An AI that never goes online.

Qwen3.5‑0.8B runs entirely on your iPhone or iPad. The model ships inside the app.

What it is

A 0.8‑billion‑parameter language model is installed along with the app. There is nothing to download afterwards, no account to create, and no network request of any kind. Turn on Airplane Mode and everything still works, because it never needed the connection in the first place.

You write it a letter. A slip of aged paper lies on a dark desk; you write on it once, the reply is written onto that same slip, and then the letter is sealed — you cannot add to it. When there is something else to say, the corner of a fresh sheet is waiting at the right edge. Pull it across and begin another.

Writing once is the point, not a missing feature. Somewhere you can keep replying gets used like a chat window: half a thought sent, then corrected, then corrected again. One letter holds one exchange, so the thought has to be finished before it goes.

Where your writing goes

Nowhere. The letters are kept in a file inside the app's own container on the device. They are not synced, not uploaded, and not readable by other apps. Deleting Answer Letter deletes them with it.

This is not a policy decision that could quietly change in an update — the app contains no networking code for it to change into. See the privacy policy for the full accounting.

Requirements

• iOS or iPadOS 18.6 or later
• About 600 MB of free space — most of it is the model
• A few seconds on first launch while the weights are mapped into memory

Built on

Qwen3.5‑0.8B under Apache‑2.0, with inference by llama.cpp (MIT) on the Metal backend.

Answer Letter 0.1.0