LM Studio is usually the easiest beginner path if you want to run local AI through a desktop app instead of starting with terminal commands. Install the app, download a model that actually fits your computer, start a local chat, and only then explore document chat, the local server, or MCP tools.
The main beginner mistake is not the installation itself. It is downloading a model that is too large for your RAM, VRAM, or Mac unified memory.
Quick recommendation
Use LM Studio if you want a desktop app for finding, downloading, and chatting with models. Use Ollama instead if you want a lighter command-line runtime, an API backend, or a common base for Open WebUI.
Start with these steps:
- Confirm your computer is supported.
- Install LM Studio.
- Download a small model first.
- Start a local chat.
- Only then try larger models, document chat, or the local server.
Before you install LM Studio
LM Studio is easier than many local AI tools, but it still depends on your hardware. A desktop app cannot make a large model fit inside a small machine.
| Question | Beginner answer |
|---|---|
| Is LM Studio beginner-friendly? | Yes. It is a desktop app for discovering, downloading, and chatting with local models. |
| Does LM Studio run models locally? | Yes, core local chat workflows can run locally once the required files are on your machine. |
| Does LM Studio work offline? | Core local functions can work offline after models and required runtimes are already downloaded. Model search, model downloads, runtime downloads, and updates require internet. |
| Can I use it on an Intel Mac? | No. Current LM Studio Mac requirements specify Apple Silicon. Ollama can run on an Intel Mac using the CPU if macOS meets its requirements. |
| Can I use it with 8GB RAM? | Only for small models and modest expectations. 16GB or more is the safer beginner baseline. |
| Do I need the local server? | No for basic chatting. The local server is for apps that need to talk to your local model through an API. |
LM Studio system requirements in plain English
Check the current official system requirements before installing. Start by checking your operating system, memory, and processor.
| Platform | Beginner requirement notes |
|---|---|
| Mac | Apple Silicon and macOS 14 or newer. Intel Macs are unsupported. |
| Windows | Windows x64 and ARM are supported, but x64 systems require AVX2. 16GB+ RAM and dedicated VRAM are recommended for a better experience. |
| Linux | The AppImage targets Ubuntu 20.04 or newer, with x64 AVX2 or ARM64 builds. Other distributions can need additional AppImage dependencies. |
| RAM | 16GB or more is the practical beginner target. 8GB users should stay with small models. |
| VRAM | Dedicated VRAM matters on Windows. Shared GPU memory should not be treated like dedicated VRAM. |
| Storage | Model files are separate downloads and can consume many gigabytes. |
Pick your first model before you download anything large
In LM Studio, the model browser can make it tempting to grab a huge model immediately. Do not do that. Start with a model that fits.
| Your machine | Safer first model choice | Avoid at first |
|---|---|---|
| 8GB system RAM | Start with a 1B/2B-class model and modest context; check larger files before downloading | 13B/14B+ models, long context, multitasking |
| 16GB system RAM | 7B/8B-class model in a compact quantization; small models for speed | Assuming every larger model will feel smooth |
| 32GB system RAM | 7B/8B comfortably; some 14B-class models may be worth testing | Treating 70B-class models as beginner-friendly |
| 16GB dedicated VRAM | Stronger local model options, depending on model and context | Confusing VRAM with system RAM |
| 16GB Apple unified memory | Good beginner tier, but memory is shared with macOS and other apps | Leaving many memory-heavy apps open |
| CPU-only or weak GPU | Small models only | Expecting cloud-chatbot speed |
What do Q4, Q5, Q6, and Q8 mean?
Many local models in LM Studio are distributed as GGUF files with quantization labels such as Q4, Q5, Q6, or Q8. In practical beginner terms:
- Q4 usually means smaller file size and lower memory use, often at some quality cost.
- Q5 is a common balance point when your machine has enough memory.
- Q6/Q8 can use more memory and may not be the right first download for a low-RAM machine.
- GGUF is the file format; Q4/Q5/Q6/Q8 are quantization choices inside that ecosystem.
Your goal on day one is not to find the “perfect” model. Your goal is to find a model that loads, responds, and does not freeze your machine.
Install LM Studio on Mac
1. Confirm that your Mac is supported
LM Studio's current macOS path should be checked against official system requirements before installing. This page treats Apple Silicon Macs as the supported Mac path.
If you have an Intel Mac, do not assume LM Studio will work. Consider Ollama’s Mac path or another local AI tool instead.
2. Download LM Studio
Download LM Studio from the official download page.
3. Install and open the app
Open the downloaded file and complete the normal macOS app installation flow. Launch LM Studio.
4. Open the model discovery area
Inside LM Studio, use the model discovery/search area to find a small first model.
5. Choose a small model first
For an 8GB Mac, choose a small model. For a 16GB Apple Silicon Mac, a 7B/8B-class model may be a practical starting point, but a small model is still safer for your first attempt.
6. Start your first local chat
After the model downloads, load it and ask a simple prompt:
Explain local AI in one sentence.
If the app responds, your basic LM Studio install works.
Install LM Studio on Windows
1. Confirm your CPU and RAM
LM Studio documents Windows x64 and ARM support, and x64 systems require AVX2. If you are using an older x64 CPU without AVX2, LM Studio may not run.
A safer beginner Windows machine has:
- 16GB or more of system RAM;
- a modern x64 or ARM Windows machine;
- dedicated GPU VRAM if you want better performance;
- enough free disk space for downloaded model files.
2. Download the Windows installer
Download LM Studio from the official download page.
3. Install and launch LM Studio
Complete the normal Windows installation flow and open the app.
4. Download a model that fits your machine
If you have 8GB or 16GB of system RAM, do not pick a large model first. Pick a small model or a compact 7B/8B-class model based on your RAM and VRAM.
5. Start a local chat
Load the model and ask a short prompt.
If the app responds slowly, do not assume the install failed. You may simply have chosen a model that is too large for your hardware.
Install LM Studio on Linux
1. Download the Linux build
LM Studio documents Linux support, including AppImage-based installation. Download the current Linux build from LM Studio and check the official docs for any distribution-specific notes.
2. Make the AppImage executable if needed
On many Linux systems, you may need to mark the AppImage as executable before launching it.
Example:
chmod +x "LM-Studio-VERSION.AppImage"
./LM-Studio-VERSION.AppImage
Replace LM-Studio-VERSION.AppImage in both commands with the exact downloaded filename, including its version and architecture. You can also use your file manager's properties to permit execution. If it reports missing FUSE or library support, follow the official AppImage notes for your distribution.
3. Launch LM Studio
Open the app and confirm that it reaches the main interface.
4. Download a small first model
As with Mac and Windows, start small. Linux users are often more technical, but the same RAM and VRAM constraints still apply.
How to know LM Studio installed correctly
Use these checks in order.
| Check | Success looks like |
|---|---|
| App launches | LM Studio opens to the main interface |
| Model discovery works | You can search for models while online |
| Model downloads | The selected model finishes downloading |
| Model loads | The app loads the selected model without crashing |
| First chat works | A short prompt gets a response |
| Offline behavior works | After files are downloaded, core local chat still works without internet |
| Local server works, if needed | API requests reach the local LM Studio server |
Where LM Studio stores models
LM Studio officially documents model storage under a path like:
~/.lmstudio/models/<publisher>/<model>/<model-file.gguf>
Treat this as the model-storage path, not necessarily the full app-data story. Chats, attached documents, and settings may be stored in other folders. Check those locations too if you need to back up or remove your data.
Use the app's My Models location to confirm your actual folder. See moving model storage for a checked copy-and-reopen procedure; model folders alone do not back up chats or attached documents.
Do you need LM Studio’s local server?
No, not for basic chatting.
LM Studio’s local server is useful when another app needs to call your local model through an API. For example, a developer tool, script, or local workflow may expect an OpenAI-style endpoint.
For beginners:
- Skip the local server on day one.
- First confirm the app launches.
- Download a model.
- Start a normal local chat.
- Only then turn on the local server if another guide specifically requires it.
LM Studio documents a desktop interface app, lms command line, headless llmster, a native REST API, and OpenAI- and Anthropic-compatible endpoints.
Optional: start the local server
If you need the local server, use LM Studio’s Developer tab or the documented command line path.
A common documented pattern is:
lms server start
Then connect your local app to the endpoint shown by LM Studio.
Beginner warning: Do not expose a local server to your network or the public internet unless you understand the security implications. “Local” is safer only when the local service stays local and is configured carefully.
Common LM Studio install problems
If the app opens but a downloaded model will not load, use the LM Studio loading checks to separate file format, runtime, incomplete downloads, and memory failures.
| Problem | Likely cause | Fix |
|---|---|---|
| LM Studio will not run on Mac | Intel Mac or unsupported macOS version | Check current system requirements; use Apple Silicon Mac or consider Ollama |
| Windows app does not launch | x64 CPU may lack AVX2 or system may be unsupported | Check CPU support and official requirements |
| Linux AppImage will not open | File may not be executable or dependencies may vary by distro | Mark executable and check official Linux notes |
| Model download is huge | Model files are separate from the app | Start with a smaller model |
| Model loads but responses are slow | Model too large, insufficient RAM/VRAM, or CPU fallback | Try a smaller or more compact model |
| App opens but there are no models | You have not downloaded a model yet | Use model discovery while online |
| Offline mode “does not work” | Required model/runtime files were not downloaded before going offline | Download model and required runtime while online first |
lms command does not work | command line may not be initialized or app may need to run first | Launch LM Studio and check command line docs |
| Local server does not respond | Server not started, wrong port, or app not running | Start the server from LM Studio and use the displayed endpoint |
| User expects ChatGPT-level quality | Smaller local models may be less capable than frontier cloud models | Use realistic model-size expectations |
Privacy notes for LM Studio
LM Studio is a strong beginner option for local AI, but privacy still depends on the full workflow.
Based on the reviewed LM Studio documentation:
- Core local chat functions can run locally once required files are downloaded.
- Local document chat can be offline-capable once the needed files are on the machine.
- Model search, model downloads, runtime downloads, and update checks require internet.
- Connecting external APIs, remote tools, remote MCP servers, or other networked integrations changes the privacy profile.
- Check where the app saves chats, document copies, and settings before backing up or deleting your data.
Use this rule:
LM Studio can be a private local AI setup, but only if the selected model, documents, embeddings, tools, and server connections stay local.
If you are working with sensitive documents, client files, legal files, medical information, financial records, or confidential business material, test the workflow with non-sensitive files first and verify exactly which providers and tools are connected.
LM Studio vs Ollama after install
| Use case | Better first choice | Why |
|---|---|---|
| “I want a desktop chat app.” | LM Studio | desktop-app model discovery and chat |
| “I want the easiest first model download.” | LM Studio | Built around browsing and downloading models |
| “I want a backend for Open WebUI.” | Ollama | Common local runtime/backend path |
| “I want terminal and API workflows.” | Ollama | Lightweight command line and local API focus |
| “I want to attach documents in the app.” | LM Studio | Document chat is part of the app workflow |
| “I want a browser-based ChatGPT-like interface.” | Ollama + Open WebUI | Ollama can power a separate interface |
What to do after installation
After LM Studio is installed and your first small model works, your next step depends on your goal.
| Goal | Next step |
|---|---|
| Choose between tools | Read Ollama vs LM Studio |
| Pick a setup for your Mac | Read Best Local AI Setup for Mac |
| Pick a setup for Windows | Read Best Local AI Setup for Windows |
| Understand low-memory limits | Read Best Local AI for 8GB RAM |
| Use local documents | Read Chat With PDFs Locally |
| Understand privacy | Read Is Local AI Actually Private? |
| Prefer a terminal/API path | Read How to Install Ollama |
What to read next
- What Is Local AI?
- Ollama vs LM Studio
- How to Install Ollama
- Best Local AI Setup for Mac
- Best Local AI Setup for Windows
- Chat With PDFs Locally
- Is Local AI Actually Private?
Sources
Sources reviewed for this page:
- LM Studio documentation: https://lmstudio.ai/docs/app
- LM Studio system requirements: https://lmstudio.ai/docs/app/system-requirements
- LM Studio desktop app privacy policy: https://lmstudio.ai/app-privacy