Verdict
Official documentation reviewed, with caveats
Evidence label: Official documentation reviewed, with caveats. Sources were reviewed on 2026-05-24. Local AI Guide test status: Not independently tested by Local AI Guide. This page does not contain local benchmark, install, privacy-audit, network-monitoring, storage-inspection, or screenshot evidence.
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 GUI-first local AI 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.
For most beginners, the safest LM Studio setup is:
- 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 | Evidence label |
|---|---|---|
| Is LM Studio beginner-friendly? | Yes. It is a GUI-first app for discovering, downloading, and chatting with local models. | Official documentation reviewed |
| Does LM Studio run models locally? | Yes, core local chat workflows can run locally once the required files are on your machine. | Official documentation reviewed |
| 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. | Official documentation reviewed |
| Can I use it on an Intel Mac? | The research packet flags current LM Studio Mac support as Apple Silicon only. Verify this against current system requirements before publication. | Official documentation reviewed |
| Can I use it with 8GB RAM? | Only for small models and modest expectations. 16GB or more is the safer beginner baseline. | Official documentation reviewed + Conservative estimate, not a benchmark |
| 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. | Official documentation reviewed + Conservative estimate, not a benchmark |
LM Studio system requirements in plain English
Check the current official system requirements before publishing or installing. Based on the research packet, these are the practical requirements to explain to beginners.
| Platform | Beginner requirement notes | Evidence label |
|---|---|---|
| Mac | Apple Silicon Mac and macOS 14 or newer are the safe current assumptions from the research packet. Intel Macs are flagged as unsupported. | Official documentation reviewed |
| Windows | Windows x64 and ARM are supported, but x64 systems require AVX2. 16GB+ RAM and dedicated VRAM are recommended for a better experience. | Official documentation reviewed |
| Linux | LM Studio provides Linux support, including AppImage-based installation. Linux setup may need executable permissions or distribution-specific troubleshooting. | Official documentation reviewed + Conservative estimate, not a benchmark |
| RAM | 16GB or more is the practical beginner target. 8GB users should stay with small models. | Official documentation reviewed + Conservative estimate, not a benchmark |
| VRAM | Dedicated VRAM matters on Windows. Shared GPU memory should not be treated like dedicated VRAM. | Official documentation reviewed + Conservative estimate, not a benchmark |
| Storage | Model files are separate downloads and can consume many gigabytes. | Conservative estimate, not a benchmark |
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 | Evidence label |
|---|---|---|---|
| 8GB system RAM | Small 3B/4B-class model, lower quantization, modest context | 13B/14B+ models, long context, multitasking | Conservative estimate, not a benchmark |
| 16GB system RAM | 7B/8B-class model in a compact quantization; small models for speed | Assuming every larger model will feel smooth | Conservative estimate, not a benchmark |
| 32GB system RAM | 7B/8B comfortably; some 14B-class models may be worth testing | Treating 70B-class models as beginner-friendly | Conservative estimate, not a benchmark |
| 16GB dedicated VRAM | Stronger local model options, depending on model and context | Confusing VRAM with system RAM | Conservative estimate, not a benchmark |
| 16GB Apple unified memory | Good beginner tier, but memory is shared with macOS and other apps | Leaving many memory-heavy apps open | Conservative estimate, not a benchmark |
| CPU-only or weak GPU | Small models only | Expecting cloud-chatbot speed | Conservative estimate, not a benchmark |
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
Based on the research packet, LM Studio’s current macOS path is for Apple Silicon Macs. Check the official LM Studio system requirements before publishing and before installing.
Evidence label: Official documentation reviewed.
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 site.
Evidence label: Official documentation reviewed.
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 first-run verification.
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
Based on the research packet, LM Studio supports Windows x64 and ARM, and x64 systems require AVX2. If you are using an older x64 CPU without AVX2, LM Studio may not run.
Evidence label: Official documentation reviewed.
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 site.
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’s research packet identifies Linux AppImage support. Download the current Linux build from LM Studio and check the official docs for any distribution-specific notes.
Evidence label: Official documentation reviewed.
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*.AppImageThen run it from your file manager or terminal.
Evidence label: Conservative estimate, not a benchmark.
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 | Evidence label |
|---|---|---|
| App launches | LM Studio opens to the main interface | Official documentation reviewed |
| Model discovery works | You can search for models while online | Official documentation reviewed |
| Model downloads | The selected model finishes downloading | Official documentation reviewed |
| Model loads | The app loads the selected model without crashing | Conservative estimate, not a benchmark |
| First chat works | A short prompt gets a response | Conservative estimate, not a benchmark |
| Offline behavior works | After files are downloaded, core local chat still works without internet | Official documentation reviewed |
| Local server works, if needed | API requests reach the local LM Studio server | Official documentation reviewed |
Where LM Studio stores models
The research packet says 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. The research packet flags chat history, attached-document storage, prompt storage, and configuration-file locations as areas that should be verified before making stronger claims.
Evidence label: Official documentation reviewed for model path; unclear for chat/document storage paths.
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.
The research packet says LM Studio includes a GUI app, lms CLI, headless llmster, a native REST API, and OpenAI- and Anthropic-compatible endpoints.
Evidence label: Official documentation reviewed.
Optional: start the local server
If you need the local server, use LM Studio’s Developer tab or the documented CLI path.
A common documented pattern is:
lms server startThen 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
| Problem | Likely cause | Fix | Evidence label |
|---|---|---|---|
| LM Studio will not run on Mac | Intel Mac or unsupported macOS version | Check current system requirements; use Apple Silicon Mac or consider Ollama | Official documentation reviewed |
| Windows app does not launch | x64 CPU may lack AVX2 or system may be unsupported | Check CPU support and official requirements | Official documentation reviewed |
| Linux AppImage will not open | File may not be executable or dependencies may vary by distro | Mark executable and check official Linux notes | Conservative estimate, not a benchmark |
| Model download is huge | Model files are separate from the app | Start with a smaller model | Conservative estimate, not a benchmark |
| Model loads but responses are slow | Model too large, insufficient RAM/VRAM, or CPU fallback | Try a smaller or more compact model | Conservative estimate, not a benchmark |
| App opens but there are no models | You have not downloaded a model yet | Use model discovery while online | Official documentation reviewed |
| Offline mode “does not work” | Required model/runtime files were not downloaded before going offline | Download model and required runtime while online first | Official documentation reviewed |
lms command does not work | CLI may not be initialized or app may need to run first | Launch LM Studio and check CLI docs | Official documentation reviewed + Conservative estimate, not a benchmark |
| 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 | Official documentation reviewed + Conservative estimate, not a benchmark |
| User expects ChatGPT-level quality | Smaller local models may be less capable than frontier cloud models | Use realistic model-size expectations | Conservative estimate, not a benchmark |
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 research packet:
- 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.
- The model-storage path is documented, but the research packet flags some chat/document/config storage details as requiring additional verification before making stronger claims.
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 | GUI-first 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 CLI 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 UI.” | Ollama + Open WebUI | Ollama can power a separate UI |
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 and evidence
Official and research sources used for this draft:
- 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
- Local AI Stack keyword map and SERP research packet
- Local AI Stack source-of-truth research packet
- Local AI Stack compatibility foundation
- Local AI Stack privacy and security research packet
- Local AI Stack repeatable testing protocol