Your business computers can run powerful AI models without sending a single prompt to the cloud. LM Studio is a free desktop app that lets you download open AI models like Llama, Qwen, DeepSeek, and Phi and chat with them entirely on your own machine — no coding, no API keys, no per-seat subscription. This tutorial shows you how to install LM Studio, download your first model, and put it to work for your business.
What It Is
LM Studio is a desktop application for running large language models locally on Mac, Windows, and Linux. It gives you a simple chat interface on top of open models you download yourself, so your conversations never leave your device. Under the hood it runs on llama.cpp — the standard open-source inference engine — plus Apple’s MLX framework on Apple Silicon Macs, wrapped in LM Studio’s own local runtime.
You download models right inside the app by searching Hugging Face. Popular picks include Meta’s Llama, Alibaba’s Qwen, DeepSeek, Microsoft’s Phi, and OpenAI’s open-weight gpt-oss, plus newer frontier open models like GLM 5.2, Kimi K3, and DeepSeek V4 Pro. As checked on October 19, 2026, LM Studio runs on Apple Silicon Macs, x64/ARM64 Windows PCs, and x64 Linux PCs, and is free to download.
The current version also ships Bionic, LM Studio’s built-in AI agent for work and code. Bionic can create and edit documents (changes auto-save), help with coding and automations, and take actions on your computer when you allow it — all powered by your chosen local model. It also offers real-time voice transcription in multiple languages, processed on-device.
Who It’s For / SMB Use Case
LM Studio fits small businesses that want AI help without cloud subscriptions, API bills, or data leaving the building:
- Private drafting assistant: Draft emails, proposals, and marketing copy with no prompts sent to a third party.
- Offline document Q&A: Ask questions about contracts, manuals, and price lists entirely offline.
- Local coding help: Get code explanations and debugging help without exposing proprietary code to external APIs.
- Internal tools backend: Use server mode to expose an OpenAI-compatible API, so your own scripts call a local model instead of a paid cloud API.
It’s not the right pick if you need maximum model quality for hard reasoning — a local model on a laptop won’t match frontier cloud models on everything. And it’s a single-user desktop app, not a team product (server mode and the headless llmster build cover team scenarios).
Installation
Go to lmstudio.ai, open the Downloads page, and grab the installer for your OS — macOS (Apple Silicon), Windows (x64 or ARM64), or Linux (x64). Install it like any normal app. If you’re unsure your machine qualifies, check the System Requirements page in the docs; a modern computer with a healthy amount of RAM will run smaller models comfortably.
No account, API key, or credit card is needed — the app works fully offline once a model is downloaded.
How to Use LM Studio
Step 1: Download your first model
Use the in-app model search to browse Hugging Face. For a first run, pick a smaller model (often labeled by size, like 7B or 8B) — smaller models load faster and need less RAM. Click download, wait for the weights, then click Load. The docs’ Download an LLM guide covers the options.
Step 2: Start chatting
Open a new chat and type — just like ChatGPT, except everything runs on your machine. Try real work: paste a rough draft and ask for a tighter rewrite, or ask it to summarize meeting notes. There’s no usage meter, so experiment freely.
Step 3: Try Bionic, the built-in agent
Switch to Bionic when you need more than chat. Ask it to draft a document and it creates and edits the file directly, saving changes automatically. It can also help with coding, automations, and computer tasks you approve. The difference: advice versus delegation.
Step 4: Chat with your documents — fully offline
Attach a PDF or document to a chat message and ask questions about it. LM Studio reads the file locally (often called RAG), so you can interrogate contracts, handbooks, or reports without uploading them anywhere. See the Chat with Documents guide for details.
Step 5: Connect tools with MCP
LM Studio supports MCP (Model Context Protocol) servers, which plug external tools and data sources into your local model — web search, databases, file systems, and more. Install an MCP server from the app and your local model can use it during chats, extending what it can do with zero cloud dependency.
Step 6: Turn on server mode for your own apps
This is the power move for businesses. LM Studio can serve your loaded model on an OpenAI-compatible local endpoint — on your machine or across your office network. Any script, app, or integration that talks to the OpenAI API can be pointed at your local server instead, swapping a paid API for your own hardware. The docs cover the OpenAI Compatibility API, the beta LM Studio REST API, and Python/TypeScript SDKs. There’s also a CLI (lms) and a headless build (llmster) for servers and CI environments.
Pricing / Limitations
LM Studio is free to download and use — no license fees, no seat fees. Your costs are the hardware you already own. Open models are free to download from Hugging Face (check each model’s license for commercial-use terms).
Limitations: quality and speed depend on the model and your hardware — a small model on an old laptop is slower and less capable than a large model on a beefy machine, and very large frontier models need serious RAM. Like all LLMs, local models can hallucinate, so keep a human in the loop for high-stakes work.
Bottom Line for SMB
LM Studio is the easiest way to run real AI models on your own computers with zero code and zero cloud dependency. Start with one practical job — private drafting, offline document Q&A, or local coding help — prove the value on a single machine, then explore server mode when your own apps should call a local model. Prefer the terminal? Read our Ollama tutorial; for a private chatbot over your whole document library, see our AnythingLLM tutorial.
Frequently Asked Questions
Is LM Studio really free?
Yes — free to download and use on Mac, Windows, and Linux, and the open models it runs are free to download from Hugging Face. Check each model’s license before commercial use.
Does LM Studio need the internet?
Only to download the app and model weights. After that, chats, document Q&A, and the Bionic agent run entirely on your device.
What’s the difference between LM Studio and Ollama?
Both run models locally. LM Studio is a full desktop app with a chat UI, document chat, MCP support, and the Bionic agent — no terminal needed. Ollama is lighter and command-line-first, better for developers and servers. We cover Ollama in a separate tutorial.
Sources
- LM Studio homepage — product overview, Bionic agent, download platforms
- LM Studio Docs — setup, model downloads, server mode, REST API, MCP, headless llmster (checked October 19, 2026)
