LM Studio is a free desktop app that runs AI models entirely on your own laptop, so prompts and documents never leave the device. This guide covers the latest version and how to check it, whether LM Studio requires admin rights to install on Windows (and why there is no official portable no-admin build), which model suits your RAM, and how to start chatting in about ten minutes.
Key takeaways
- LM Studio runs open-weight models locally with a point-and-click interface. No command line needed.
- On Windows, a per-user install generally does not need administrator rights, but company device policies can still block it.
- We are not aware of an official portable Windows build. Avoid unofficial repackaged versions on work devices.
- 16 GB of RAM is the practical starting point for useful 7 to 8 billion parameter models.
- LM Studio has been free for work use since July 2025. Check your organisation's AI policy before installing.
Why is LM Studio the easiest starting point for local AI?
There are several tools for running AI locally, including Ollama, Jan, GPT4All and LM Studio, covered in our guide to running AI offline at work. If you have never used a command line and want something that feels like ChatGPT but runs on your laptop, start with LM Studio.
Our team sees the same pattern with non-technical staff: Ollama's terminal stops them, while LM Studio has them chatting with a local model within minutes. It is a purpose-built desktop app where you search for models, download them and chat from one window.
- Section 1Install and versionsWindows, Mac and Linux, admin rights, latest version
- Section 2Interface tourWhat each tab does
- Section 3Your first modelPick the right model for your hardware
- Section 4First conversationLoad a model and start chatting
- Section 5Advanced featuresDocuments, local server and corporate use
What is the latest version of LM Studio?
In January 2026 the current release was LM Studio 0.3.36, which added bug fixes and support for newer models. LM Studio releases updates frequently, so the version number changes every few weeks.
To check the latest version:
- Open LM Studio and check for updates from the app menu. The app prompts you when a new version is available.
- Visit lmstudio.ai and its changelog or blog for release notes.
On managed work devices, IT may control updates. If your version is behind, ask IT before updating, because newer releases can change model formats and features.
What hardware does LM Studio need?
| Minimum | Recommended | Impact | |
|---|---|---|---|
| RAM | 8 GB (very limited) | 16 GB or more | For 7-8B models |
| Storage | 10 GB free | 50 GB+ free | Models are roughly 2-40 GB each |
| CPU | Modern CPU with AVX2 (Windows/Linux) | Apple Silicon or recent Intel/AMD | Faster responses |
| GPU | Not required | Dedicated GPU with 6 GB+ VRAM | Much faster |
Mac: current builds run on Apple Silicon (M1 or later) with a recent macOS version. Unified memory lets a 16 GB Apple Silicon laptop run 7 to 8 billion parameter models comfortably.
Windows: Windows 10 or 11 with an AVX2-capable CPU (almost all processors from the past decade) and ideally 16 GB RAM. An NVIDIA GPU helps but is not required.
Linux: distributed as an AppImage for x64 (and ARM64 builds), with the same RAM guidance as Windows.
Check lmstudio.ai for current system requirements, as they change between releases.
How do you install LM Studio?
- Go to lmstudio.ai. The site detects your operating system.
- Download the installer (a few hundred megabytes).
- Install:
- Mac: open the .dmg, drag LM Studio to Applications, then open it.
- Windows: run the .exe and follow the installer (see the admin rights section below).
- Linux: make the AppImage executable with
chmod +x LM-Studio-x.x.x.AppImage, then run it.
- On first launch, let the app update if prompted.
Does LM Studio require admin rights to install on Windows?
Usually not, if you install for your own account. Like many desktop apps built on the same framework, the LM Studio Windows installer can install per user into your user profile, which generally does not need administrator rights. Installing for all users of the PC does need admin rights and triggers a Windows UAC prompt.
Even without a UAC prompt, a work laptop may still block the install. Common reasons:
- Application control such as Microsoft Defender Application Control, AppLocker or Intune policies that only allow approved software.
- Endpoint protection that flags or quarantines new executables.
- Network controls that block model downloads from Hugging Face.
If the install is blocked, ask IT. Trying to work around the controls usually breaches your organisation's acceptable use policy.
Is there a portable LM Studio for Windows with no admin rights?
We are not aware of an official portable LM Studio build for Windows. Repackaged "portable" versions from third-party sites are a security risk because you cannot verify what has been changed. On a work device, use the official installer with IT's approval. If IT wants local AI centrally managed, a server-based option is often a better fit (see the corporate section below).
How is the LM Studio interface laid out?
- Left sidebarNavigation between tabs
- Main workspaceWhere you chat and configure
- Top barModel selector and settings
- Bottom inputType your messages here
| Tab | What it does |
|---|---|
| Chat | ChatGPT-style conversation window |
| Discover | Search and download models |
| Developer / Local Server | Run LM Studio as a local API server |
| My Models | Manage downloaded models |
Discover is the model store: search for names such as "llama", "qwen", "gemma" or "mistral", and each model card shows size and whether it fits your hardware.
Chat has a model selector at the top, conversation history on the left, the chat in the centre and a collapsible settings panel on the right.
My Models shows file sizes and locations, and lets you delete models you no longer need.
Which model should you download first in LM Studio?
How much RAM does your laptop have?
- 8 GB RAMA 3-4B model such as Phi-4-mini, Qwen 2.5 3B or a small Gemma variant
- 16 GB RAMA 7-8B model such as Llama 3.1 8B Instruct, Qwen 2.5 7B or Mistral 7B
- 32 GB+ RAMA 14-32B model such as Qwen 2.5 14B or Mistral Small
With 16 GB RAM, Llama 3.1 8B Instruct is a solid first download: about a 5 GB file in Q4 quantisation, good at drafting, summarising and Q&A, with a large support community. Newer families such as Qwen 3 and Gemma 3 are also worth trying as they appear in Discover.
To download: open Discover, search the model name, pick a quantised version (Q4_K_M is a good default) and click Download.
Reading model names:
llama-3.1-8b-instruct-q4_k_m.gguf
│ │ │ │ │
│ │ │ │ └── Quantisation (smaller file, slightly less accurate)
│ │ │ └── Tuned for instructions and conversation
│ │ └── 8 billion parameters
│ └── Version
└── Model familyQuantisation guide (8B model):
- Q4_K_M: around 5 GB, good balance for most users.
- Q5_K_M: a little larger, slightly better quality.
- Q8_0: around 8-9 GB, highest quality quantised.
- F16: around 16 GB, full precision for high-end hardware.
How do you start your first conversation?
- Open the Chat tab.
- Choose your downloaded model in the selector and wait 10 to 60 seconds while it loads into memory.
- Type a prompt such as "Summarise the key risks in a supplier contract in three bullet points" and press Enter.
- Ask follow-ups. LM Studio keeps conversation context within the chat.
On a 16 GB Apple Silicon laptop, an 8B model typically responds fast enough to feel interactive.
Which LM Studio settings matter?
| Setting | What it does | Suggested value |
|---|---|---|
| Temperature | Randomness: lower is more focused | 0.3 for factual tasks, 0.7 for general writing |
| Max tokens | Maximum response length | 2048 for most tasks |
| Context length | How much conversation the model considers | 4096-8192, hardware dependent |
| GPU offload | How much of the model runs on the GPU | Auto, or increase for speed |
System prompt example for work:
You are a professional business assistant. Respond in a concise,
formal tone suitable for corporate communication. Use Australian
English spelling.Can LM Studio read your documents?
Yes. Attach PDF, DOCX or TXT files in the chat and ask questions, for example "List the payment terms in this contract" or "Summarise the action items from these meeting notes".
Limits to know:
- Long documents may be cut down to fit the model's context window.
- Local models are less accurate than large cloud models on complex analysis.
- Tables and complex layouts may not parse cleanly.
How do you use the LM Studio local server?
LM Studio can run as a local API server compatible with the OpenAI API format, so tools built for OpenAI can point at your local model instead.
- Open the Developer or Local Server tab.
- Select a model and start the server.
- Point applications at
http://localhost:1234/v1with any placeholder API key.
Uses include VS Code extensions, note-taking apps and internal scripts, with no usage fees or rate limits.
How do you fix common LM Studio problems?
Model fails to load or crashes: not enough memory. Close other apps, choose a smaller or more quantised model, reduce context length, or lower GPU offload.
Very slow responses: the model is running on CPU or memory is tight. Increase GPU offload if you have VRAM, use Q4_K_M, and close background apps.
Download fails or stalls: retry, try another quantisation, or check whether your network blocks Hugging Face.
Gibberish or repetition: use an "instruct" version, lower temperature to 0.5-0.7, or try a different model.
Installer blocked on Windows: see the admin rights section above and contact IT.
Is LM Studio suitable for corporate use?
For individuals, often yes, once IT approves it:
- After models are downloaded, LM Studio works offline.
- Chats stay on your device.
- No account is needed for local use.
- LM Studio has been free for work use since July 2025.
Good uses: drafting emails and documents, summarising notes, explaining technical concepts, code explanation.
Use an approved cloud or server AI instead for: very long document analysis, tasks needing current information, complex multi-step reasoning, and anything where accuracy is critical without review.
Before installing at work, check your organisation's AI acceptable use policy. Our guide on staff leaking business data into ChatGPT explains why approved tools matter.
When should an organisation move beyond LM Studio on laptops?
LM Studio suits individuals. Rolling local AI out to 50 or 500 staff raises different questions: who approves models, how hardware is specified, how usage is logged, and how AI connects to your documents and systems. At that point a central deployment on your own servers or private cloud in Australia, managed by IT, is usually easier to govern than hundreds of laptop installs. Compare options in OpenAI vs Claude vs Ollama and why everyone should run Ollama locally.
How Solve8 can help
When local AI needs to work across a whole organisation, our engineers design and deploy private AI infrastructure that runs on your infrastructure in Australia, and our custom AI development team connects it to your documents and systems. Our consultants help with model choice, policy and rollout through our AI strategy service. Talk to us about local AI for your teams.
Sources: LM Studio; LM Studio documentation; LM Studio 0.3.36 release notes.
