LumaBrowser gives you a ChatGPT-style chat box that runs on the PC or Mac you already own. Download it, click Easy Setup, and it picks a model that fits your machine, downloads it, and opens a chat. No subscription, no API key, and nothing you type goes anywhere.
Rather see it first? Try a small model in your browser, no install. Want images, stories, or games too? Start on the Create page.
Three kinds of machine, in plain words. You do not need to know what is inside your computer: Easy Setup checks for you and tells you which row you are before it downloads anything.
Three plain questions, one click, then a wait while the model downloads. There is no terminal, no settings file, and nothing to choose that you do not understand.
Memory, graphics card, free disk space. Nothing is sent anywhere; it just reads what is there.
You get the biggest model your machine can run comfortably, and one sentence explaining the choice and how big the download is.
The download can be paused, and it resumes if your connection drops or you close the app. Nothing starts over from zero.
A chat box opens. Ask a question, paste a document, or ask "how good are you?" and it will tell you honestly.
Step-by-step, with the install prompts for each operating system: the getting-started guide.
Most people who quit local AI quit in the first session, because nobody told them what "good" looks like on their machine. Here is the straight version.
A small model on a laptop is a capable helper for questions, rewrites, summaries of short text, and drafts. It is not ChatGPT, and it will get tricky reasoning wrong more often.
A mid model on a gaming PC is close to what a free chatbot tier gives you. A large model on a high-end PC or a 32 GB+ Mac feels like a paid chatbot for most everyday work. None of them have browsed today's news unless you turn on web browsing.
With a gaming graphics card the reply appears faster than you can read. On a high-end card it is effectively instant. Without a graphics card the reply arrives like someone typing, and that is normal, not a fault.
The first reply after opening the app takes a few extra seconds while the model loads.
Your conversations are files on your own disk. There is no account, no cloud copy, and no one at LumaByte can read them. The app sends one small check-in with its version number so we can count installs; that is all. Privacy policy.
Free, and no subscription. You will not save money versus paying per message unless you chat a lot; you do this for privacy, for working offline, and because it is yours. Electricity is the only bill.
Everything below is already in the same download, switched off until you ask for it. None of it is required to chat.
Drop in PDFs, notes, or text files and ask questions about them. Answers cite the source. Knowledge base.
Let the assistant open real web pages to look something up. It shows you what it read. Chat and artifacts.
Add image generation later with one click if you have a graphics card with 8 GB or more. Create page.
Talk instead of typing, and hear the reply, all offline. Windows and Linux in this release. Voice mode.
Not quite, and it depends on your machine. Laptop without a gaming card: a small model, good for questions and drafts, weaker on long documents and hard reasoning. Gaming PC: a mid model close to a free chatbot tier. High-end PC or 32 GB+ Mac: a large model that feels like a paid chatbot for most tasks. What you gain at every level is privacy and no subscription.
Speed comes from your graphics card. Without one, the model runs on the processor and reads like slow typing, which is normal. If you have a gaming card and it is still slow, the model is probably too big for the card and spilling into system memory: open the Setup tab and pick a smaller model, or run Easy Setup again. Old graphics drivers also slow things down; update them.
The app is under 200 MB. The model is 2 to 5 GB on a laptop, 6 to 20 GB on a gaming PC, 20 to 50 GB on a high-end machine. Easy Setup shows the exact number before it downloads. Keep 10, 30, or 80 GB free.
The installer is not code-signed yet, so Windows SmartScreen and some antivirus products warn about it like any new unsigned file. Click More info, then Run anyway. The download page lists the size and checksum of every file so you can confirm you got the one we published.
Yes. Models live in one folder you can move to another drive from Settings, and each model can be deleted from the Setup tab. Uninstalling the app does not remove that folder unless you delete it.
No. The model runs on your machine and conversations stay there. The app sends a small version check-in so we can count installs; nothing else.
It resumes from where it stopped, even after closing the app or rebooting. You can also pause it on purpose.
While a reply is being written, yes, the same as a game would. Between replies it idles. Plug in for long sessions; battery drains faster than normal use.
Yes. Paste your own OpenAI or Anthropic key and switch per conversation. We do not mark up their prices. Local is the default and needs no key.
Download LumaBrowser, click Easy Setup, wait for the download, start chatting. Free, private, and yours.
Curious how it works underneath? Models, memory, and speed explained. Want every knob? The advanced page.