Support
Answers to the common questions — and a human if you need one.
What do I need to run Bastion?
An iPhone on iOS 17 or later. The zero-download “Instant” model needs an iPhone with Apple Intelligence (iPhone 15 Pro or newer). Downloadable models run on a wider range of recent iPhones — Bastion automatically offers only the models your particular iPhone has the memory to run.
Why is the first model download so large?
The AI itself lives on your phone — that's the entire point. Open models range from about 0.5 GB (Qwen3 0.6B) to several GB for the largest ones. Downloads run over Wi-Fi by default and only start when you ask.
How do I free up the space later?
Open the model picker (tap the model name in the message bar), then tap the trash button on any downloaded model. You'll see exactly how much space it frees. Your conversations are untouched — only the model download is removed, and you can re-download it any time.
Does it really work offline?
Yes. Once a model is on your phone, chat works in airplane mode. The chat itself makes no network requests at all — try it. That makes Bastion a genuinely useful travel companion: long flights, field work, remote areas, and networks you'd rather not touch. Download a model at home over Wi-Fi; after that, no connection, no roaming, no problem.
What's your threat model? What can Bastion actually protect?
What we protect against by architecture: your prompts and conversations being read, logged, sold, breached, or compelled from a server — there is no server. No account links you to what you ask. Network surveillance of your chats sees nothing, because the chat produces no traffic.
What we can't protect against: someone holding your unlocked iPhone. Chats are stored in the app's private storage, protected by your device passcode and iOS hardware encryption — the same protection as your photos and messages — and they exist nowhere else. Use a strong passcode. If you delete a chat, it's gone; if you delete the app, everything is gone. (An in-app lock is on the roadmap.)
One honest caveat: downloading a model is the one visible network act — an observer could tell that a device fetched public model files from Hugging Face, the way they could tell you downloaded any app. What you do with it afterward produces nothing to observe.
Can it see my photos?
Bastion reads the text in images you attach (using on-device OCR) — great for screenshots, receipts, signs, and scanned PDFs. It doesn't yet describe what a picture looks like. When that changes, it will still happen entirely on-device.
Why do answers differ from ChatGPT or Claude?
On-device models are much smaller than data-center ones — that's the trade for total privacy. They're genuinely capable for everyday writing, summarizing, and thinking out loud, but they will sometimes be wrong. Long-press a reply to regenerate it, or switch to a larger model in the picker.
Where are my conversations stored?
On your iPhone, in the app's private storage — never on a server. Deleting a chat (swipe it in History) deletes it. Deleting the app deletes everything.
Something's broken
Email me at brian.kemler@gmail.com with your iPhone model, iOS version, and which AI model was active. Because Bastion has no analytics or crash reporting, your report is genuinely the only way I learn about problems — thank you.