Your AI – your best friend in your phone

Your AI, your name

The AI in your SCYLAR One belongs to you – so it gets the name you choose. Out of the box she is called Aurora – you give her your own name. Anything from 2 to 24 characters. From then on, that is what you call her, and that is how she answers.

And you decide how she works for you:

• What she may do: take care of tasks on your phone, answer your questions – or both.
• When she speaks: only when you open her, also for new notifications – or never out loud.
• What she does without asking: out of the box, timers and alarms. Everything else you confirm beforehand, until you change that.
• Where she thinks: only on the phone – that is the default – or, when needed, with a larger AI via our servers, and even then only with your consent.

Her character comes built in: a companion who is honest, answers in your language and never pretends to be a person. Like a best friend in your phone who gets done what you say and knows your everyday life – your appointments, your contacts, your messages.

What can your AI do?

Your AI is built right into ScylarOS. You talk to her or type – whatever suits the moment. She can make calls, write messages, add appointments, find files and take care of tasks in the ScylarOS apps. She has tools in the system for that, so she does not just answer, she acts.

For example: “Call Mum.” – “Text Jonas that I’ll be ten minutes late.” – “Put the dentist in for Friday at 3 pm.” – “Find the invoice from October.”

Your AI is not an app you download and not an add-on; she is part of the system from day one. What she is not: a replacement for a large cloud AI in every situation. The model on the device is small. Where that reaches its limits is explained further down – along with when she gets help via our servers, with your consent.

Which AI runs directly on the phone?

Two AI models run directly on SCYLAR One:

• Language model for text and tools: Qwen2.5-1.5B-Instruct, Apache 2.0 licence, in a reduced (quantised) version of about 0.9 GB.
• Speech recognition: Whisper small, MIT licence.

These two models work on the phone without an internet connection.

That means: when you say something to your AI, speech recognition on the phone turns your voice into text. The language model on the phone understands the task and calls the right tool – phone, messages or calendar, for example. For these steps, your request does not need to leave the device.

Both are open models with their own licences that we have built into ScylarOS. We deliberately do not publish speed figures until we have properly measured values.

When does your AI ask our servers – and what leaves the device then?

For more difficult tasks, your AI can ask a larger AI via our servers. This only happens if you agree. With the setting “Ask when needed”, she asks you before every use.

If you say yes, your request leaves the device and goes via our servers to the larger AI. If you say no, she sticks to what the model on the phone can do.

A second place where data can leave the device is the live translator: it also uses our servers unless it is set to “local only”. If you don’t want that, set it to “local only”.

We say this so clearly because for us “AI on the device” should not mean that nothing ever leaves the phone. It means: speech recognition and the language model run on the device. For the larger AI, your AI asks you beforehand, and you set up the translator yourself.

What does your AI know about me – and where is it stored?

To help you, your AI needs context: who you mean by “Mum”, when your next appointment is, what you last talked about.

What she knows about you – your contacts, your appointments and the history – is stored encrypted on your device, in the encrypted data vault of ScylarOS. Its key depends on your PIN and a device secret.

There is no ad tracking and no data for ad networks. What your AI knows is there to take work off your hands.

If you allow her to ask the larger AI via our servers for a difficult task, that request leaves the device – only with your consent. More on encryption is on the page about ScylarOS.

Am I talking to a person or to an AI?

To an AI. Even if you give her a name: your AI is not a person and never claims to be one. Her answers are generated by an AI model – on the device or, if you agree, by a larger AI via our servers. We state this explicitly because you should know it before you talk to her. The European AI Act (Article 50) requires it too. That is why the notice “You are talking to an AI” is right at the top of this page.

What this means in practice: AI answers can be wrong. An AI model can mix things up or sound confident while being mistaken. So check important information yourself – especially about health, money and legal matters. And when your AI has done something for you, such as adding an appointment, take a quick look if in doubt.

Where does your AI reach its limits?

• Small model on the device: for complex questions, your AI on the phone is less capable than large cloud AIs. The larger AI via our servers helps – but only if you agree.
• Languages: which languages your AI understands and how well, we will publish once we have tested it.
• Speed: we do not publish figures until we have properly measured values.
• Own apps: your AI takes care of tasks in the ScylarOS apps. We make no promise for apps from the optional app extension.
• Mistakes: AI answers can be wrong, even when they sound confident. Check important things yourself.
• Beta: ScylarOS is still in beta. What the system as a whole cannot do yet is listed in the beta notes (German).

Questions about your AI? We are happy to help: support@scylaros.com.