A model won't load
If a model fails to load, the chat screen says why instead of showing a raw error, with shortcuts to Settings and to the setup wizard.
- The file is truncated or corrupted. Usually a download that was interrupted. Delete the model in Settings → Models and download it again while online.
- Not enough memory ("Hardware RAM Exceeded"). Close other apps, or switch to a smaller model in Settings. The default model needs the least.
- The architecture isn't supported. Some models from Hugging Face use designs this build of llama.cpp can't run yet (Instella-MoE-16B is one). Pick another model.
The setup download is stuck
Each download shows a Retry button when it has made no progress for 60 seconds. Check your connection and tap Retry. Keep BOAR open: downloads pause while the app is in the background and continue when you come back.
Answers are slow
- The first word takes the longest. The model reads the retrieved passages before it writes; that's most of the wait.
- Bigger models are slower. On our test phone the default model wrote about 11–17 tokens a second; Phi-3.5-mini about 4 and Qwen2.5-7B under 3. See Models.
- Heat slows the phone down. Over a long session speed dropped by about a third as the phone warmed up. Let it cool, and take it out of a case or the sun.
- Deep Research is slower on purpose. It makes several passes. Turn it off for quick questions.
An answer stopped with "Paused — app was backgrounded"
Android pauses apps you leave. Ask again to continue, and keep BOAR in front while it answers.
Voice input says it's unavailable
BOAR uses your phone's own speech recognizer. GrapheneOS and other phones without a system speech service don't have one, so voice isn't available there. Typing always works.
An answer is wrong
It happens, especially on long comparisons and on arithmetic with several steps. Check the sources under the answer, ask a narrower question, or try a larger model. If you think BOAR should have got it right, open an issue with the question and the answer; wrong answers are how the evaluation set grows.
Start over
- Change setup or re-download the defaults: Settings → App → Re-run Setup Wizard. Nothing else is lost.
- Wipe everything: Settings → App → Danger Zone → Clear All Data & Reset App. It deletes every model, the whole library including your own collections, and all conversations and settings, then opens the setup wizard.