LM Studio

What it does

LM Studio downloads open language models such as Llama, Qwen, Mistral or DeepSeek and runs them on your computer – in GGUF format via llama.cpp, and on Apple Silicon Macs also in MLX format. In the chat you attach documents and ask questions about them offline. A local server makes the models available on your network through an OpenAI-like interface; there are also SDKs for Python and TypeScript, the lms command-line tool, a headless mode for servers and support for MCP servers. The Bionic agent edits documents and transcribes speech on the device.

Who it suits

Developers and teams who want to test language models without the cloud or build them into their own applications, and anyone who prefers not to send confidential texts to outside servers.

Data protection for businesses

Local models process chats and documents only on your device; according to the vendor, only anonymised data is collected for updates and model downloads. The paid cloud models and web search run through providers in the US without data retention, and inputs are not used for training. The vendor does not offer EU data hosting. The vendor does not mention a data processing agreement. Organisations get central billing and a quote on request; team subscriptions have been announced.

Limits

Speed and quality depend heavily on memory and graphics card. Local models often fall short of large cloud assistants. The app is not open source and may not be modified or redistributed.