Magita is an AI OS for your phone. Agents work in a Sandbox of their own and hand you the finished file — the PDF, the font, the sprite. Pay once. No account, no subscription.
An agent, not a chatbot
Chat hands you words. Magita hands you the thing itself, made on this phone and kept in your Library. Tell it a job in plain words; it works in steps you can watch, and each one says whether it finished.
Things people ask it for
• "Make this into a PDF" — and the file is on your phone, not markup to convert later.
• "Export this typeface as a font" — a real font file, out of the tool that drew it.
• "Send me that sprite" — the agent reaches into the pixel-art studio and files the image for you.
• "Boot DOS" — a FreeDOS machine, running in the Sandbox next to your work.
A Sandbox, a Toolbox, a Library
The Sandbox is the room the agents work in: run more than one job at a time, each on its own surface, and take the wheel of any of them. The Toolbox is what they work with. The Library keeps what they made, what you handed them, and your conversations — on the phone, with no cloud sync.
There is a real computer inside
A terminal with a shell, Python, ffmpeg, 7z and git, and everything you have kept sitting at /library. Guest machines boot in the Sandbox, a FreeDOS computer among them.
Six tools that need no model at all
Files, images, video and audio, documents, sprites, and typefaces you export as a real font. No key, no signal, nothing to spend.
It gets cheaper the more you use it
Ask for the same job twice and the second time is free: Magita keeps only the ways of working it can prove it repeats. Have a reply read aloud, or speak instead of typing.
It works the phone when you ask
Open a link or a settings page, pick photos or bin them, make and rename files in a folder you granted, take one screenshot and file it. Android asks your permission each time, naming the app before it looks, and every step is written down.
You stay in control
Decide up front how the agent behaves: get on with everyday work and tell you afterwards, or stop and check with you first. It asks before deleting anything or sending anything to a service you connected. What it may touch is a list you can switch off one item at a time. One page shows what has left the phone and what it did for you — on a fresh install it says, truthfully, that nothing has.
Your own key — or none at all
For real work, paste in a key you already pay for: any OpenAI-compatible or Anthropic endpoint. Your instruction, conversation context and tool results go to the endpoint you chose. You pay that provider directly. Magita resells nothing and adds nothing to your bill, and your key is encrypted on the phone by a key the phone will not hand out.
A compact model is also built in, so the app answers the first time you open it with no download and no signal. It is a floor, not a rival to a real model: it gets facts wrong and is poor at jobs needing several moves in a row. One switch holds the whole app offline when you want it.
Pay once
One purchase, and then it is yours. No subscription, no ads, no analytics, no feed, no account to make.
What it does not do
• Work pauses when the app goes to the background. It is not getting things done while your phone is locked.
• Your own provider costs whatever that provider charges you. There is no included allowance.
• It does not tap around your other apps, and there is no unattended device control. What it does on the phone is the short list above, each step permissioned and recorded.
• The record it keeps is this app's account of its own work, not a firewall over the whole phone.
Before you buy
Android 7.0 or later, on a 64-bit device. Uninstalling removes the app's own data, so export anything you want to keep.
If you would rather not hold an API key, judge Magita on the built-in model alone. Decide before you buy, not after.
An AI OS for your phone: an agent that finishes the job. Pay once, no account.