The Locally Uncensored handbook
Everything you need to install the app, run a model on your own machine, use the agent, the Code tab and the Create studios, add LU Cloud when your hardware runs out, and fix the usual problems yourself. Written for version 3.0.5 from the app's own source.
Locally Uncensored is a desktop app for Windows, Linux and Apple Silicon Macs that runs AI models on your own computer. Chat, an agent that can use tools, a coding tab, and a Create tab for images, video and audio all sit in one window. When a model is too big for your graphics card, the same window can hand the work to LU Cloud, the hosted twin of the app that also runs in a browser at lu-labs.ai.
This handbook is written for someone who has never used a local AI app. Each chapter explains what a part of the app is for, then walks through it in numbered steps with the button labels exactly as they appear on screen, and ends with the things that usually go wrong and how to fix them. Every fact comes from the app's source code for version 3.0.5, so where the code does not settle a question, the handbook says so instead of guessing.
Chapters
1. What Locally Uncensored is
Local on your machine versus hosted on ours, what needs a GPU, and what costs nothing.
2. Install and update
Windows installer, Linux packages, the SmartScreen warning, updates, file locations, uninstall.
3. First start: the setup wizard
The wizard step by step, with the exact button labels, and why the starter model is 7B.
4. Chat
Model picker, context window, thinking, sampling controls, personas, memory, documents, voice, export.
5. Agent mode
Tools, the workspace folder and the sandbox, approvals, Stop, background agents, loops.
6. The Code tab
What it is for, the modes, the plan bar, the folder rules, the token limit note, the sign in.
7. Create: images, video, audio and characters
Image, video, audio, the Character Studio and the LoRA trainer, with local requirements and the hosted route.
8. LU Cloud
Account, plans and packs, the Flash allowance, the No Filter mark, the content policy setting, API keys.
9. Settings and troubleshooting
Every settings tab, the log file, the exact engine and trainer errors with fixes, Linux notes, storage paths, reset.
10. FAQ and glossary
The questions from the help channels, and the words the app uses.
If you only read one page
Install the app (chapter 2), let the wizard set up the LU Engine and download the starter model (chapter 3), and send your first message (chapter 4). That is a working local AI on most machines with a graphics card that has 6 GB of memory or more. Everything else in the handbook is optional and can wait until you want it.
Local by default, with optional network features
The app runs locally by default. Chat, agent runs and image and video generation on a local model happen on your machine, and the app has no analytics kit and sends no crash reports. Model downloads, update checks against GitHub Releases, and the count that is sent on each press of the Cloud switch still use the network. While you are signed in to LU Cloud, the app also checks your account every five minutes, in local mode too. Hosted requests leave your machine and are sent to the selected service, including any content you attach to that request. Remote access can also expose app content to a device you paired. Review the selected provider and the tool permissions before sharing private material.
For what the hosted service does and does not do, read the cloud overview. Local chat needs a downloaded model, and document chat also needs an embedding model. Image and video generation need ComfyUI and additional model downloads, which the app installs for you. None of this is a promise that every computer performs the same way.
Before you install: the security warnings
The Windows installer is not yet signed with a Microsoft code signing certificate, so Windows SmartScreen shows an "unrecognized app" warning on first run, and some antivirus engines flag the file with a heuristic. Chapter 2 explains what is behind that. The short version: confirm that the file came from the project's official Releases page and matches the exact version you meant to download. Source availability alone does not verify a downloaded executable. Stop if the source or checksum does not match.
If your antivirus flags the app
- Keep protection enabled and leave the file blocked or quarantined while you check. Do not add a blanket folder exclusion.
- Look up the detection name and the affected file in your security product's history, update its definitions and scan again with its normal controls.
- A false positive is possible on this installer, and it must be investigated rather than assumed. Report the app version, the release file name and the detection name in the project's issue tracker. Do not post private files, credentials or unredacted logs.
- Microsoft's own guidance is at Virus and threat protection in the Windows Security app. Exclusions reduce protection and are not a normal installation step.
Where to get help
Most problems are covered in chapter 9. When you ask for help in the Discord or in a GitHub issue, say which surface you are on (the desktop app or the browser at lu-labs.ai), which version, which operating system, which graphics card, and attach today's log file from Settings, General, Troubleshoot. The log holds what the app was doing right before the problem, which a screenshot cannot show.