The short version
- Want to try lots of models: PocketPal (free, open source) or Private LLM (paid once).
- Want a plain private assistant: Locally AI or Enclave.
- Want a bigger, feature-heavy app and have a new phone: Layla.
- Want to talk to characters who stay in role: Castmates.
Comparison table
| App | Built for | RAM / device | Price | Downside |
|---|---|---|---|---|
| Private LLM | General chat, many models | Some 1B models on 4 GB iPhones, 3B and up need 6 GB+ | $4.99 one-off | Reviewers report short memory and repetitive replies from the small models |
| PocketPal | Running any GGUF model you pick | Depends on the model you choose | Free, open source | You choose and manage models yourself |
| Locally AI | Simple private assistant | Needs iOS 18.1+, no RAM floor stated | Free | Some reviewers on older iPhones say it won't open |
| Enclave | Private chat with a model picker | No RAM figure published | Free plus subscription | Reviewers on older devices report slow replies |
| Layla | Large local model plus image and music tools | Newest iPhones, 8 GB+ recommended | $19.99 | About a 4 GB first download, degraded below 8 GB |
| Castmates | Character roleplay | 6 GB+ (iPhone 15 Pro and similar) | 30 free messages a day, Pro from $6.99 | One built-in model, no model choice, not a general assistant |
Rival details come from each app's App Store listing, its developer's own site and reviews, read on 9 October 2026. Versions change, so check the listing before you buy.
Private LLM
A paid app that offers a long list of models, including small Llama and Gemma builds for 4 GB iPhones and larger ones for phones with 6 GB or more. The developer says the 1B Llama runs on all supported iOS devices. It costs $4.99 once, which is the cheapest route to a wide choice of models. The downside is the model class: some reviewers say the smaller models lose the thread within a couple of replies.
PocketPal
Free and open source. You download a GGUF model of your choice and chat with it, so it suits people who like to compare models and quantizations. There's no preloaded recommendation, so you decide what fits your phone. Community speed tests show 1B models running at a usable pace on a 4 GB iPhone 13 and larger ones much slower. If you don't want to think about model files, it's more work than the others.
Locally AI
A free assistant app built on Apple's MLX that offers models from the Llama, Gemma and Qwen families. It shows a badge for whether a model should run well on your phone. It requires iOS 18.1, and some reviewers on older iPhones report that it won't run for them at all.
Enclave
A private chat app with free use and an optional subscription, supporting several model families. It doesn't publish a RAM requirement. Reviewers on an iPhone 11 and an older iPad call it slow but one of the few that works there. Context length is a common complaint: when a chat gets long, replies can stop.
Layla
The most ambitious of the group. It downloads a large model of roughly 4 GB on first launch and adds image and music tools. The listing says it's aimed at the latest phones with 8 GB or more, and warns of degraded performance below that. It costs $19.99, and its rating is lower than the others, with several reviewers reporting errors on older phones.
Castmates
This is ours, so weigh it accordingly. Castmates runs one built-in model, tuned for roleplay, with 35 characters who stay in role and remember your story. There's no account and nothing you type leaves the phone. It needs an iPhone with 6 GB of RAM or more and a download of about 2 GB. Pricing is 30 free messages a day, then Pro at $6.99 a month, $29.99 a year or $44.99 once.
It also does voice calls that run on the phone (English voices, about a 360 MB one-time download) and imports SillyTavern and Chub character cards. For an outside view, PromptQuorum has an independent Castmates review that lists what it could and couldn't verify.
What you give up: you can't choose or import models, it isn't a general assistant for research, code or documents, and phones with 4 GB of RAM can't run it. If any of those matter more to you than characters, one of the apps above will serve you better. You can meet the cast first, or read how it compares with a SillyTavern setup on iPhone.
How to choose
- Check your iPhone's RAM first. Under Settings, General, About you can see the model name, then look up its memory.
- Decide whether you want to manage models (PocketPal, Private LLM) or just open an app and chat.
- Expect small models to forget and repeat more than a cloud chatbot. That is true of every app here.
- Battery use is higher than texting in all of them, because the phone does the work.