Assistant Offline

Prototype — scripted replies

20/08/2026

Explain what a GGUF file actually is
A single-file package for a language model: the weights, the tokenizer, and the metadata your app needs to load it. Read straight from disk.
Where does the model run?
On this device. The weights are read from local storage, and nothing is sent anywhere.
What about with no connection?
Identical. Airplane mode changes nothing, because no request is made when a reply is produced.
What gets stored?
The conversation stays on the phone. There is no server to sync it to.

The app ships with no weights. Models are downloaded once and then read from local storage.

Qwen 3 · 4B

Downloaded · ready

Active

Runs through llama.cpp, on device

Qwen 3 · 1.7B

Smaller and faster, for older phones

No account. Nothing is uploaded.

Each mode changes how the assistant answers. The kid mode is locked by a parent PIN.

Assistant

General questions, plain answers.

On

Kid‑safe

Simpler answers, narrower topics. Ages 8–10.

Network

Requests at reply timeNone
AccountNone
AnalyticsNone
Cloud syncNone

Storage

ConversationsOn this device
Model weightsOn this device

Turn on airplane mode and nothing changes.