No limits
Cloud apps count your messages, and slow you down when you use too many. Local AI has no meter. Ask as much as you like.
The short guide
Everything you might wonder about running AI on your own Mac, in plain English. No jargon, and no experience needed.
About a five-minute read.
1 · The basics
ChatGPT, Gemini and Claude are cloud AI. When you send a message, it travels over the internet to powerful computers in a data center. The AI writes its answer there, and the answer travels back.
Local AI is the same idea with one change: the AI is a file on your own computer, and your computer does the thinking. Nothing has to travel anywhere.
Think of streaming a movie versus owning it. Streaming works anywhere with a connection, for as long as you keep paying. Owning means it’s on your shelf, it plays any time, and nobody can take it back.
Your question goes to a data center. The AI answers there, and the reply comes back.
The AI is a file on your Mac. Your Mac does the thinking, so there is nothing to send.
2 · The upside
Cloud apps count your messages, and slow you down when you use too many. Local AI has no meter. Ask as much as you like.
Your questions, documents and photos stay on your computer. No company stores them, and no one else can read them.
There is no subscription, because there is no server bill. The AI models are free to download. What matters is a Mac with enough memory.
On a plane, in a cabin, on a bad connection: it still works, because nothing needs the internet once the AI is set up.
3 · The honest part
Local AI isn’t magic. Here is what you give up, so nothing surprises you.
The biggest cloud AIs run on enormous computers, and they are smarter than anything that fits on a laptop. Local models are smaller. They are very good at everyday jobs, and weaker at the hardest problems.
The AI has to fit in your Mac’s memory while it works. 8 GB runs small, quick models. 16 GB or more runs much smarter ones.
The AI itself is a file of a few gigabytes. You download it once, and then it is yours. Mighty comes with a first one already inside.
Local AI doesn’t browse the web. It knows what it learned up to a point in the past, and nothing after that.
Every AI sometimes says something false with complete confidence. Check anything that matters.
| What is compared | Cloud AI | Local AI, like Mighty |
|---|---|---|
| Message limits | Yes, on most plans | None |
| Where your chats go | A company’s servers | Stay on your Mac |
| Cost | A free tier with limits, or a monthly fee | Free |
| Works offline | No | Yes |
| Hardest problems | Best | Good, not the best |
| Today’s news | Often, by searching the web | No |
| Getting started | Sign up for an account | Download and open |
Finish the sentence
4 · Under the hood
An AI model has read a huge amount of text and learned which words tend to follow which. When you ask it something, it doesn’t look up an answer. It predicts what comes next, one small piece at a time, and each piece shapes the next.
Try it. Finish the sentence, and watch the AI weigh its options.
A toy example. A real AI weighs tens of thousands of options at every step, using your whole conversation. Those small pieces are called tokens.
5 · Your Mac
Almost certainly. What matters most is memory, also called RAM. Choose yours to see what to expect.
Not sure? Open the Apple menu, then About This Mac.
Mighty looks at your Mac when you first open it, and suggests models that fit.
6 · Jargon, translated
You don’t need any of these to use Mighty. Here they are in plain words, in case you run into them.
The AI itself: a big file full of everything it learned. “Download a model” just means “download an AI.”
Running on your own computer, instead of on someone else’s. The opposite of cloud.
A rough measure of a model’s size. “3B” means 3 billion. More parameters usually means smarter, slower, and hungrier for memory.
A small piece of text, about three quarters of a word. AI reads and writes in tokens. Mighty’s dashboard counts them.
How much text the AI can keep in mind at once: your conversation so far, and any files. When it fills up, the oldest parts fall out. Mighty can compress a long chat to make room.
Your Mac’s short-term working space. An AI model has to fit in it while it runs. It’s different from storage, which is where your files live.
A model whose files anyone can download and run, like Llama, Gemma and Qwen. Open models are what make local AI possible.
An AI that can look at pictures as well as read text. Useful for photos, screenshots and scanned pages.
An AI that works through a problem step by step before it answers. Slower, but sharper on hard questions. Mighty shows when it’s thinking.
Free software that runs open models on a computer. Mighty has its own copy inside, so you never have to install or touch it.
7 · Where Mighty fits
You can run local AI by hand: install an engine, find a model, learn a few commands, and chat in a Terminal window. It works, but it’s not for most people.
Mighty does all of that for you, and puts it in a chat app that feels like the ones you already know.
The software that runs the AI is built in. You never install or start anything.
It looks at your Mac and suggests AI models that fit it.
Characters, attachments, a quick-chat tab on your screen edge, and more.
8 · Still curious?
For everyday jobs like writing, summarizing and explaining, it is close enough that many people don’t notice the difference. For the very hardest problems, the biggest cloud AIs are still ahead. A Mac with more memory narrows the gap.
Mighty downloads models from the Ollama library, a well-known public collection, and checks the engine it downloads against a checksum. As with any download, stick to sources you trust.
While it is writing an answer, yes: the AI works your Mac’s chip hard, so a laptop’s battery drains faster and the fans may spin. When you are not chatting, it rests.
Only the files you put into a chat. It doesn’t browse the web, and it doesn’t look around your Mac on its own.
Open models are published by companies like Meta (Llama), Google (Gemma), Alibaba (Qwen), Microsoft (Phi), OpenAI (GPT-OSS) and Liquid AI (LFM). Anyone can download them.
Because the AI itself is inside. A few gigabytes holds everything a model has learned.