What Locally Uncensored is
One app for chat, an agent, a coding tab and image, video and audio generation. This chapter explains what runs where, what needs a graphics card, and what is free.
One window, four tabs
The top bar of the app has these entries: Chat, Create, Compare, Benchmark, Models and Settings. Most of your time goes into the first two.
Chat is a conversation with a language model. At the top of a chat you can switch the conversation into Agent mode, where the model gets tools (web search, files, a shell, screenshots, image and video generation), or into the Code tab, where a coding agent works inside a project folder with a file explorer and a plan. Chapters 4, 5 and 6 cover those three.
Create is where images, video and audio are made. Its toolbar has twelve lanes: Image, Edit, Cutout, Upscale, Erase, Video, Animate, Character, Lipsync, Music, Extend and Motion. Chapter 7 covers them.
Compare puts two to four models side by side on the same prompt. Benchmark measures a model on your own hardware for speed, cost and correctness. Models is where you download, delete and inspect models. Settings has four tabs, which chapter 9 lists.
Local on your machine versus hosted on ours
The app has two ways to do its work, and a switch next to the model picker moves between them.
Local means the model runs on your computer. The app ships its own engine, called the LU Engine, which runs the model files you download. It can also talk to an engine you already have, such as Ollama or LM Studio, or to any server that speaks the OpenAI API. Images and video run through a ComfyUI that the app installs, starts and repairs for you. In local mode nothing you type leaves your machine. The only network traffic the app makes on its own is the check for updates against GitHub Releases, the model downloads you start, and one anonymous daily count when you press the Cloud switch (which way it was pressed, platform and app version, nothing else). Settings, General, Privacy states this in the app itself.
Hosted means the same window, but the model runs on GPUs that LU Labs rents. This is LU Cloud. You sign in with an account, the work is metered in credits, and the same account also works in a browser at lu-labs.ai without the desktop app. Hosted requests leave your machine together with anything you attached to them. Chapter 8 explains the account, the credits and the settings that come with it.
Going into the cloud takes a second click within six seconds, because the switch sits next to the model picker and one stray click used to be enough to start billing. Going back to local is a single click, and the message box shows which side you are on.
What needs a graphics card
Local chat works without a dedicated graphics card. The LU Engine runs the model on the processor when it has to, and the app says so in the status line when that happens. It is slower, and a 7B model on a processor alone answers at a walking pace, but it works.
Local chat is much faster on a graphics card with enough memory (VRAM) to hold the model. The starter model the wizard offers needs about 6 GB of VRAM for a full GPU run. The Models tab shows, for every model, how much memory it needs, and marks what fits your card.
Local images and video need ComfyUI and a graphics card. Image generation starts at around 6 GB of VRAM with the Juggernaut XL starter checkpoint. Video wants more, and the setup card for each lane names the model it downloads and the VRAM it needs. Two lanes, Upscale and Erase Object, run hosted only; the Character Studio runs locally on a card with 12 GB or more and also hosted. On an AMD card the picture is more complicated, and the FAQ in chapter 10 has the honest version.
If your card is too small, or you have no card at all, the hosted route runs the same lanes and the same catalogue of chat models on rented GPUs. That is what LU Cloud is for.
What is free
The desktop app is free and open source under the AGPL-3.0 licence. Everything that runs locally is free: chat, the agent, the Code tab, images, video, music, the Character Studio trainer, voice in and out, document chat, memory and remote access from your phone. There is no account, no usage limit and no time limit on the local side.
LU Cloud costs money, because the GPUs cost money. You pay either with a monthly plan or with a credit pack that has no subscription behind it. What the plans and packs cost is on the pricing page, which takes its numbers from the same source the checkout uses, so this handbook does not repeat them.
Cloud providers of your own (OpenAI, Anthropic, OpenRouter, Groq, Together, DeepSeek, Mistral, or any OpenAI compatible address you paste in) are billed by that provider under your own key. The app only carries the key and the request.
Which computers it runs on
Windows 10 and 11 on x64, and Linux on x64 as an AppImage, a .deb or an .rpm. There is no Mac build and there has never been one; on a Mac the hosted studio runs in the browser at lu-labs.ai. A phone is not a place to run the app either, but a paired phone can drive the app on your PC over your network, which the Remote tab in the sidebar sets up.
Next chapter: Install. Back to the handbook overview.