Private AI

Private AI, on your own computer.

An AI model that runs on a computer in your business, for the documents and records you would rather not send to an AI company. Set up by Andrei, with the trade-offs said up front: it is a small model, it needs a capable computer, and it is only as quick as that computer.

01 — What it is

What private AI on your own computer is

A model on your computer

An AI model is a file you download to your computer. A free app loads it into the computer's memory, and it answers there, on the machine.

What you type stays there

LM Studio, the free app we set up, says nothing you enter when chatting leaves your device, and a document you drop in to chat with stays on your machine.

The internet for the app's own jobs

The app does go online to search for and download models, to get the engines that run them and to check for updates. The work itself needs no connection.

Privacy law still applies

Running AI locally doesn't switch off privacy law. Australia's privacy regulator, the OAIC, says privacy obligations apply to any personal information put into an AI system.

Try it yourself: run AI locally, step by step →

02 — Who it's for

For records that must stay in the building.

A business with a rule that certain records stay off the cloud. If that is you, tell us, and the system is built around your rules.

Good first jobs, we think

The ones where privacy is the point: a summary of a contract before you call your lawyer, a first draft of a staff policy, notes on a supplier's proposal.

Not the right fit if

You want AI inside your email, calendar or books: set up this way, it sees only what you type or attach. Or privacy isn't what's holding you back: then a business account of ChatGPT or Claude, set up properly, will do more for you.

03 — What Andrei sets up

The app, a model that fits, and a test with the internet off.

01

Your computer checked first

LM Studio runs on Macs with an Apple Silicon chip and macOS 14.0 or newer, and on Windows PCs with an x64 processor that supports AVX2 or an ARM one; it recommends 16 GB of memory or more. If yours falls short, the plan says so before anything is installed.

02

The app and a model that fits

LM Studio and a small open model such as Google's Gemma 4 E4B — a 5.90 GB download — in a version your computer can run.

03

Tested with the internet off

Wi-Fi off, cable out, then a question. If it still answers with no connection at all, the answer can only be coming from your computer. That test comes before anything private goes in.

04

Your first jobs, on your documents

Summaries and first drafts of the documents you choose, and your team shown how to check what it wrote against the document.

05

Yours to switch off

It runs on your computer, under your rules, and can be switched off or removed at any time.

Running Where it already runs

Our own website's chat: when you type a question, an AI model on Andrei's own computer reads it and picks which of our approved answers fits. It never writes a reply of its own, and when that computer is off the chat answers as it always has.

04 — The trade-offs, said plainly

A small model, and only as quick as your computer.

Speed depends on the computer

When we tested the model now behind our chat on 30 real questions (10 November), writing each answer took a median of about 2 seconds on our computer's graphics card, and about 11 on its processor alone.

A small model

It won't match Claude or ChatGPT for most jobs, we think. Google says its models with fewer parameters are less capable, and that open-ended or highly complex tasks might be challenging.

It needs a capable computer

LM Studio recommends 16 GB of memory or more. On an 8 GB Mac it says to stick to smaller models and modest context sizes, and it doesn't support Intel-based Macs at present.

Older knowledge, long documents

Google says Gemma 4's pre-training data has a cutoff date of January 2025, and offline it can't look anything up. For a very long document, LM Studio picks out the parts that look relevant instead of reading the whole thing at once, which it says sometimes needs some tuning and experimentation.

How we handle your data, in full →

05 — How it starts

It starts with the AI Setup Session.

Our usual first step — two hours with Andrei, at your business in Melbourne or remotely — set up for a model on your own computer instead of Claude or ChatGPT. It starts with your computer: if it meets the requirements, the app and a model go on, the test with the internet off comes next, then your first jobs on your own documents. If it doesn't, you hear that first.

06 — Questions people ask

Is it as good as Claude or ChatGPT?

Not for most jobs, we think. Google says that in its own Gemma 4 family, models with fewer parameters and lower precision are less capable but may be enough for your task, and that open-ended or highly complex tasks might be challenging. For a draft, or a summary of a document you attach, a small model is worth trying.

Does anything I type leave the computer?

LM Studio says nothing you enter when chatting leaves your device, and a document you drop in to chat with stays on your machine. It does use the internet for some things, among them searching for models, downloading them and checking for app updates.

Do I need a new computer?

Not if yours meets LM Studio's requirements: a Mac with an Apple Silicon chip and macOS 14.0 or newer, or a Windows PC with a processor it supports, with 16 GB of memory or more recommended. The setup starts by checking yours.

Does this make us compliant with privacy law?

We don't promise that. Running AI on your own computer changes where your data goes, not what the law asks of you: Australia's privacy regulator, the OAIC, says privacy obligations apply to any personal information put into an AI system. We set it up around your own rules.

Can it work inside my email or my books?

Not set up this way: it sees only what you type or attach. If you want AI inside your email, calendar or books, Claude can be connected to Xero, MYOB, Gmail and Outlook — a different set-up, in your own accounts.

Can we use it for our business?

Yes. LM Studio has been free to use at home and at work since July 2025, and its terms allow personal and internal business use. Google says Gemma models permit responsible commercial use.

How much does it cost?

There are no prices on this page, because no two set-ups are the same. After a 15-minute call you get a fixed quote in writing, before any work starts. The first plan costs nothing, and there are no lock-ins.

Put a private AI on your own computer.

Two hours with Andrei, at your business or remotely, starting with a check of your computer.

SEE THE AI SETUP SESSION →

From the blog