Setting up a private AI on your computer

What installation actually looks like · about a 5-minute read

The short version: with most private-AI tools, "setup" means opening a terminal and choosing model files. With Garnet, there's nothing to choose and nothing to configure — you open it once and it sets itself up for your machine. This page walks through exactly what that looks like, and how to prove the privacy for yourself once it's running. Garnet is in early access, so the first step is getting on the list for the download.

1. Check your computer can run it

A private AI runs the model in your computer's memory (RAM), so the only real requirement is having enough room. The guide-rails:

8 GB of memory — runs a genuinely useful everyday model. This is where most laptops land.
16 GB — a noticeably smarter model; the sweet spot for most people.
32 GB and up — room for larger, sharper models.

It runs on Mac or Windows, and Apple-Silicon Macs (the M-series) punch above their memory because they share it cleverly with the chip. Not sure what you have? The 10-second checker on the homepage tells you in plain words where you land — smooth, fine, or maybe go with a setup — without you needing to know a single spec.

2. Get the app

Garnet is in early access right now, which means the download isn't open to everyone yet. Join the early-access list and we'll email you the link the day it's ready — one email, the day it lands, nothing else. No account to create, nothing to subscribe to.

Why a list and not an instant download? We'd rather a small group has a flawless first run than a big one has a rough one. You'll be early, not last.

3. Install it in one step

This is the part that's different. With the do-it-yourself tools, installing is where you meet the command line, the model catalogue, and the settings. With Garnet, you open the app and it sets itself up:

No terminal. Nothing to type, no commands to copy from a forum.
No model-picking. Garnet looks at what your computer can carry and fits a model to it, so you get the sharpest version your hardware allows — without having to know what any of that means.
No configuration. There's no settings file to edit before it works. It's ready when it finishes setting up.

The first setup downloads the model — a single large file — so it takes a few minutes on a normal connection. After that, it's all on your computer and starts instantly, online or off.

4. Start using it

You talk to it like any assistant — a plain chat window. A good first thing to try is something real you'd never paste into a cloud chat box: a sensitive email to rewrite, a confusing letter to explain, a decision to think through out loud. If you want a running start, the starter guide has ten prompts to copy straight in.

5. Prove the privacy yourself

Don't take our word that it's private — test it. Once Garnet is running, pull your internet cable (or turn off Wi-Fi) and ask it something. It keeps right on answering. That's the whole argument in one move: if it still works with nothing connected, then nothing you type was ever going anywhere. There's no account to hijack and no database of your chats to leak, because neither exists. If you want the precise version of what "private" means here, here's exactly what we can and can't see.

Before the optional upsell, the honest part: for almost everyone, the free app sets itself up fine and you'll never need anything else. The next section is only for people who'd genuinely rather hand the whole thing to someone — no judgment, and no funnel if you skip it.

Or have us set it up for you

If "open the app and it sets itself up" is still more than you want to deal with — or you want it tuned around your specific work — we'll do it on your computer, with you, in a guided session. The same conversation also scopes the bigger end: a model larger than a laptop can hold, or a dedicated, always-on system on its own hardware, built by the engineer who stands behind it.

Book a setup session A scoping call is free, no card. You only pay if you decide to go ahead.

If something doesn't fit

A few honest edge cases:

An older or smaller computer still runs Garnet — just with a smaller model, so expect quick-and-useful rather than its sharpest. The checker will tell you plainly if that's you.
Want a model bigger than your machine can hold? That's not a setup problem, it's a hardware one — and it's exactly what a dedicated build is for.
Comparing it to the developer tools first? Here's an honest look at Garnet vs Ollama, LM Studio, Jan and LocalAI so you can pick the right one.

More: the plain-English starter guide · using it for sensitive documents · what we can and can't see · back to top.