You can run an AI model right on your phone. No internet, no subscriptions, and your data never leaves your device. When that beats ChatGPT, and when it doesn't.
Picture this: you need to get something specific done, the kind of thing that’s easiest to hand off to AI. Analyze some text, pull data out of a photo, make sense of a chart. But you’re somewhere with bad signal, or no internet at all. Or even worse, you don’t have a subscription to ChatGPT or Claude.
What now? Open an AI that lives right on your phone.
Why run AI on your own device when there’s ChatGPT?
You can put an AI model not just on a computer, but on a huge number of modern phones. You open an app that already has the model inside, ask it to do something — and it does it. No internet, no subscriptions, no money at all. Cool? I think so.
The second thing is your data. When you type into ChatGPT, your data goes to its servers. It’s encrypted on the way, which makes sense and is the right thing to do, so intercepting it is close to impossible. But it still gets stored on the server. And by default your chats are also used to train their models, unless you turn that off in settings.
With a local model, your data stays only on your phone or computer. That’s it, it doesn’t go anywhere. You can check: turn off the internet, run a local model — and it’ll do everything you tell it to.
Isn’t it dumber, though?
Obviously, yes. On a powerful computer with tons of RAM and video memory you can run a smarter model, but even that one will fall short of the pro models from ChatGPT or Claude. And on a phone you won’t run a powerful model at all.
But for analyzing data it’s more than enough. For recognizing images — enough. For working with text — same. You can even just chat with it. The only catch: its knowledge may be outdated. But if you need something fresh, you can just give it your own files, and it’ll answer based on them.
These models are best used for routine tasks. Instead of writing some complicated algorithm.
That’s exactly what I did in my app Tenot. You hit record, dictate your note — and the local model turns it into text and puts in all the punctuation by itself. You don’t have to do a thing.
I also keep a skincare journal and photograph my skin every day: I need to track whether there’s any inflammation or flaking. I just attach a photo, tap it once — the model analyzes it and writes down, point by point, what it sees on my skin. Later I can export the whole journal and hand it to a big subscription model, so it can analyze what was going on and give me a conclusion.
What kind of phone or computer do I need?
For a phone — at least 6 GB of RAM, ideally 8 or more. And preferably, though not required, a dedicated processor for AI. Both iPhones and Android flagships have one. From my experience, the iPhone is faster here. But it’s not worth paying extra just for that: if you like Android, get Android — it’ll still be plenty fast.
Either way, it’s a trade-off. Models get smarter when they can load more parameters. And for that you need more RAM or video memory. So when it comes to hardware — the newer and the more, the better.
You can’t just install a local model on its own. A model is one big file, and to run it you need a special app. These apps show you whether your phone or computer can handle a specific model or not. Tenot will also tell you whether your phone is up to it when you try to turn on an AI feature.
So should I install one or not?
I, for one, installed an app where I can chat with different models. Basically ChatGPT, just local. You can drop in a photo so it recognizes something specific. Analyze data. Write code, if you need to. Or a birthday message for a friend.
So:
- Routine stuff — going through text, photos, data — a local model will handle it just fine.
- No internet or no subscription — it’ll save you.
- Don’t want your data flying off somewhere — local only.
- Hard task — go to ChatGPT or Claude. A local model is dumber, and not just a little — seriously dumber.