Start here · 3 min read
What is a local model?
A plain-English explanation of AI models, the difference between local and cloud, and why running AI on your own Mac matters for privacy and cost.
You'll see the word "model" everywhere in Osaurus. Here's what it means, why some models run on your Mac and others don't, and how to think about the difference. No technical background needed.
A model is the brain
A model is the part of an AI that reads your message and writes the reply. ChatGPT, Claude, and Gemini are apps built around models. So is Osaurus.
The important idea: Osaurus and the model are two separate things. Osaurus is the app. It keeps your chats, remembers what matters to you, runs your agents, and connects to your apps and files. The model is the brain it asks for each reply.
Because they're separate, you can swap the brain whenever you like. Your agents, memory, and tools stay exactly where they are.
Local or cloud: where the thinking happens
Models run in one of two places.
- A local model is downloaded to your Mac and runs there. It works with no internet connection, and nothing you type leaves your computer.
- A cloud model runs on another company's servers, like OpenAI's or Anthropic's. Your message travels over the internet to them, and the reply comes back.
On Macs running macOS 26 or later, there's a third option: Apple Intelligence, Apple's built-in model. It runs on your Mac too, and there's nothing to download. It shows up as Foundation in the model picker.
Why local is a big deal
Privacy. With a local model, your questions, documents, and replies never leave your Mac. That makes it safe to ask about contracts, health questions, client work, or anything you'd rather not upload.
It works offline. On a plane, on a train, or on bad hotel Wi-Fi, a local model works the same.
It's free to run. Once a model is downloaded, using it costs nothing. There's no per-message charge and no monthly cap.
It's yours. The model file sits on your disk. No one can change it, remove it, or raise its price.
When cloud makes more sense
Local models are excellent at everyday work: writing, summarizing, sorting mail, and working with files. The largest cloud models are still stronger at some things, such as very long, complex reasoning or tricky coding problems.
They also need no disk space or memory on your Mac, which helps on older or smaller machines.
The trade is that your message, and anything sent with it, goes to that company. Their terms apply.
Tip: You don't have to choose forever. Many people use a local model by default and switch to a cloud model for the occasional hard question. It's one click in the chat box.
The words you'll see
| Word | What it means |
|---|---|
| Model | The AI brain that reads and writes |
| Local model | A model downloaded to your Mac. Works offline. |
| Cloud model | A model on a company's servers. Needs internet. |
| Parameters | A rough size, written like "4B" (4 billion). Bigger is usually smarter but needs more memory. |
| Quantization | Shrinking a model so it fits, written like "4-bit". Smaller and faster, with a small loss in quality. |
| MLX | Apple's technology for running models fast on M-series chips. Osaurus uses it. |
| Context window | How much of the conversation the model can keep in mind at once. |
Run into another word you don't know? The Osaurus glossary explains them all in plain English.
What to do next
If you picked Keep it private, on this Mac during setup, you're already running a local model: Raptor 0.7. Nothing else to do.
If you want to try a different one, or you're wondering whether your Mac can handle a bigger model, read Choosing a model.
New posts, in your inbox
Subscribe to get new blog posts and product updates by email.