LM Studio is large language model inference software for running and using large language models locally on personal computers. It provides tools for downloading and managing models, interacting with them through a chatbot interface, and serving locally running models to other applications through an application programming interface (API).^[1][2]

LM Studio uses llama.cpp as an inference engine for models in the GGUF format and also supports Apple's MLX framework on Apple silicon.^[3] In addition to its graphical interface, the platform includes command-line tools, software development kits for Python and TypeScript, and a local inference server with native, OpenAI-compatible, and Anthropic-compatible APIs.^[4][5]

LM Studio was launched in May 2023.^[6] It was initially developed as a desktop environment combining local model inference with model discovery, a chat interface, local model management, and an API server.^[7]

Support for the MLX inference framework on Apple silicon was introduced in October 2024, alongside its existing llama.cpp based inference system.^[8] Subsequent releases expanded the software's developer-oriented capabilities, including headless operation, on-demand model loading, command-line model downloading, tool calling, and speculative decoding.^[9][10]

In April 2026, Element Labs, the developer of LM Studio, acquired Locally AI,^[11] whose developer joined the company to work on native AI applications across Apple devices.^[12]

LM Studio provides an integrated interface for finding and downloading AI models, loading models into memory, configuring inference settings, and interacting with models through a chat interface. Models can be obtained through integrations with repositories such as Hugging Face.^[13] The application can also use retrieval-augmented generation to provide local documents as context to language models.^[14]

The software can operate as a local inference server, allowing applications to access loaded models through LM Studio's REST API or compatibility interfaces modeled on the OpenAI and Anthropic APIs.^[15] Its Python and TypeScript software development kits provide programmatic model loading, text generation, embeddings, and agentic tool use.^[16]

LM Studio can also operate without its graphical interface. Its headless component, llmster, packages the core LM Studio runtime for use as a background service on local computers, servers, and GPU systems.^[17]

LM Studio has expanded beyond its original desktop application to include additional software and services. LM Link allows models running on another computer to be accessed remotely,^[18] while the company's mobile software grew from its acquisition of Locally AI in 2026.^[19]

The company also develops Bionic, an AI agent for document creation, computer programming, workflow automation, and computer interaction that can use locally running models through the LM Studio runtime.^[20]

PCWorld included LM Studio among applications for running language models locally, highlighting its model selection, configuration controls, developer interface, and support for retrieval-augmented generation, while describing it as oriented more toward advanced users than beginners.^[21]

In a comparison with Ollama, Windows Central highlighted LM Studio's graphical interface and its support for GPU acceleration on systems using integrated AMD and Intel graphics, describing it as a unified environment for discovering, installing, and interacting with local language models.^[22]

LM Studio has also been studied in research on local AI software. A 2026 digital-forensics study compared LM Studio with Ollama and llama.cpp, examining artifacts left by the applications including model caches, configuration data, prompt histories, logs, and network activity.^[23]

• SGLang
• vLLM

• Official website
• LM Studio documentation