localBotsWe answer, and so do the bots.
Most questions are covered below. For anything else, write to us and a human replies.
roberto.pillitteri@hotmail.it
Include your iPhone model and iOS version, and the model you downloaded if the question is about a download or a slow reply.
Which iPhone do I need?
Any iPhone that runs iOS 26.4 or later. Models under 1 GB (Qwen3 0.6B and 1.7B) run on every supported phone; 3–4B models need an iPhone with 6 GB of RAM or more (iPhone 13 Pro or later, iPhone 15 or later).
The download seems slow or stuck
Models are 0.4–3 GB. The library shows megabytes downloaded and the current speed; Wi-Fi is recommended. If your connection drops, tap Retry: the download resumes where it stopped. A download that shows no data for a while is flagged in the library.
Why don't jobs run while the app is closed?
iOS does not allow apps to run a language model in the background. Reminders still arrive on time as notifications. "Job" routines notify you at the scheduled time and run as soon as you open the app or tap the notification.
Does anything leave my phone?
Only model downloads from Hugging Face and, if you turn on DeepSearch, your search query to DuckDuckGo or Wikipedia. See the privacy policy.
The first reply takes a while
The model is loaded into memory on first use, which takes a few seconds. Replies after that stream immediately. If the phone is low on memory, iOS may unload the model between chats; it reloads on the next message.
How do I free space?
Open ⋯ → Models and swipe a downloaded model to delete it. Chats and bots are tiny; models are the only large files.
Can I use a model that isn't in the list?
Yes. In Models, tap Add from Hugging Face and paste any MLX repository id such as mlx-community/SmolLM3-3B-4bit. 4-bit models under 3 GB work best on iPhone.