Capability · Framework — fine-tuning

LM Studio

LM Studio is a free (for personal and approved work use) desktop application that makes running local LLMs feel like using the App Store. It integrates Hugging Face model search, automatic GGUF / MLX backend selection, an OpenAI-compatible REST server, structured-output support, and a headless `lms` CLI for scripted deployments. Widely used by developers who want a GUI on top of llama.cpp without the CLI.

Framework facts

Category: fine-tuning
Language: TypeScript / C++
License: Proprietary (free for personal and some commercial use)

Install

# Download the installer from https://lmstudio.ai
# Or via Homebrew cask:
brew install --cask lm-studio

Quickstart

# From the app: Developer → Start Server, then:
curl http://localhost:1234/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model": "local-model", "messages": [{"role":"user","content":"hello"}]}'

# Or headless
lms server start && lms load llama-3.2-3b-instruct

Alternatives

Jan — open-source alternative
Ollama — CLI-first
llamafile — single-binary

Frequently asked questions

Is LM Studio open source?

No — the app is proprietary, though the inference engines it wraps (llama.cpp, MLX) are open. A free license is available for personal and some commercial use; check their Terms for details.

Does LM Studio support Apple Silicon acceleration?

Yes — it auto-selects Metal for GGUF models and can use Apple MLX for native quantised models, typically giving 2-4x speedups on M-series chips over CPU.

Sources

LM Studio docs — accessed 2026-04-20
LM Studio home — accessed 2026-04-20