Install private AI on your computer

Private AI is an AI chat that runs on your own computer instead of on someone else’s servers, so what you type stays with you. This guide takes you from nothing to a working private AI chat in about 20 minutes, using a free app with a familiar chat window, and shows you how to prove it really is running on your computer.

Before you start

• A computer that can run private AI. If you’re not sure, start with the “Check if your computer can run private AI” guide.
• At least 20 GB of free disk space, as a rough guide. Models are large files.
• An internet connection for the download. After that, chatting works offline.
• Permission to install apps on the computer. On a work computer, check with whoever looks after it first.
• About 20 minutes, plus download time.

Want to do it yourself?

Difficulty

6 out of 10

Estimated completion time

45–90 minutes, plus model downloads

Skills you’ll need: Install a desktop app, choose a model that fits your memory, and confirm it works offline on your computer.

I’ll do it myselfI’d rather get help

How this helps your brand

A tested local chat gives you another way to draft and summarise everyday material on your own computer. Confirming offline operation helps you understand the setup before using business information with it.

1. 1

   Step 1: Choose one app to start with

   Several free apps let you download an AI model and chat with it on your own computer. Pick one; you can try others later.

   • LM Studio (lmstudio.ai): a chat window that will feel familiar, with a built-in search for models. The easiest place to start for most people. Check its requirements page first, as it doesn’t support every computer, such as older Intel Macs.
   • Ollama (ollama.com): lighter, and often used behind other apps. A good pick if another tool you use asks for it.
   • AnythingLLM (anythingllm.com): built around chatting with your own documents. See the “Ask questions about your own documents privately” guide.

   Tip: This guide follows LM Studio. The others work in a similar way, and each has getting-started pages on its official site.

2. 2

   Step 2: Download it from the official website

   Go to lmstudio.ai and download the version for your computer. Only download from the official site; copies elsewhere can carry unwanted software.

   Open the file you downloaded and follow the usual install steps for your computer. On a Mac, that usually means dragging the app into your Applications folder. On Windows, run the installer and follow the prompts.

3. 3

   Step 3: Download one small model

   The app is just the window. The model is the “brain” that writes the answers, and you download it separately.

   In LM Studio, open the Discover section and browse or search for a model. Pick a small, popular one from a well-known maker. The app often shows whether a model is likely to fit your computer. Download just one for now; it may take a while.

   Tip: Sizes are shown with a number and a B, like 3B or 8B. With 8 GB of memory, start with the smallest. With 16 GB or more, a mid-size model is a sensible first try. These are rough starting points; check the model’s page for its current needs.

   Illustrated exampleThe app and the model do different jobs

   Install onceLM Studio

   The app gives you the chat window and model picker.

   +

   Download separatelyOne small model

   The model is the file that generates the answers.

   1. Discover → downloadSave the model to your disk.
   2. Chat → loadPut that model into memory.
   3. Send a messageGet your first local answer.

   Start small: a label such as 3B describes model size. Check the model page and the app’s fit guidance before downloading.

4. 4

   Step 4: Load the model and say hello

   Go to the Chat section, choose your downloaded model from the model picker at the top, and wait for it to load. Loading puts the model into your computer’s memory, which can take a moment.

   Then type a simple message and press Enter. If you get an answer, it’s working.

5. 5

   Step 5: Prove it runs on your computer

   Turn off your Wi-Fi or unplug your network cable, then ask another question. If it still answers, the model is running entirely on your computer. LM Studio’s help pages confirm that chatting with a downloaded model works offline.

   Turn your internet back on afterwards. The app still uses the internet to search for and download models and to check for updates.

   Illustrated exampleTest a fresh reply while you are offline

   1. 1
      Load your downloaded modelGet a first answer while connected.
   2. 2
      Disconnect the internetSwitch Wi-Fi off or unplug the cable.
   3. 3
      Ask a new questionA fresh reply is the useful test.

   Internet disconnected

   Downloaded model · loaded

   You

   Give me three subject lines for a customer thank-you email.

   Local model

   1. Thanks for choosing us
   2. A quick thank-you
   3. We appreciate your support

   A new reply, with no internet

   Reconnect afterwards for model downloads and app updates. This is an illustrated test, not an exact LM Studio screen.

6. 6

   Step 6: Try it on a real task

   Give it something you’d normally do yourself: tidy up a rough email, summarise a page of notes, or suggest three subject lines. Compare the result with what you’d get from the online AI tool you usually use.

   Expect it to be a little slower and a little less polished. If it’s too slow or the answers are weak, the “Get better answers from private AI” guide covers what to change.

Here are my rough notes for an email to a customer. Turn them into a short, friendly email in plain English. Keep it under 120 words and don’t add any facts that aren’t in my notes. If something is missing, put it in [brackets].

Who it’s to: [for example, a regular customer]
My notes: [paste your rough notes]