Is there a Mac version of the Locally Uncensored desktop app?

No. The free open source desktop app is Windows and Linux only. There is no native Mac build and we do not want to imply otherwise. On a Mac the answer is the browser Studio at lu-labs.ai, used next to whatever local runner you already have installed. The desktop app's Cloud switch and the browser Studio use the same LU Labs account, so a Windows machine and a Mac in the same household share one balance.

What is the best LM Studio alternative on a Mac?

It depends on what LM Studio is failing to do for you. If you want a different local chat client, Jan and Ollama with a front end are the closest swaps and stay fully offline. If you hit the ceiling of your unified memory or you want images and video, a local client is not the answer at all, and the browser Studio at lu-labs.ai covers chat with 38 hosted models, an image lane with 10 models and a video lane with 5.

Do I need to install anything to run the browser Studio on a Mac?

No. It runs in Safari or Chrome. No Homebrew, no Xcode command line tools, no Python environment, no model download. If you want it to feel like an app, add the page to the Dock: File then Add to Dock in Safari, or the three dot menu then Cast Save and Share then Install page as app in Chrome.

Can I use it from the Mac terminal or from SillyTavern?

Yes. Create a key under Settings then Cloud API keys. The key has an lu_ prefix and is shown once. The base URL is https://lu-labs.ai/api/inference/v1 and the shape is OpenAI chat completions, so curl works, and so do SillyTavern in Chat Completion mode as a custom OpenAI compatible source, Aider, and LibreChat. Useful model ids are moonshotai/Kimi-K3 and zai-org/GLM-5.3-Flash.

Can LM Studio generate images on a Mac?

No. LM Studio is a text app. It runs language models and serves them over a local OpenAI compatible endpoint, and that is the whole scope. For diffusion work on a Mac you either install a separate stack such as Draw Things or ComfyUI and accept slow generation, or you send the prompt to a rented GPU. The hosted image lane runs on managed NVIDIA H100, A100 and B200 class hardware.

What does the hosted layer cost?

A 5 EUR credit pack is 165,000 credits with no subscription, and credits never expire. That is roughly 550 Flux Schnell images at 300 credits each, or about 137 Flux 2 Dev images at 1,200 credits each, or roughly 3.3 million output tokens on GLM 5.3 Flash. The Hosted plan is 19 EUR a month with 900,000 credits and includes 2 LoRA trainings a month. Full list at lu-labs.ai/pricing.

Should I stop using LM Studio if I set this up?

No, and that is the point of the hybrid setup. Keep a small model loaded in LM Studio or Ollama for offline work, drafts and anything you do not want leaving the machine. Send the heavy prompts, the long agent loops and every image to the hosted lane. The two do not conflict, and on a Mac this combination beats either half on its own.

LM Studio Alternative for Mac: The Browser Studio Setup

LM Studio is a good app. If you are on a Mac and you want a tidy local chat client with a model browser bolted on, it is one of the shortest routes there, and nothing below is an argument against keeping it installed. People search for an alternative on a Mac for two reasons that have very little to do with the app itself: they run into the memory ceiling of the machine, or they want a picture and a text client does not make pictures.

This guide is the setup we recommend on Apple hardware: the browser Studio at LU Labs next to whatever local runner you already use. For a straight comparison of desktop clients, see the seven LM Studio alternatives roundup.

What LM Studio Already Does Well

Credit where it is due. On a Mac, LM Studio gives you a clean chat window, a model browser that surfaces both GGUF and MLX builds with the quant sizes visible, and a local server speaking the OpenAI chat completions shape on localhost. Everything stays on the machine. Pull the network cable and it keeps answering. For a laptop that mostly needs a private assistant for text, that is a complete product and you do not need us.

The Two Things It Does Not Do

It does not generate images or video. LM Studio runs language models, full stop. Diffusion on a Mac means installing a separate stack such as Draw Things or ComfyUI, and then waiting, because Apple Silicon is not built for that workload.

It cannot run a model larger than your machine. On Apple Silicon the model weights live in unified memory shared with macOS, your browser and everything else. A 16 GB MacBook Air realistically tops out around a 7B to 12B model at 4-bit before the system starts swapping. There is no local path to a 405B model on a laptop, at any quant, ever. We wrote the honest tier list in what a Mac can actually run locally in 2026.

What the Browser Studio Gives You Instead

Lane	What is in it
Chat	38 hosted models on every plan and on a credit pack, including Kimi K3, GLM 5.3 Flash (vision plus a 1M context), DeepSeek V3.2, Gemma 4 26B, gpt-oss 120B, Euryale 70B, Lunaris 8B and MythoMax 13B
Image	10 models: Flux 2 Dev, Flux Dev, Flux Schnell, Qwen Image, HiDream, HunyuanImage 2.1, Z-Image Turbo, Chroma, Prefect Pony XL, Neta Lumina
Editing	Inpainting on Flux Dev, Qwen Image Edit, background removal, eraser, upscale to 2k, 4k or 8k
Video	5 models: Wan 2.2 720p, Wan 2.2 Fast, LTX-2 with audio, LTX 2.3, HunyuanVideo 1.5. Clips are 5 or 8 seconds
Hardware	Managed NVIDIA H100, A100 and B200 class GPUs. Shared, so expect a short queue at busy hours

On filtering, the honest version: hosted text models answer the way they were trained and we add no moderation layer on top of them. Image and video prompts follow the Terms of Service and are checked before anything renders. Content involving minors is refused everywhere, always. The local desktop app runs whatever you install with nothing in the middle, which is exactly why the hybrid below makes sense.

Setup on a Mac, Step by Step

Open the Studio. Go to lu-labs.ai in Safari or Chrome. Nothing to install. No Homebrew formula, no Xcode command line tools, no 20 GB download sitting in your Downloads folder.
Sign in. Accounts and billing are in the EU. Prompts are not used for training and are never sold.
Pick a model. GLM 5.3 Flash is the sensible first pick: it takes images as input and holds a 1M token context. If you want the largest thing in the catalog, Kimi K3 needs exactly zero gigabytes of your RAM.
Send the first message. Usage is metered per token in credits, so a short chat costs a fraction of a cent rather than a subscription.
Add it to the Dock, if you want it to feel like an app. Safari: File, then Add to Dock. Chrome: the three dot menu, then Cast Save and Share, then Install page as app. You get an icon and a window with no tab strip.
Switch lanes. The lane switcher at the top moves you from chat to image or video. This is the part a local text client cannot cover at all.

The API Key Path, So It Works From the Terminal Too

The browser is not the only door. Open Settings, then Cloud API keys, and create a key. It carries an lu_ prefix and is shown once, so copy it before you close the dialog. The endpoint is OpenAI chat completions shaped, at base URL https://lu-labs.ai/api/inference/v1.

curl https://lu-labs.ai/api/inference/v1/chat/completions \
  -H "Authorization: Bearer lu_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moonshotai/Kimi-K3",
    "messages": [{"role": "user", "content": "Summarise this shell script."}]
  }'

The two model ids worth memorising are moonshotai/Kimi-K3 and zai-org/GLM-5.3-Flash. Because the shape is standard, the usual suspects work without a shim: SillyTavern in Chat Completion mode as a custom OpenAI compatible source, Aider, and LibreChat. If you already point one of those at LM Studio's localhost server, you are changing a base URL and a key, nothing else.

The Hybrid Setup, Which Is the Actual Answer on a Mac

Do not pick a side. Keep a small model loaded in the runner you already have, LM Studio or Ollama or MLX, for offline work, drafts, notes, and anything you would rather not send anywhere. Send the heavy prompts, the long agent loops and every single image to the hosted lane. Your fans stay quiet and your memory pressure graph stays green.

The dividing line is concrete. Qwen 3.8 27B at Q4_K_M is 17.1 GB of weights resident, so a 32 GB Mac holds it comfortably and a 16 GB Mac needs the smaller IQ4_XS at 15.7 GB with almost no context left. Our Qwen 3.8 27B local guide has the full quant table. Above that size, hosted is not a compromise, it is the only option that exists.

What It Costs

Option	Credits	Roughly
5 EUR pack, no subscription	165,000	About 550 Flux Schnell images, or 137 Flux 2 Dev images, or 3.3 million output tokens on GLM 5.3 Flash
10 EUR pack	350,000	Same rates, better ratio
25 EUR pack	900,000	Same volume as a Hosted month, without the subscription
Hosted, 19 EUR a month	900,000	About 3,000 Flux Schnell images or 750 Flux 2 Dev images, plus 2 LoRA trainings a month

Credits never expire and any pack is available from the first purchase, so you can start at 5 EUR without committing to anything. Per model rates are listed at https://lu-labs.ai/pricing. Flux Schnell is 300 credits an image, Flux 2 Dev 1,200, a 5 second LTX-2 clip 8,000. LoRA training is Hosted only, not on packs.

One Thing to Be Clear About

The free open source desktop app is Windows and Linux only. There is no native Mac build, and we are not going to pretend otherwise in a post aimed at Mac owners. The desktop app has a Cloud switch that signs into the same LU Labs account, so if you also have a Windows box the balance is shared. On the Mac itself, the browser Studio is the app.