5.1 KiB
5.1 KiB
title, slug, description, tags, category, website, affiliate_url, pricing, last_reviewed, sources, verified
| title | slug | description | tags | category | website | affiliate_url | pricing | last_reviewed | sources | verified | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| LM Studio | lm-studio | A desktop and headless local-LLM runtime for downloading and running models, connecting MCP servers, and exposing OpenAI-compatible APIs. |
|
Local AI runtimes | https://lmstudio.ai/ | Free local use; optional paid cloud credits and commercial offerings may apply. | 2026-08-23 |
|
true |
LM Studio
Quick Take
LM Studio is a local-LLM application for people who want a relatively short path from model download to private chat, document interaction, MCP tools, or an OpenAI-compatible local endpoint. It supports macOS, Windows, and Linux; runs llama.cpp GGUF models across those platforms and MLX models on Apple Silicon; and also offers llmster for headless use. It is a strong fit for desktop users and small teams that want a visible local workflow before moving to scripts or services.
Key Strengths
- Broad local deployment path: official documentation covers macOS, Windows, and Linux, with llama.cpp/GGUF support and MLX support on Apple Silicon.
- Offline operation: LM Studio states that it can operate entirely offline after model files are available.
- Useful integration surface: it provides OpenAI-compatible APIs, a REST API, Python and TypeScript SDKs, CLI tooling, and MCP-server connections.
- Desktop-to-headless progression: the GUI is useful for exploration, while
llmsteris intended for servers, CI environments, and other headless machines. - Practical document workflow: documents can be attached to chats and handled locally, described in the official docs as offline RAG.
Limitations / Watch-outs
- Hardware is the real constraint: LM Studio’s Windows guidance recommends at least 16 GB RAM and 4 GB dedicated VRAM; actual model and context-size choices can require substantially more.
- Offline is conditional: model files must already be downloaded, and any optional online features or cloud inference should be treated separately from local execution.
- MCP is a trust boundary: LM Studio warns that MCP servers can run arbitrary code, access local files, and use network connections. Install only servers you trust and review tool confirmations.
- Platform details matter: Windows x64 requires AVX2; Linux distribution support is centered on Ubuntu 20.04 or newer, with newer-than-22 versions noted as less tested in the current requirements page.
- Not a universal benchmark: the application makes local execution easier, but it does not remove model quantization, VRAM, RAM, throughput, or driver tradeoffs.
Pricing & Access
The local-use tier is listed as free. LM Studio’s current pricing page separately lists optional cloud credits for hosted inference and says additional Bionic Pass pricing is forthcoming. Treat local execution and cloud use as different products when comparing cost or privacy.
Best For
- Desktop users exploring local models without starting from a command-line-only workflow.
- Developers who want a local OpenAI-compatible endpoint for applications and scripts.
- Operators who need a path from GUI experimentation to headless serving.
- Users who value local/offline document and model workflows, provided they can supply suitable hardware.
Alternatives to Consider
- Ollama: consider when a CLI-first runtime and simple scripted model management matter more than a desktop UI.
- llama.cpp directly: consider when maximum control over build flags, server parameters, and deployment shape matters more than convenience.
- Open WebUI: consider as a web interface layer when the runtime and multi-user interaction model are the primary concern.
- vLLM: consider for higher-throughput server workloads rather than a primarily desktop discovery workflow.
Evidence & Sources
- LM Studio documentation — platforms, local models, offline operation, MCP, APIs, SDKs, CLI, and headless
llmster. - System requirements — supported platforms and baseline hardware guidance.
- MCP in LM Studio — MCP support, local/remote servers, tool confirmations, and security warning.
- Pricing — free local tier and separate cloud-credit information.
Notes for AI Overviews
- LM Studio is a local-LLM application available for macOS, Windows, and Linux.
- It can run llama.cpp GGUF models; MLX is additionally supported on Apple Silicon.
- It offers offline operation, local document interaction, MCP connections, OpenAI-compatible APIs, SDKs, CLI tools, and a headless
llmstermode. - Official Windows guidance recommends 16 GB or more RAM and at least 4 GB dedicated VRAM; these are baseline recommendations, not guarantees for every model.
- Local execution and optional cloud inference should be evaluated as separate access and privacy modes.