LM Studio

LM Studio, built by Element Labs, Inc., is a desktop application and headless runtime that lets individuals and teams run open-weight large language models entirely on their own hardware — no cloud required. It supports macOS, Windows, and Linux, and is free for both home and work use under its published terms.

What It Is

LM Studio is a local AI inference platform that wraps llama.cpp (for GGUF models) and Apple MLX (on Apple Silicon) into a polished desktop GUI and a headless daemon called llmster. Users can browse, download, and run models like Llama, Qwen, DeepSeek, Gemma, Mistral, and OpenAI's gpt-oss directly from the app or via a CLI tool called lms. The platform also exposes an OpenAI-compatible REST API so existing apps and scripts can point at local models with minimal code changes.

Core Capabilities

• Model discovery and download via Hugging Face integration, directly inside the app
• Chat interface with document attachment (RAG) for fully offline document Q&A
• MCP client support — install and use Model Context Protocol servers with local models
• OpenAI-compatible API for drop-in compatibility with tools built for OpenAI endpoints
• Python SDK (pip install lmstudio) and JavaScript/TypeScript SDK (npm install @lmstudio/sdk) for programmatic access
• Headless mode (llmster) for server, CI, and cloud deployments without a GUI

Deployment Model

LM Studio covers two distinct deployment paths. The desktop app targets developers and knowledge workers who want a GUI-driven experience on their laptop or workstation. The llmster daemon targets server and CI environments where no display is available — it can be installed on Linux boxes or cloud VMs via a one-line shell script. Both paths share the same model management and API surface, so workflows built locally can be promoted to server deployments without rewriting integration code.

Developer and Integration Surface

The platform ships with first-party SDKs for Python and TypeScript, a CLI (lms) for chat, model downloads, daemon management, and server control, and documented integrations with tools like Codex, Claude Code, and OpenClaw. LM Studio also acts as an MCP client, letting developers connect external tool servers to local models. A "LM Link" feature, introduced as a new capability on the homepage, routes local AI workloads across devices so users can connect to remote LM Studio instances and use them as if they were local.

Update: Version 0.4.7 and llmster

The homepage download badge shows version 0.4.7 for Linux (x86_64) as the current release. The 0.4.0 release introduced llmster — the headless, no-GUI deployment mode — along with server and CI deployment support, which the changelog describes as "deploy on servers, deploy in CI, deploy anywhere." The product direction signal is clear: LM Studio is expanding beyond a desktop-only tool toward a full local inference stack usable in automated and server environments. The addition of LM Link for remote instance connectivity further extends the platform's reach beyond single-machine use.