LM Studio Interface Explained: Every Button and Setting

1. 
   1

   Chat

   The chat screen you are on now. Talk to a model that runs on your own computer.

2. 
   2

   Model Server

   Run a local API server so other apps on your computer can use your models.

3. 
   3

   Library

   Every model you have downloaded, with its details and load settings.

4. 
   4

   LM Link

   Use models running on your other computers as if they were on this one.

5. 
   5

   Model Search

   Find and download new models from Hugging Face.

6. 
   6

   Settings

   Opens Settings: updates, appearance, runtimes, hardware and more.

7. 
   7

   Hide sidebar

   Hides the sidebar to give the chat more room. Click again to bring it back.

1. 
   1

   New chat

   Starts a new chat. The shortcut is Cmd + N (Ctrl + N on Windows).

2. 
   2

   Search chats

   Find an old chat by searching your chat history.

3. 
   3

   New Folder

   Makes a folder for your chats. Drag chats in and out of it. The shortcut is Cmd + Shift + N.

4. 
   4

   Folder

   A folder of chats. Click the arrow to open or close it.

5. 
   5

   Chat menu

   Options for this chat, like Duplicate, which copies the chat along with any files in it.

1. 
   1

   Model picker

   The model this chat is using. Click it to switch to another model or load one.

2. 
   2

   Eject

   Unloads the current model from memory. Do this to free up RAM when you are done.

3. 
   3

   Split view

   Shows two chats side by side, for example to compare two models on the same question.

4. 
   4

   Downloads

   Shows your model downloads and how far along they are.

5. 
   5

   Right panel

   Shows or hides the panel on the right with integrations and model settings.

1. 
   1

   Integrations tab

   The plugins and tools this chat can use.

2. 
   2

   Model Parameters tab

   The settings for how the model answers in this chat, like the system prompt and temperature.

3. 
   3

   Install

   Add more plugins or MCP servers. Edit mcp.json lives here, where you paste in MCP server settings.

4. 
   4

   Filter plugins

   Type to find a plugin in the list.

5. 
   5

   Plugin switch

   Turns a plugin on or off. js-code-sandbox lets the model run JavaScript code, and rag-v1 is LM Studio's built-in plugin for chatting with documents.

1. 
   1

   Regenerate

   Writes a new answer to the same message. The old answers are kept so you can flip between them.

2. 
   2

   Continue

   Makes the model keep writing from where this reply ends.

3. 
   3

   Branch

   Starts a new chat with everything up to this message, so you can try a different direction.

4. 
   4

   Copy

   Copies the reply text.

5. 
   5

   Edit

   Change the reply text yourself. The model sees your edited version from then on.

6. 
   6

   Delete

   Deletes this message from the chat.

Status symbols. These only show information. They are not buttons.

1. 
   7

   Reply stats

   How the reply went: 26.09 tokens per second, 1,027 tokens in total, 0.41 seconds before the first word, and why it stopped. EOS Token Found means the model simply finished.

1. 
   1

   Message box

   Type your question or instructions for the model here.

2. 
   2

   Attach

   Opens a small menu to attach a file or an image.

3. 
   3

   Integrations menu

   A quick menu to turn plugins on or off without opening the right panel.

4. 
   4

   Think

   Turns thinking on or off for models that can reason before they answer. Thinking gives better answers to hard questions but takes longer.

5. 
   5

   Vision

   Shows the loaded model can understand images, so you can attach pictures and ask about them.

6. 
   6

   Chat with Files

   Lets the model read documents you attach. The x removes it.

7. 
   7

   Code Interpreter

   Lets the model write and run JavaScript in a sandbox to work things out, like math. The x removes it.

1. 
   1

   Attach file

   Add a .pdf, .docx or .txt file. Short files go to the model in full. For long ones, LM Studio picks out the parts that match your question.

2. 
   2

   Attach image

   Add a picture for the model to look at. This needs a model with vision, like Gemma 4 E4B here.

3. 
   3

   Send

   Sends your message to the model.

Status symbols. These only show information. They are not buttons.

1. 
   4

   Context used

   How much of the model's context length this chat has used, 15% here. When it fills up, the Context Overflow setting decides what gets dropped.

1. 
   1

   Preset

   Pick a saved preset: a system prompt plus settings you can reuse in any chat. Unsaved Preset means you have changed something.

2. 
   2

   Discard unsaved

   Throws away changes you have not saved to a preset.

3. 
   3

   Save Preset As

   Saves the current system prompt and settings as a new preset with its own name.

4. 
   4

   System Prompt

   Standing instructions the model follows for the whole chat, like "Only answer in rhymes" or "Keep answers short."

5. 
   5

   Editor

   Opens a bigger editor for long system prompts. The shortcut is Cmd + E.

Status symbols. These only show information. They are not buttons.

1. 
   6

   Token count

   How many tokens your system prompt uses. A token is about three quarters of a word, and these count toward the context length.

1. 
   1

   Temperature

   How creative the answers are. Higher values make them more random, lower values make them more predictable. It is set to 1 here.

2. 
   2

   Limit Response Length

   Tick it to cap how many tokens the model can write in one answer.

3. 
   3

   Context Overflow

   What happens when the chat gets longer than the context length. Truncate Middle keeps your system prompt and first message and drops the middle. Rolling Window drops the oldest messages, and Stop at Limit ends the answer.

4. 
   4

   Stop Strings

   Words or symbols that make the model stop writing as soon as it produces them. Type one and press Return.

5. 
   5

   Sampling

   Fine controls for how the model picks each next word, such as Top K, Top P, Min P and Repeat Penalty. Most people can leave these alone.

6. 
   6

   Conversation Notes

   Add your own notes to this chat.

1. 
   1

   Filter models

   Type to find one of your downloaded models by name.

2. 
   2

   Loaded model

   The model in memory right now, which answers your chats.

3. 
   3

   Eject button

   Unloads the model and frees its memory.

4. 
   4

   Sort

   Sort your models by Recency, Size or Downloaded.

5. 
   5

   Close

   Closes the picker without changing the model.

Status symbols. These only show information. They are not buttons.

1. 
   6

   Context badge

   The context length the model was loaded with: 8,192 tokens. That is how much text it can keep in mind at once.

2. 
   7

   Memory Consumption

   How much memory your loaded models use, 7.89 GB, out of the 24 GB in this Mac.

1. 
   1

   Model

   Click a model to load it. The row also shows the publisher, the size in parameters (4B), the model family and the file size.

2. 
   2

   Choose load parameters

   When this is on, picking a model shows its load settings, like context length, before it loads. Holding Option does the same for one load.

Status symbols. These only show information. They are not buttons.

1. 
   3

   Quantization

   How much the model file is compressed. 4-bit files are much smaller but lose a little quality. LM Studio suggests 4-bit or higher if your computer can handle it.

2. 
   4

   Vision badge

   The eye means the model can understand images.

3. 
   5

   Tool use badge

   The hammer means the model can use tools, like plugins and MCP servers.

4. 
   6

   Reasoning badge

   The green brain means the model can think through a problem before it answers.

5. 
   7

   Model format

   The file format. MLX is Apple's format and runs only on Apple silicon Macs. GGUF runs on Mac, Windows and Linux.

1. 
   1

   Search models

   Search by name, like gemma 4 e4b. You can also paste a user/model name or a Hugging Face link.

2. 
   2

   Format filter

   Show GGUF models, MLX models or both. On a Mac you can use either; on Windows and Linux only GGUF works.

3. 
   3

   Sort results

   Changes the order of the results. Best Match puts the closest matches first.

4. 
   4

   Community file

   A single model file from Hugging Face, here uploaded by lmstudio-community. The heart shows likes and the arrow shows downloads.

Status symbols. These only show information. They are not buttons.

1. 
   5

   Capability icons

   Eye, hammer and brain show support for images, tools and reasoning.

1. 
   1

   Copy model name

   Copies the model's full name, like google/gemma-4-e4b.

2. 
   2

   Use in New Chat

   This model is already downloaded, so the button opens a new chat with it.

Status symbols. These only show information. They are not buttons.

1. 
   3

   Popularity

   How many times the model was downloaded and how many people liked it.

2. 
   4

   Staff Pick

   LM Studio's team recommends this model.

3. 
   5

   Parameters

   The model's size: 7.9B means 7.9 billion parameters. Bigger models are usually smarter but need more memory.

4. 
   6

   Formats

   The formats this model comes in, GGUF and MLX. MLX is built for Apple silicon Macs.

5. 
   7

   Capabilities

   What the model can do: Vision (understand images), Tool Use (call tools) and Reasoning (think before it answers).

1. 
   1

   Download options

   Pick which version to download. This one is the MLX file, 4-bit, 15.64 GB.

2. 
   2

   Download

   Downloads the model, 15.64 GB here. Check you have that much free disk space first.

Status symbols. These only show information. They are not buttons.

1. 
   3

   Full GPU Offload Possible

   The whole model fits in your graphics memory, so the GPU can run all of it. GPU offload means moving the model onto the GPU, which is much faster than the CPU.

1. 
   1

   Model types

   Filter by type. LLMs are chat models, Text Embedding models turn text into numbers for document search, and Drafters are small helper models for speculative decoding.

2. 
   2

   Filter my models

   Type to find a model. The shortcut is Cmd + F.

3. 
   3

   Model settings

   Sets this model's default load settings, like context length. They apply every time the model loads.

4. 
   4

   Models folder

   Where LM Studio saves your models on disk.

Status symbols. These only show information. They are not buttons.

1. 
   5

   Arch

   The model family it is built on, like qwen3 or gemma4.

2. 
   6

   Params

   The model's size in billions of parameters, like 0.6B or 4B.

1. 
   1

   Use in New Chat button

   Opens a new chat with this model.

2. 
   2

   Load Model

   Loads the model into memory without opening a chat, for example to use it from the local server.

3. 
   3

   Info tab

   The model's format, quantization, family, capabilities and size on disk.

4. 
   4

   Load tab

   Settings used when the model loads, like context length.

5. 
   5

   Inference tab

   Settings for how this model answers, like the system prompt and temperature.

1. 
   1

   Context Length

   How much text the model can keep in mind at once, in tokens. This model supports up to 40,960, but a longer context uses more memory.

2. 
   2

   Max Concurrent Predictions

   How many requests the model works on at the same time instead of making them wait in line. The default is 4.

3. 
   3

   Seed

   Tick it and set a number to get the same answer every time for the same prompt. Unticked, LM Studio uses a random seed.

4. 
   4

   KV Cache Quantization

   Compresses the model's working memory of the chat (the KV cache) so long chats need less RAM. It can cost a little quality and is still experimental.

5. 
   5

   Prompt Disk Cache

   Saves the already processed part of a chat to disk, so follow-up messages do not have to process it again.

1. 
   1

   Model system prompt

   A system prompt for this model, used in its chats.

2. 
   2

   Model temperature

   How creative this model's answers are. 0.8 here; lower is more predictable.

3. 
   3

   Reasoning Parsing

   Tells LM Studio how to spot the model's thinking, so it shows apart from the final answer.

4. 
   4

   Sampling settings

   Fine controls for how the model picks each next word, such as Top K, Top P and Repeat Penalty.

1. 
   1

   Local Server

   The page you are on, with the server controls and the models it can use.

2. 
   2

   Status

   Starts or stops the local server. It shows Stopped here. Once it runs, apps can reach your models at localhost:1234.

3. 
   3

   Server Settings

   Server options, such as the port and serving on your local network so other devices can connect.

4. 
   4

   mcp.json

   Opens the file where you add MCP servers, which give models extra tools. It uses the same format as Cursor.

5. 
   5

   Load Model button

   Loads a model into memory so the server can use it.

6. 
   6

   Copy model ID

   Copies the model's name, to paste into your code.

7. 
   7

   Eject model

   Unloads the model from memory. READY on the left means it is loaded and can answer requests.

1. 
   1

   Developer Docs

   LM Studio's developer documentation, right inside the app.

2. 
   2

   API reference

   Short guides for each API request. POST and GET show the type of request.

3. 
   3

   API type tabs

   Switch between LM Studio's own API and the OpenAI-compatible and Anthropic-compatible ones, so apps built for those services work with your local models.

4. 
   4

   Clear logs

   Clears the server log below.

5. 
   5

   API Model Identifier

   The name to use in the model field of your API requests. The button on the right copies it.

1. 
   1

   Check for updates

   Looks for a newer version of LM Studio. This Mac has 0.4.28 installed.

2. 
   2

   Updates channel

   Which releases LM Studio updates to. It is set to Stable here.

3. 
   3

   App Language

   Changes the language of the app. LM Studio offers more than 30 languages, and this feature is still in development.

4. 
   4

   Full model loader

   Skips the quick model picker and always opens the full loader with every load setting.

5. 
   5

   Hugging Face Proxy

   Downloads models from Hugging Face through LM Studio's own proxy, for more reliable downloads.

6. 
   6

   Models directory

   The folder where downloaded models are saved. Use the menu at the right to change it.

7. 
   7

   Account

   Shows if you are signed in to an LM Studio account. You need one for LM Link.

1. 
   1

   Appearance

   Pick the color theme for the app.

2. 
   2

   Developer

   Turns on Developer Mode, which shows every advanced setting, like load and inference options.

3. 
   3

   Model Defaults

   The default settings LM Studio uses for models.

4. 
   4

   Runtime

   The engines that run models. The shortcut is Cmd + Shift + R.

5. 
   5

   Hardware

   What your computer has and how much of it models can use. The shortcut is Cmd + Shift + H.

1. 
   1

   GPU switch

   Turns the GPU on or off for running models. Leave it on: the GPU (Metal on a Mac) is much faster than the CPU.

2. 
   2

   Guardrails

   How careful LM Studio is about loading a model too big for your memory. OFF has no protection against overloading your computer, and Strict has the most.

3. 
   3

   Copy Info

   Copies your hardware details, handy when you ask for help.

4. 
   4

   Community

   Opens LM Studio's community, where you can ask for help.

Status symbols. These only show information. They are not buttons.

1. 
   5

   CPU check

   Confirms your processor can run LM Studio, an Apple M5 here.

2. 
   6

   Memory Capacity

   Your RAM, 24 GB. On a Mac the memory is shared (unified memory), so VRAM shows the part the graphics chip can use, 17.76 GB.

3. 
   7

   Resource Monitor

   How much memory and CPU LM Studio is using right now.

1. 
   1

   GGUF runtime

   The engine for GGUF models. On a Mac it is llama.cpp running on Metal, Apple's graphics system.

2. 
   2

   MLX runtime

   The engine for MLX models, Apple's format for Apple silicon. This version adds support for the M5's Neural Accelerators.

3. 
   3

   Auto-update runtimes

   Keeps the selected engines up to date on their own.

4. 
   4

   Runtime updates

   Checks for newer engine versions. The menu next to it picks the update channel, Stable here.

5. 
   5

   Engines & Frameworks

   Every engine and add-on. Compatible only hides the ones that cannot run on your computer.

6. 
   6

   Download engine

   Installs an extra engine or add-on, like Office Document Processor for Word and PowerPoint files.

Status symbols. These only show information. They are not buttons.

1. 
   7

   Latest version

   This engine is already up to date.