기본 콘텐츠로 건너뛰기

LM Studio Tutorial: Run a Local LLM and Chat With Your Files

Quick summary: LM Studio is a desktop app that downloads open-weight language models from Hugging Face and runs them on your own Mac, Windows or Linux computer, with a ChatGPT-style chat window and an optional local API server. Once a model is on your disk, chatting, attaching documents and serving the model to other apps all work without an internet connection. This guide walks through the setup in the order the official docs recommend, explains which download option to pick for your hardware, and shows how to point existing OpenAI-based tools at your own machine.

What's in this guide

  1. Check your computer before you download anything
  2. Install LM Studio and get your first model
  3. Load the model and chat
  4. Chat with your own documents
  5. Turn LM Studio into a local API for other apps
  6. LM Studio vs Ollama vs Jan
  7. Editor's note, FAQ and related guides

Step 1: Check your computer before you download anything

Local models live in your RAM (and, if you have one, your graphics card's memory), so hardware decides what you can run. LM Studio's system requirements page lists these minimums:

  • Mac: Apple Silicon (M1, M2, M3 or M4) on macOS 14.0 or newer. 16GB of RAM or more is recommended; 8GB Macs can work if you stick to smaller models and modest context sizes. Intel Macs are not supported.
  • Windows: x64 or ARM (Snapdragon X Elite) PCs. On x64 the CPU must support the AVX2 instruction set. At least 16GB of RAM and at least 4GB of dedicated VRAM are recommended.
  • Linux: x64 or ARM64, distributed as an AppImage, Ubuntu 20.04 or newer. The docs note that Ubuntu versions newer than 22 are not well tested.

If your machine falls short of the recommended RAM, LM Studio may still install, but you'll be limited to small models. That's worth knowing before you spend twenty minutes downloading a multi-gigabyte file that won't load.

Step 2: Install LM Studio and get your first model

Install the app

Download the installer for your operating system from the official download page (lmstudio.ai/download). At the time of writing, the page offered LM Studio version 0.4.26, available under LM Studio's terms of use. The same page also promotes LM Studio Bionic, which is a separate, newer agent app from the same company; for this tutorial you want the regular LM Studio download, described on the page as "Experiment with LLMs on your computer. Chat interface and programmable API."

Find a model in the Discover tab

LM Studio has a built-in downloader that pulls supported models from Hugging Face. Open the Discover tab (shortcut: Cmd+2 on Mac, Ctrl+2 on Windows and Linux). You can pick one of the curated options or search in three ways:

  • by keyword, such as llama, gemma or qwen;
  • by a specific user/model string from Hugging Face;
  • by pasting a full Hugging Face URL into the search bar.

Pick the right download option

Most models show several files with names like Q3_K_S, Q4_K_M or Q8_0. These are all the same model compressed to different degrees, a technique called quantization. Smaller numbers mean smaller files and lower memory use, at the cost of some quality. The official guidance is short: choose a 4-bit option or higher if your machine can handle it. In practice, a Q4 file is the usual starting point, and you move up to Q6 or Q8 only if you have memory to spare.

LM Studio runs GGUF models through llama.cpp on all three platforms, and on Apple Silicon Macs it can also run MLX models. If you're on a Mac, you may see both formats for the same model.

Step 3: Load the model and chat

Downloading a model only puts it on your disk. To use it, you load it into memory:

  1. Go to the Chat tab.
  2. Open the model loader (Cmd+L on Mac, Ctrl+L on Windows and Linux).
  3. Select the model you downloaded.
  4. Optionally, adjust load settings before it loads, such as GPU offload and context size.

Once it's loaded, type a message and the conversation runs entirely on your computer. New chats are Cmd+N / Ctrl+N, and you can organise conversations into folders (Cmd+Shift+N / Ctrl+Shift+N) and drag chats between them. Conversations are saved as JSON files in ~/.lmstudio/conversations/ on Mac and Linux, or %USERPROFILE%\.lmstudio\conversations on Windows.

Save load settings per model

If a model only fits with a particular GPU offload or context length, you don't have to set it every time. In the My Models tab, click the gear icon next to a model to set its default load parameters. The docs mention three common reasons to do this: a specific GPU offload level, a specific context size, and whether to use Flash Attention. Those defaults apply wherever the model is loaded, including from the command line.

Reuse prompts with Presets

Presets bundle a system prompt with inference settings such as Temperature, Top P and Max Tokens. If you use the same model for, say, email drafts and code review, save each setup as a Preset and switch between them instead of rewriting the system prompt. Presets are plain JSON files, stored in ~/.lmstudio/config-presets (Mac/Linux) or %USERPROFILE%\.lmstudio\config-presets (Windows).

Step 4: Chat with your own documents

You can attach .docx, .pdf and .txt files to a chat. What happens next depends on length:

  • If the document fits in the model's context window, LM Studio adds the whole file to the conversation.
  • If it's too long, LM Studio switches to retrieval-augmented generation (RAG): it tries to pull out only the relevant passages and hands those to the model.

LM Studio's offline documentation states that dropped-in documents stay on your machine and all document processing is done locally. That makes this the feature most people install LM Studio for: summarising a contract, a lab report or meeting notes you'd rather not upload to a cloud chatbot.

The docs are candid that RAG "sometimes requires some tuning and experimentation." Their one concrete tip is to put as much context in your question as possible, including the terms and phrases you expect to appear in the relevant part of the document. Asking "What does section 4 say about the termination notice period?" will usually retrieve better passages than "Summarise this."

Step 5: Turn LM Studio into a local API for other apps

This is where LM Studio becomes more than a chat window. It can serve a loaded model on OpenAI-compatible endpoints, so tools built for the OpenAI API can talk to your own computer instead.

  1. Open the Developer tab and switch the server toggle on. The status shows as running, along with the address it is reachable at.
  2. Or start it from a terminal with lms server start (and stop it with lms server stop). The lms command-line tool ships with LM Studio, but you need to open the app at least once before it works.
  3. In your client code, change the base URL. The official examples assume port 1234, so a Python OpenAI client becomes OpenAI(base_url="http://localhost:1234/v1"), and the model field takes the model identifier shown in LM Studio.

The OpenAI-compatible endpoints listed in the docs are /v1/models, /v1/responses, /v1/chat/completions, /v1/embeddings and /v1/completions, plus an LM Studio-specific /v1/decisions. The docs also list Anthropic-compatible endpoints and say Codex works through the /v1/responses endpoint.

Server settings worth knowing

  • Just-in-Time model loading is on by default. Apps can request any downloaded model and LM Studio loads it on demand, so /v1/models lists all downloaded models, not just loaded ones.
  • Idle TTL unloads a JIT-loaded model after it sits unused; the default is 60 minutes, and you can set a different value per request with a ttl field.
  • Auto-Evict (on by default) unloads the previous JIT-loaded model before loading a new one, so switching models from a client app doesn't pile them up in memory.
  • Serve on Local Network lets other devices on your network use the server, and Require Authentication makes clients send an API token. Turn on the second whenever you turn on the first.

LM Studio vs Ollama vs Jan

All three run open models locally, but they differ in interface, licensing and how they make money. The table below uses only what each official site states today.

  LM Studio Ollama Jan
Local use Desktop app download; no price listed for the app. Terms allow personal and internal business use. Free plan; the pricing FAQ says running models on your own hardware is always unlimited. Desktop app described as free and open source (Apache 2.0).
Paid options Pricing page covers the separate Bionic app: Bionic+ $20/month and Pro $100/month for US-hosted cloud models. Pro $20/month ($200/year) with $60 of cloud usage credits; Max $100/month with $300 of credits. None listed on the Jan Desktop pages checked.
Main interface GUI with Discover, Chat, Developer and My Models tabs, plus the lms CLI. Installed via a one-line script or manual download; runs models from the command line. Desktop app with projects, assistants, web search and file upload.
API / cloud OpenAI- and Anthropic-compatible endpoints, default port 1234 in the docs. Runs models locally or in Ollama's cloud; cloud model names end in :cloud. Local OpenAI-compatible API server.
Best fit People who want to browse, compare and tune models visually. Developers who live in the terminal or want a cloud fallback. People who want a fully open-source chat app.

If you already tried Ollama and found the command line awkward, LM Studio is the natural next step: same idea, but you can see the download options, file sizes and load settings before you commit.

Editor's note

Good for: anyone who wants a private assistant for sensitive documents, developers who want to test OpenAI-style code against a free local model before paying for API calls, and people who travel or work offline.

Not for: Intel Mac owners (not supported), PCs whose x64 CPU lacks AVX2, or anyone expecting cloud-chatbot quality from a small model on an 8GB laptop.

Gotchas I found in the docs:

  • Two apps, one company. The download and pricing pages lead with LM Studio Bionic. The paid plans you see there are for Bionic's cloud models, not for running local models in regular LM Studio.
  • Offline has limits. Searching the Discover tab, downloading models or runtimes, and checking for updates all need internet. Download what you need before you go offline.
  • Linux updates are manual. The offline docs say the in-app updater covers macOS and Windows, while the Linux updater is still in the works.
  • Memory creep from other apps. With JIT loading on, a connected app can load models in the background. Lower the Idle TTL if your machine slows down after using editor plugins.

Common errors and fixes:

  • Model won't load or the computer freezes: pick a smaller quantization (a lower Q number) or a smaller model, and reduce the context size in the load settings.
  • lms command not found or not working: open the LM Studio app at least once, as the CLI docs require.
  • Document answers miss the point: the file was probably too long and went through RAG. Ask with specific terms from the document, or split it into smaller files.
  • Another device can't reach the server: turn on Serve on Local Network in Server Settings, and use the port shown in the Developer tab.

For a visual reference, see the official LM Studio screenshots at the end of this post.

FAQ

Does LM Studio send my chats to the cloud?

Not for local models. LM Studio's offline documentation says that once a model is on your machine, nothing you enter while chatting leaves your device, and attached documents are processed locally. Internet access is only needed for searching and downloading models, downloading runtimes and checking for app updates.

Can I use LM Studio at work?

The LM Studio desktop app's terms (version dated August 23, 2026) grant a license for "personal and / or internal business purposes." They also restrict things like reselling the software or offering it as a service, so read the terms if you plan anything beyond internal use.

Which model should I download first?

Start small. A model with a few billion parameters at a 4-bit quantization (a Q4 file) loads quickly on most 16GB machines and lets you confirm everything works. Once you know how your hardware behaves, try a larger model or a higher quantization.

Related guides on lightmind24

Download one small model, attach a document you'd never paste into a cloud chatbot, and you'll know in ten minutes whether local AI earns a place on your computer.

Images: official screenshots from the LM Studio documentation (lmstudio.ai/docs), included at the end of this post.

Last verified: 2026-10-26
Sources:
LM Studio Docs: Welcome
LM Studio Docs: System Requirements
LM Studio Docs: Get started
LM Studio Docs: Download an LLM
LM Studio Docs: Manage chats
LM Studio Docs: Chat with Documents
LM Studio Docs: Offline Operation
LM Studio Docs: Per-model Defaults
LM Studio Docs: Presets
LM Studio Docs: Local Server
LM Studio Docs: Server Settings
LM Studio Docs: Idle TTL and Auto-Evict
LM Studio Docs: OpenAI Compatibility
LM Studio Docs: lms CLI
LM Studio download page
LM Studio (Bionic) pricing
LM Studio Desktop App Terms of Service
Ollama pricing
Ollama download
Jan Docs: Overview

Uploaded Image Uploaded Image Uploaded Image Uploaded Image

댓글

이 블로그의 인기 게시물

브루(Vrew) 요금제 정리: 무료 200크레딧으로 어디까지 쓸 수 있나

요약: 영상 자막 편집기 브루(Vrew)는 2026년 4월 22일부터 기능별 사용량 대신 '통합 크레딧' 하나로 요금을 매겨요. 무료 플랜은 매달 200크레딧이 들어오는데, 음성 분석(자동 자막)만 쓰면 최대 200분 분량이라 짧은 영상 몇 편은 충분하지만, 영상 초반의 Vrew 마크를 지우려면 유료 플랜이 필요해요. 이 글은 브루 공식 가격 페이지와 고객센터 공지를 2026년 10월 23일에 다시 확인해, 크레딧 계산법과 플랜 고르는 기준을 정리했어요. 목차 브루 요금 구조: 통합 크레딧이란 1크레딧으로 할 수 있는 일 무료 플랜으로 되는 것과 안 되는 것 유료 플랜 4종 비교표 월 구독·연 구독·기간 이용권, 무엇이 이득일까 상황별 추천: 학생·직장인·유튜버 크레딧이 모자랄 때: 사용 기록 지우개 업그레이드·환불 전에 알아둘 규칙 운영자 메모, 자주 묻는 질문 1. 브루 요금 구조: 통합 크레딧이란 예전 브루는 플랜마다 음성 분석 몇 분, AI 목소리 몇 자, 번역 몇 자처럼 기능별 한도를 따로 줬어요. 그런데 보이저엑스(브루 운영사)가 실제 이용 데이터를 보니 한두 가지 기능만 집중해서 쓰는 사람이 훨씬 많았다고 해요. 그래서 2026년 4월 22일 업데이트부터는 플랜마다 정해진 크레딧을 받고, 어떤 기능에 쓸지는 내가 정하는 방식으로 바뀌었어요. 플랜별 매달 지급량은 Free 200 · Light 2,000 · Standard 10,000 · Business 12,000크레딧 이에요. 공식 공지에 따르면 무료 사용자는 매달 1일에, 유료 사용자는 매월 결제일에 충전되고, 남은 크레딧은 다음 달로 넘어가지 않아요. 바뀐 뒤 무엇이 좋아졌나 공지에 나온 예시가 이해하기 쉬워요. Standard 플랜에서 AI 목소리만 쓰면 예전에는 최대 50만 자였지만, 지금은 1만 크레딧을 모두 AI 목소리에 몰아 쓸 수 있어 최대 100만 자가 돼요. 반대로 여러 기능을 골고루 쓰는 사람은 크레딧이 기능끼리 ...

에이닷 전화 vs 익시오: 통화 녹음·AI 요약, 내 통신사·폰에 맞는 앱 고르기

요약: 통화 녹음과 AI 요약을 쓰고 싶다면 먼저 내 통신사와 휴대폰 종류부터 확인해야 해요. LG유플러스 고객은 무료 앱 익시오, SKT 고객은 에이닷 전화가 가장 기능이 많고, 다른 통신사의 아이폰 사용자는 iOS에 들어 있는 기본 통화 녹음·전사 기능이 현실적인 선택이에요. 이 글은 2026-10-30에 SKT·LG유플러스·Apple 공식 페이지와 도움말을 직접 확인해 세 가지를 비교했어요. 왜 '앱 기능'보다 '내 조건'을 먼저 봐야 할까요 통화 요약 앱 비교 글을 보면 기능 목록부터 나열하는 경우가 많아요. 그런데 실제로 써 보려는 사람에게 가장 먼저 걸리는 건 기능이 아니라 이용 조건 이에요. 익시오는 LG유플러스 고객 전용이고, 에이닷 전화는 누구나 설치할 수 있지만 통역콜·baro 같은 기능은 SKT 고객에게만 열려 있어요. 아이폰은 운영체제 구조상 앱이 통화를 마음대로 녹음하기 어려워서, 통신사 앱이 안 되면 iOS 기본 기능에 기대야 해요. 그래서 이 글은 "어떤 앱이 제일 좋은가"보다 "내 통신사·내 폰에서 무엇을 쓸 수 있나" 순서로 정리했어요. 기능 차이는 그다음이에요. 한눈에 비교: 에이닷 전화 vs 익시오 vs 아이폰 기본 통화 녹음 구분 에이닷 전화 (SKT) 익시오 ixi-O (LG유플러스) 아이폰 기본 통화 녹음 (Apple) 누가 쓸 수 있나 통신사 구분 없이 앱 설치 가능. 단, 통역콜·baro·그룹통화 발신 등은 SKT 고객 전용 U+ 모바일 고객. 타 통신사, 만 14세 미만, 선불요금제 고객은 제한 (공식 사이트 FAQ상 알뜰폰 중 KB리브엠은 예외) 지원 국가·언어의 iPhone 사용자. 1명과 통화할 때만 녹음 가능 지원 OS·기기 Android 10 이상, iOS 16 이상 iOS 17 이상 iPhone 12 이상 / Android...

타입캐스트 vs 클로바더빙: 수익 나는 채널에 쓸 때 무료·유료 조건 비교

요약 타입캐스트와 클로바더빙은 둘 다 글만 입력하면 한국어 AI 목소리를 만들어 주는 서비스지만, 무료로 쓸 수 있는 범위가 꽤 다릅니다. 클로바더빙 무료는 매달 15,000자까지 쓸 수 있는 대신 수익이 생기는 채널에는 쓸 수 없고, 타입캐스트 무료는 처음에 주는 약 5분 분량의 다운로드로 끝나는 체험판에 가깝습니다. 광고 수익이 붙은 유튜브나 쇼핑몰 영상에 넣을 목소리라면 처음부터 유료 플랜 조건을 보고 고르는 편이 안전합니다. 왜 이 두 서비스를 같이 비교하나 쇼츠 내레이션, 수업 자료 더빙, 상품 소개 영상처럼 "내 목소리 대신 AI 목소리"가 필요할 때 한국 사용자들이 가장 많이 찾는 곳이 네오사피엔스의 타입캐스트 와 네이버의 클로바더빙 입니다. 기능 소개만 보면 비슷해 보이지만, 실제로 문제가 되는 건 음질보다 "이 목소리를 내 채널에 써도 되나" 입니다. 그래서 이 글은 기능 나열보다 무료 사용 범위, 출처 표기 의무, 유료 플랜 가격과 다운로드 한도를 중심으로 정리했습니다. 모든 수치는 2026년 10월 25일에 두 서비스의 공식 요금 페이지와 고객센터 도움말에서 직접 다시 확인한 내용입니다. 한눈에 보는 비교표 항목 타입캐스트 무료 타입캐스트 베이직 클로바더빙 Free 클로바더빙 Standard 월 요금 0원 9,900원 (연간 결제 시 10% 할인) 0원 19,900원 (네이버 클라우드 플랫폼에서 결제) 무료·포함 사용량 음성 생성·재생 무제한, 다운로드는 최초 3,000크레딧(약 5분) 1회 지급 월 30,000크레딧(약 60분) 월 15,000자, 다운로드 월 20회, 프로젝트 누적 5개 (매월 1일 갱신) 월 30,000자, 다운로드 월 40회, 프로젝트 누적 20개 쓸 수 있...