Home / Getting started

Getting started: from download to your first private chat

This is the whole first run, screen by screen, for someone who has never installed a local AI before. It takes about 15 minutes on a normal connection, and most of that is the model downloading. You do not need to know anything about models, memory, or graphics cards; the app checks your machine and picks for you.

Here for images? Getting started with images. Building an agent? Agent quickstart.

Before you download: will it run?

Almost any PC or Mac from the last six or seven years can chat. What changes with your hardware is how big and how fast the model is:

• Laptop, no gaming graphics card, 8 to 16 GB memory: a small model, 2 to 5 GB download, replies arrive like slow typing. Keep 10 GB free.
• Gaming PC with an NVIDIA or AMD card (8 to 16 GB video memory): a mid model, 6 to 20 GB download, replies faster than you read. Keep 30 GB free.
• High-end PC (24 GB+ video memory) or Mac with 32 GB+: a large model, 20 to 50 GB download, effectively instant. Keep 80 GB free.

Not sure which you are? It does not matter; the wizard tells you before it downloads anything. More on what each tier feels like on the what to expect section.

Windows

1. Go to the download page and click Download for Windows. The file is called LumaBrowser Setup 1.8.1.exe and is about 130 MB.
2. Open it from your Downloads folder.
3. A blue box may appear saying "Windows protected your PC". This is SmartScreen, and it says this about any new program that is not yet code-signed. Click More info, then Run anyway.
4. The installer runs without asking anything else, adds LumaBrowser to the Start menu, and opens it.

macOS

1. On the download page click Download for macOS. LumaBrowser needs an Apple Silicon Mac (M1 or later); to check, open the Apple menu and choose About This Mac. "Chip: Apple M..." means you are set.
2. The download is a .zip. Double-click it in Downloads and it unpacks to LumaBrowser.app.
3. Drag LumaBrowser.app to your Applications folder.
4. Right-click (or hold Control and click) the app and choose Open. A dialog says the developer cannot be verified; click Open again. You only do this once. A normal double-click on the very first launch will only show a warning with no Open button.

If macOS still refuses, open Terminal (Applications, Utilities) and paste:

Voice input is not available on macOS in this release. Everything else, including chat on Apple Silicon, is.

Linux

The download is an AppImage, a single file that runs without installing. Stock Ubuntu 24.04 (and most recent Debian-based distributions) is missing a few libraries it needs, so install those first. This one line covers the app and the chat engine:

Then make the file executable and run it:

• Ubuntu 22.04 and older: the packages are libfuse2 and libasound2.
• Fedora: sudo dnf install fuse-libs nspr nss alsa-lib libgomp.
• libgomp1 is used by the chat engine, not the app, so if you skip it the failure shows up only after the model has downloaded. Install it now.
• If the app still refuses to start, the error names the missing library. Email it to us and we will add it to this list.

The setup wizard

The first time LumaBrowser opens you get a short wizard. Here is every screen and what to click.

Screen 1: Welcome

Two choices: Automatic Local Setup and Guided Setup. Pick Automatic. (Guided is for people who want to type in an API key for ChatGPT or Claude, or choose their own model. You can always do that later from Settings.)

Screen 2: License

This screen asks for a license key. You do not have one and you do not need one. LumaBrowser is free with no paid plan. Leave the field empty and click Continue (or Skip). This screen is left over from an earlier plan and will be moved out of the wizard.

Screen 3: Three questions

Automatic Local Setup asks three plain questions. There are no wrong answers; they only nudge the choice of model:

1. What will you mostly use it for? Everyday questions, writing, coding, and so on.
2. What is the slowest reply speed you would put up with? Slower answers let the wizard pick a bigger, smarter model.
3. How much conversation memory do you want? Longer memory costs speed on smaller machines.

Below the questions: Also set up image generation? Say yes only if you have a gaming graphics card with 8 GB of video memory or more (and, for now, you are not on Linux). You can add images later from chat with one click. If a music question appears, say no; it is a preview feature that needs a very large card.

Screen 4: The plan

The wizard shows what it detected (your memory, graphics card, and free disk) and what it picked: the model name, one sentence on why, the download size, and the disk space needed. If the download looks too big for your connection or your disk, click Customize and choose a smaller model; otherwise click Start.

Screen 5: Downloading

The wizard first fetches the chat engine (a few hundred megabytes), then the model. The model is the long part: a few minutes on a laptop, up to half an hour or more for the largest picks. You can pause it, close the app, or reboot; it carries on from where it stopped. When it says "file 2 of 2" it is not downloading the same thing twice; the engine and the model are separate files.

Screen 6: Review

A summary of what was installed. Click Finish. If it says images were skipped or failed, chat still works; the image guide covers adding them afterwards.

Your first chat

The chat opens with a text box and a few suggestion chips. The first reply takes a few extra seconds while the model loads into memory; after that replies start immediately. Try:

• "Explain what you can and cannot do in three sentences."
• Paste an email and ask for a shorter, friendlier version.
• "Plan a three-day trip to Lisbon for two people who like food and walking."

What to expect from the answers depends on your tier: a laptop model is a capable helper but not ChatGPT; a gaming PC model is close to a free chatbot tier; a high-end model feels like a paid one for most tasks. The honest version is on the what to expect section.

If something goes wrong

"It is really slow"

Without a graphics card, slow typing speed is normal. With a gaming card, slowness usually means the model is too big for the card and part of it is running in system memory. Open the Setup tab and pick a smaller model, or update your graphics driver; old drivers are the second most common cause.

"The download stopped"

Click Resume, or just reopen the app; it continues from where it was. If the disk is full, free some space or move the models folder to another drive from Settings.

"It gave a wrong answer"

Small models do that more than ChatGPT does; it is the trade for privacy and no subscription. Ask it to check its answer, or turn on web browsing for anything that needs current facts.

"I want it off my computer"

Uninstall the app the normal way for your OS, then delete the models folder (Settings shows its location; on Windows it is under %LOCALAPPDATA%\LumaBrowser). Models are big, so do not skip that second step.