Untethered LLM · For iPhone and iPad

An AI chat app with nowhere to send your words.

Untethered LLM runs the whole language model on your device. Your prompts never leave your iPhone, because there is no server side to send them to.

It ships with a capable built-in model, opens common files for on-device analysis, and keeps every chat, model, and attachment inside the app until you decide otherwise. Forty interface languages. One purchase. No account.

iPhone & iPad · iOS 18+ · One-time purchase · No account

Download on the App Store Get support → Read the privacy policy →

Qwen3 4B built in · 40 interface languages · GGUF import · Passphrase-locked exports

Screenshot coming soon

In brief

The short version

Where it runs: On your iPhone or iPad. The model ships with the app and runs with llama.cpp — no connection needed after install.
What it remembers: Only what you keep. Chats and files stay in the app sandbox until you export or delete them.
What it costs: One purchase, once. No subscription, no account, no usage meter.

Overview

What the app actually does

Five things, plainly stated. Everything below happens on your device.

Chat with a model that lives on your phone

Qwen3 4B ships with the app — a 4-billion-parameter model that handles everyday questions, writing, and reasoning. A smaller fallback model covers older devices.

Bring your own models

If you have a favorite GGUF file, import it and chat with that instead. Anything up to about 6B parameters runs well on a recent iPhone; the Models page lists tested picks.

Ask about your files

Attach documents, spreadsheets, photos, audio, or video and ask questions about them. Extraction and analysis happen on-device; the originals stay yours.

Reshape replies, or write your own lens

Built-in transformers — executive summary, technical deep-dive, and more — restyle any answer. Custom transformers in Settings let you define your own, all inside the local prompt path.

Keep, export, or delete — your call

Chats save locally. Export as text or Markdown, optionally locked with a passphrase. Delete one chat or wipe everything; deletion removes the local files.

Honest limits

What it won't do

Honest limits, so there are no surprises:

It can't browse the web or check today's news — the model only knows what it was trained on, plus what you give it.
Big models need a big phone: the built-in 4B model wants an iPhone with 8 GB of RAM. Older phones get the smaller fallback (SmolLM2 135M).
It can be confidently wrong, like every language model. Double-check anything that matters.
Files stay on your device, which means no sync between devices and no web backup. Export the chats you want to keep.

Legal

The fine print, up front

Three short documents cover the app, your privacy, and the open-source components inside. They are written to be read, not to be scrolled past.

Support

Talk to a human

Support, press, privacy, or licensing — one inbox, read by the people who make the app.