LM Studio provides a desktop environment for exploring and running local language models.
LM Studio
Explore features, practical uses and pricing below.
LM Studio provides tools for downloading, running and connecting to language models on a user's own hardware. Its desktop interface supports model exploration and chat, while developer tools expose local inference to applications and scripts. The current product site also introduces Bionic, an agent application for work and code. It suits people evaluating open models, developers building local AI workflows and users who want to choose where inference runs.
The LM Studio application guide describes downloading models, managing chats and configuring local inference. It supports llama.cpp model workflows and MLX on supported Apple Silicon systems. Users can attach documents and ask questions about them rather than copying their contents into a separate hosted chat service.
The offline-operation guide distinguishes downloaded-model inference and document processing from actions that require connectivity, such as searching for models, downloading weights or fetching runtimes. Prepare the model and required runtime before expecting a disconnected workflow to work. Model files and available memory remain practical constraints even when no cloud API is involved.
The developer stack includes REST APIs, TypeScript and Python SDKs, and compatible endpoints for selected OpenAI- and Anthropic-style workflows. These provide a route from desktop experimentation to a local application or script. The llmster daemon supports headless operation, and the lms CLI manages tasks such as model downloads and server control.
Bionic is documented as a new, separate application that can be used alongside LM Studio. It organizes agent work into projects and sessions. Its coding tools can search a chosen local codebase, edit files and run shell tools. Document projects can also generate or revise files, supporting work beyond a chat response.
Bionic's model picker separates local, remote and cloud models. Local inference uses a downloaded model on the computer. Remote inference uses a model shared by another device through LM Link. Cloud inference uses hosted models and requires an internet connection, account and credits. Check the selected option before assuming that a session stays on the current machine.
For a local document experiment, download a suitable model, attach a sample policy and ask for an answer with supporting passages. Check the answer against the original text and repeat a few representative questions. A developer could then use the local API to test the same model in a small application. For Bionic coding work, start with a focused repository task and review its diff and test output before accepting changes.
Consult the system requirements for the platform and hardware in use. Memory, model size and context length affect what can run; model support for tools and images also differs. Running locally does not improve an inaccurate model's judgment or make a tool action harmless.
Bionic's document guide distinguishes managed project files from external files: editing an external file changes its original location. Keep recoverable copies when experimenting with file-changing tasks.
The pricing page describes free local access and optional paid Bionic cloud plans. The credit guide says local and LM Link models do not consume Bionic cloud credits; cloud usage depends on the model and tokens processed. Check current allowances and any model-license restrictions separately from the application's access terms.