§ support

Support

Unbound runs entirely on your own hardware, so most issues come down to the model download, storage, or device performance. Start here, then email us if it is still stuck.

Email [email protected]

Include your device model, iOS version, app version, and the model you selected

The chat says “Download model”. What now?

Unbound ships without a model so the download stays your choice. Go to Settings → Model, pick one, and tap Download. Use Wi-Fi: files run from about 0.5 GB (Unbound Q 0.8B) up to several GB for the vision models. Return to chat when it finishes — the first message loads the model into memory, which takes a few seconds, then tokens stream.

The download stopped or failed.

Downloads are resumable: reopen Settings → Model and start it again, and it continues where it left off. If it keeps failing, check free storage (you need roughly the model size again while it assembles), stay on Wi-Fi, and keep the app in the foreground.

Generation is slow.

Speed depends on device and model. Pick a smaller model or a lower quantisation (Q4_K_M is a good default, Q2_K is fastest and least accurate), keep other heavy apps closed, and expect the first message after loading to be slower because the prompt has to be processed before the first token. The tokens-per-second readout under a reply shows what your device is actually doing.

Does it work offline?

Yes. Once a model is downloaded, chatting works in airplane mode. The only optional network feature is web search, which needs your own Tavily API key.

How do I free up space?

Settings → Model → delete the model you no longer want. Chats and settings are tiny by comparison; model files are what use the storage.

How do I turn on web search?

Settings → Web search, paste your own Tavily API key, and enable it. The key is stored on your device. With no key, the app makes no search requests.

Why is it asking for calendar, contacts, reminders, or health access?

Only when you ask the assistant to do something that needs it — “what’s on my calendar tomorrow”, “remind me at 6”, “how did I sleep this week”. Decline and the rest of the app works normally. Everything read this way stays on the device.

Can I use my own model?

Yes. Settings → Model → add a model from a public Hugging Face GGUF URL. Use the resolve/main/<file>.gguf form, not blob/main. Quality and prompt formatting are up to whatever model you point it at.

The model said something wrong or offensive.

Unbound runs open-weight models locally without server-side moderation, so output can be inaccurate or explicit and is not reviewed by us. It is intended for adults. Verify anything that matters, and do not use it for medical, legal, or financial decisions.

How do I delete my data?

Settings → Privacy → erase all chats, and Settings → Model → delete the model. There is nothing stored on our side to delete, because nothing is uploaded. See the privacy policy.