Verdict: a good all-in-one studio if you have a Windows or Linux PC with a strong graphics card. On an ordinary laptop it starts without any setup but answers slowly, and its starter models are ones with the refusals removed. For plain private chat on any computer, LM Studio or Jan is the calmer first choice.
Other good fits
Trying a local model with no setup knowledge: the chat engine is built in and the first-start wizard downloads a model for you.
Asking questions about a document that must stay on your own machine.
+ Pros
Read pros
Locally Uncensored is a desktop app for Windows and Linux that puts a chat window, image and video generation, a coding agent and phone access to your own PC into one program. The chat engine is built in and based on llama.cpp, so there is nothing else to install before the first answer. Images and video run through ComfyUI, which the app can install for you. The app is free and open source under AGPL-3.0. The same maker sells hosted graphics cards as "LU Cloud".
Pick it when
You have a PC with a strong graphics card and want chat, pictures and video in one place.
You want a first local model without any setup: install, click through four steps, type.
You want to ask questions about a file that should not leave your computer.
− Cons
Read cons
On a laptop without a graphics card it is slow. The app's own benchmark measured 2.8 tokens per second, which is about two words a second. The starter model answered an English statistics question correctly, but a short German email came back with grammar mistakes.
Both starter models in the wizard are "abliterated", which means their refusals were removed. Standard models such as Gemma or Granite are in the Models tab under Mainstream. That tab lists everything, including models of more than 200 GB, so use its size filter. The wizard also recommended installing image generation on a laptop that cannot run it.
The Windows installer is not code-signed, so Windows and some antivirus programs warn about it. The project says so itself and publishes a checksum for each release. Ours matched, and Microsoft Defender found nothing.
Pick something else when
You use a Mac, or you want standard models and clear hardware hints: use LM Studio.
You need open-source software under a permissive licence: use Jan.
A whole team should share one assistant with logins: run Open WebUI on a server.
Avoid if
Your laptop has no graphics card and you need quick answers; our test laptop managed 2.8 tokens per second. Use a cloud assistant such as Claude or ChatGPT.
Your lab or company needs models that keep their safety behaviour, or software with a signed installer. The starter models have their refusals removed and the Windows installer is unsigned.
Requirements
Windows 10 or 11, or Linux. There is no Mac desktop build. We ran version 3.0.5 on a Windows 11 laptop with an Intel i7-8650U processor, 16 GB of memory and no separate graphics card. The app took 88 MB of disk and the starter model 4.4 GB. While answering, the model held about 7 GB of memory. The project's own guide says an 8B model sits comfortably on a 6 GB graphics card, and that image generation starts at 6 to 8 GB of graphics memory.
First run
Hands-on review — see the recorded run conditions.
Download the installer from the project's GitHub releases page and install it. Windows may warn about an unknown publisher; the reason is under "Where it falls short".
In the first-start wizard, keep the built-in engine, skip image and video if you have no graphics card, pick one starter model and install the small document model (84 MB).
In Chat, attach a short text file or PDF with the paperclip.
Ask a question whose answer you can look up in the file.
On our laptop the model download took just under 5 minutes, and a two-sentence answer took about 50 seconds.
Integration
Everything listens on your own machine only. While the app answered, we saw no connection from it to the outside. An OpenAI-style local address for other programs exists, but it stays off until you create a token and start it. Phone access is off until you switch it on. The app also finds Ollama, LM Studio and ten other local engines if you already run one. A switch labelled Cloud sends work to the paid hosted service instead.
Check the result
Compare the answer with the file. In our run the app read six lines of lab rules and gave the right due date and the right late penalty. Then ask for something the file does not say. A good answer tells you it is not in there.
Links & evidence
Hands-on review · Oct 10, 2026
Installed and run once for this review on 10 October 2026: version 3.0.5 on a Windows 11 Pro laptop (Intel i7-8650U, 16 GB of memory, no separate graphics card). Run: the first-start wizard, the starter model download, two chat questions, the built-in benchmark, one document question and a read of the settings. Not run: image and video generation, the coding agent, agent mode, phone access, LU Cloud, Linux, or any machine with a graphics card. Platforms, licence, prices and the unsigned installer come from the project's website, GitHub page and security notes, checked the same day.
Locally Uncensored website · Oct 10, 2026 — Features, Windows and Linux only, AGPL-3.0, LU Cloud credit packs from 15 EUR and a hosted plan from 19 EUR a month.
Locally Uncensored GitHub repository · Oct 10, 2026 — Licence AGPL-3.0-only; no Mac desktop build; memory guidance for chat and image models; engines the app detects.
Security notes · Oct 10, 2026 — The Windows installer is not code-signed; known antivirus and SmartScreen warnings; how to verify a download.
Release 3.0.5 · Oct 10, 2026 — Installer file name, size and SHA-256 that the downloaded file was compared with.
Further review notes & alternatives
Costs and operating effort
Prices checked 10 October 2026: the app is free. LU Cloud, the optional hosted service, sells credit packs from 15 EUR and a plan from 19 EUR a month.
You pay with disk space and patience instead. Each model takes several gigabytes; ours used 4.6 GB in the app's data folder. Updates are downloaded automatically by default and installed only when you click. The installer added no autostart entry, no service and no scheduled task.
Alternatives and trade-offs
LM Studio wins on polish, Mac support and hardware guidance. It is closed source and has no image or video generation.
Jan wins when you want open source under Apache 2.0 on Windows, Mac and Linux.
ComfyUI wins when images and video are the main job and you want control over every step.
Locally Uncensored is a free, open-source desktop app for Windows and Linux that runs AI models on your own computer. One program holds a chat window with document questions, image and video generation through ComfyUI, a coding agent and phone access to your own PC. The chat engine is built in, and a first-start wizard downloads a model. The maker also sells hosted graphics cards as LU Cloud.