A desktop app from Element Labs for running open models on your own PC. It supports Mac, Windows, and Linux, and ships with two inference engines: MLX for Apple Silicon and llama.cpp for everything else. You can search for models, download them, and start chatting entirely through the GUI, which makes local LLMs approachable even if you’re not comfortable on the command line. Its RAG feature for querying your own documents also runs fully offline, so it suits work where you’d rather not send data anywhere. An OpenAI-compatible API server, the lms CLI, and MIT-licensed Python and TypeScript SDKs round it out, and it has become a common choice for development work as well. Since July 2025, commercial use at work has been free too.
Key Features
- A complete local LLM environment in a GUI: Search for open models on Hugging Face (GGUF or MLX format), download them in-app, and chat right away. The app shows whether a given model is likely to run on your machine, so picking one is less of a guessing game.
- Two built-in engines, MLX and llama.cpp: It automatically selects the right inference engine for your setup — MLX on Apple Silicon, llama.cpp elsewhere. GPU offload and quantization levels are adjustable from the GUI.
- Fully offline document chat (RAG): Load your own PDFs, docx files, and similar, then ask questions about them. Everything is processed locally, and document contents never leave your machine.
- OpenAI-compatible API server: The local server speaks the same format as the OpenAI SDK, so you can try local models by swapping the endpoint in an existing app.
- The
lmsCLI plus Python and TypeScript SDKs: Control model loading and server startup from scripts. The SDKs are published under the MIT license and can be used to build apps around local LLMs. - MCP server support: Add Model Context Protocol (MCP) servers to give local models access to external tools.
- Bionic, an AI agent, plus cloud inference: Bionic, an AI agent feature, launched on July 16, 2026 and is available even on the free plan. When your own machine isn’t enough, you can supplement it with zero-data-retention cloud inference billed per token.
Pricing
Pricing was revised alongside the July 16, 2026 launch of the Bionic AI agent feature, pairing a free tier for local use with pay-as-you-go cloud inference.
| Plan | Price | Key Features |
|---|---|---|
| Free | $0 | Every local feature (Bionic agent, llama.cpp / MLX, offline speech transcription, ZDR web search, up to 5 LM Link devices). Free for both personal and commercial use at work (since July 2025) |
| Cloud pay-as-you-go | Per token | Cloud inference with zero data retention (ZDR). Rates vary by model (for example, DeepSeek V4 Flash is $0.13 input / $0.26 output per 1M tokens, and Kimi K3 is $3.00 input / $15.00 output per 1M tokens) |
| Bionic Pass | Not announced | Listed as “coming soon” on the official page; neither price nor contents have been disclosed |
| Enterprise | Contact sales | Aimed at organizations. Details available only through the sales contact flow |
Pricing is current as of August 21, 2026. Check the official site for the latest rates.
Pros and Cons
✅ Pros
- Everything from finding a model to chatting with it happens in the GUI, which lowers the barrier to entry for local LLMs
- Your data never leaves the device, making it a comfortable fit for confidential documents and personal information
- Once a model is downloaded it runs offline, with no API charges
- The OpenAI-compatible API, CLI, and SDKs cover the path from prototype to production
- Commercial use at work is free, so there’s little license paperwork when installing it on a company PC
⚠️ Cons
- Running at a practical speed requires a reasonably capable machine (memory and GPU)
- Only open models will run; cloud-only models such as GPT and Claude are out of reach (cloud inference is likewise limited to open models like DeepSeek, GLM, and Kimi)
- Large models can weigh tens of gigabytes and eat into your storage
- The app itself isn’t open source (though the SDKs and CLI are published under the MIT license)
Comparison with Similar Tools
| Criterion | LM Studio | Ollama | Jan | GPT4All |
|---|---|---|---|---|
| Primary interface | GUI | CLI (GUI available) | GUI | GUI |
| Supported OS | Mac / Windows / Linux | Mac / Windows / Linux | Mac / Windows / Linux | Mac / Windows / Linux |
| OpenAI-compatible API | Yes | Yes | Yes | Yes |
| Open-source app | No (SDKs are MIT) | Open source | Open source | Open source |
| Notable for | MLX support, document RAG, MCP | Lightweight, developer-oriented, rich ecosystem | Extensions for added functionality | Focus on local document chat |
Who It’s For
- Beginners who want to try a local LLM without touching the command line
- Anyone handling confidential documents or personal information who needs AI without sending data outside
- People who want to experiment freely with open models without watching an API bill
- Developers building apps or prototypes against local models through an OpenAI-compatible API
- Anyone who wants to run a local LLM on a company machine without licensing worries
Conclusion
LM Studio is a desktop app that has meaningfully lowered the barrier to local LLMs. Its strength lies in combining the convenience of doing everything — obtaining models, chatting, document RAG — from a GUI with a serious developer toolkit in the form of an OpenAI-compatible API, a CLI, and SDKs. Practical use does assume a reasonably capable machine, but for anyone who wants to use AI while keeping their data close, it’s the first option worth trying. Local execution is free for both personal and workplace use, so if you have a compatible machine, it’s an easy one to install and explore. And when your machine falls short, you can top it up with pay-as-you-go cloud inference.