Why LM Studio
LM Studio provides a graphical interface for the same llama.cpp inferenceInferenceThe process of running a trained AI model to generate outputs — what happens when you send a prompt and receive a response.Learn more → engine that Ollama uses under the hood. It is ideal for users who prefer not to use a terminal and for teams that want to evaluate models visually before deploying them.
The key feature that sets LM Studio apart is its built-in Hugging Face browser, which lets you search, filter, and download any GGUFGGUFA file format for distributing quantized language models optimized for efficient local inference, popularized by llama.cpp.Learn more → model without leaving the app. It also includes a local server that exposes an OpenAI-compatible API.
Install LM Studio
Download the installer from lmstudio.ai. LM Studio is available for macOS (Apple Silicon and Intel), Windows, and Linux. On macOS, it is distributed as a .dmg file; on Windows as an NSIS installer.
After installation, open LM Studio. On first launch it will ask for a download directory for model files — pick a location on a drive with at least 20 GB free if you plan to experiment with larger models.
Download a model
Click the magnifying glass icon in the left sidebar to open the model search. Type a model name such as 'Phi-4 Mini' or 'Mistral Small'. LM Studio searches Hugging Face and shows available GGUF quantisations. For most users, the Q4_K_M quantisation offers the best balance of quality and size.
Click Download next to your chosen quantisation. Progress shows in the bottom bar. Models are stored in `~/Documents/LM Studio/Models` by default. The download is a one-time operation — the model stays cached locally.
Chat with your model
Click the chat bubble icon, select your downloaded model from the dropdown at the top, and start typing. LM Studio shows tokenTokenThe basic unit of text that an LLM processes — roughly corresponding to a word or word-piece. Models read input and produce output in tokens, which is also how API usage is measured and billed.Learn more → throughputThroughputThe number of tokens or requests an AI system can process per unit time, measuring overall capacity rather than per-request speed.Learn more → and memory usage in real time.
You can customise the system promptSystem PromptA special instruction given to a language model before the user conversation begins, establishing the model's persona, capabilities, constraints, and context.Learn more →, temperatureTemperatureA parameter controlling the randomness of model outputs — lower values produce more focused, deterministic responses; higher values produce more creative, varied text.Learn more →, context length, and GPU layer offload from the right sidebar. Setting GPU layers to the maximum your VRAMVRAMVideo Random Access Memory (VRAM) is the dedicated memory on graphics cards that stores model weights and the KV cache during LLM inference.Learn more → allows dramatically speeds up inference — if a 7B model fits in 8 GB VRAM, set layers to 33 to run it entirely on the GPU.
Start the local API server
Click the server icon (looks like `</>`) in the left sidebar, select a loaded model, and click Start Server. The server runs at `http://localhost:1234` by default and exposes an OpenAI-compatible API.
You can now point any OpenAI SDK client at `http://localhost:1234/v1` with any API key (it is ignored). This lets you test local models with your existing application code without modifying anything except the base URL.