an Android app today. An iOS version may come one day, but there is no plan or date for it.
Why not the Play Store?
Not for now. tensai comes as a signed APK from this site and GitHub releases, with F-Droid later. The official stores may follow one day, with no date.