I’d recommend locally.ai[1] - it’s really good and has a wide range of models. Also has shortcuts support.

1. https://apps.apple.com/gb/app/locally-ai-local-ai-chat/id674...

My son just started using 2B on his Android. I mentioned that it was an impressively compact model and next thing I knew he had figured out how to use it on his inexpensive 2024 Motorolla and was using it to practice reading and writing in foreign languages.

Nice! Tried on iPhone 16 pro with 30 TPS from Gemma-4-E2B-it model.

Although the phone got considerably hot while inferencing. It’s quite an impressive performance and cannot wait to try it myself in one of my personal apps.

These new models are very impressive. There should be a massive speedup coming as well, AI Edge Gallery is running on GPU, but NPUs in recent high end processors should be much faster. A16 chip for example (Macbook Neo and iphone 16 series) has 35 TOPS of Neural Engine vs 7 TFLOPS gpu. Similar story for Qualcomm.

I encourage everybody to try this, if they have an iPhone. If you’re like me and don’t have the time to tinker with the latest and greatest all the time; this app lowers the barrier to entry significantly and provides a glimpse into what’s possible locally, on device.

Honestly, I was extremely impressed by the speed and quality of the answers considering this thing runs on a phone. It honestly makes me want to sit down and spin up my own homegrown AI setup to go fully independent. Crazy.

The description says it's private, but the legalese it makes you agree to makes no promise. Rather, the opposite:

> We collect information about your activity in our services

Source: https://policies.google.com/privacy#infocollect

If this works smoothly on iPhone, it could change how we think about mobile apps. Less backend dependency, more on-device intelligence.

It doesn’t render Markdown or LaTeX. The scrolling is unusable during generation. E4B failed to correctly account for convection and conduction when reasoning about the effects of thermal radiation (31b was very good). After 3 questions in a session (with thinking) E4B went off the rails and started emitting nonsense fragment before the stated token limit was hit (unless it isn’t actually checking).

Is it me or does the App Store website look... fake? The text in the header ("Productiviteit", "Alleen voor iPhone") looks pixelated, like it was edited on Paint, the header background is flickering, the app icon and screenshots are very low quality, the title of the website is incomplete ("App Store voor iPho...")

> Note: I tried to hook this one up to OpenClaw and ran into issues

Anyone worked on hooking up OpenClaw to gemma4 running locally?