How does an AI language tutor work offline on your phone?
Three small AI models run on your phone, one after another, every time you talk to the tutor. One turns your voice into text, one writes a reply, and one reads the reply out loud. Once they are downloaded, none of them needs the internet. That is how a spoken conversation can work in airplane mode.
What happens when you say one sentence?
Say you tell the tutor, in Spanish, that you would like a coffee. Four things happen on the phone:
- Listening. A speech recognition model called Whisper turns the sound into text.
- Thinking. A language model reads that text, along with the lesson or scenario you are in and the conversation so far. In Sayelle this is Qwen3.5 4B.
- Replying. The same model writes the tutor's answer, and any correction, one word at a time.
- Speaking. A text-to-speech model called Kokoro reads the answer aloud.
With a typical cloud app, steps one to three happen on a company's servers. Your audio goes up and the reply comes back. Here the whole loop stays on the phone, which is why it keeps working with no signal.
Why is the download about 4 GB?
The language model is the biggest piece. Qwen3.5 4B has about four billion numbers (weights) inside it. Stored at 4 bits each, which is a compression technique called quantization, those weights come to roughly 2 GB on their own. The speech recognition and voice models, plus the supporting files, make up the rest. Lessons and scenarios you keep offline are extra. You download it once, ideally over Wi-Fi.
Why does my phone's memory matter?
A model has to sit in memory while it works, not just in storage. That is why Sayelle recommends 8 GB of memory or more on Android, and iOS 18 or later on iPhone. On an older or lower-memory phone the tutor can be slower, and the app may close during voice practice. If your phone is on the edge, the requirements are listed in full.
Is an offline tutor as good as a cloud one?
No, and it is better to say so. The largest cloud models are far bigger than anything that fits on a phone, so they know more and make fewer mistakes on hard questions. A phone-sized model is a good fit for a narrower job: holding a simple conversation, staying inside a lesson, and correcting short sentences. That is what beginners and early intermediate learners mostly need. It is not the tool for translating a legal contract or settling a fine point of grammar. It can also be wrong, which is why About Sayelle AI spells out its limits. If something it tells you looks odd, check it.
What are the benefits of running it on the phone?
You can practise anywhere: on a flight, on a train through a tunnel, abroad without roaming. Your raw microphone audio is processed on the phone and is not uploaded. And the main feature does not depend on a server being up, or on a per-message cost, so conversations with the tutor can be free and unlimited.
What still needs internet?
Downloading the models and any new lessons or scenarios, signing in, purchases, ads, and Cloud Sync. If you are signed in with Cloud Sync on, your text transcripts and progress sync next time you are connected. The conversation itself never waits on any of that.
Who is it for?
Learners who want to speak out loud and get an answer back, especially when they travel, commute, or do not want to talk to a stranger yet. Sayelle teaches English, Spanish, French, Italian, Portuguese, German, Chinese, Hindi and Japanese, and each one also works as the language the lessons are explained in. See all nine languages.
Try it on your phone
Sayelle is free to start, on iPhone and Android. Conversations with the tutor are free and unlimited.
