LM Studio

What it does

LM Studio puts a desktop face on llama.cpp: search a model, download a quantised build, start chatting. It also exposes a local OpenAI-compatible endpoint, so a prototype can point at localhost and later move to a hosted model without rewriting code.

Quality depends entirely on your RAM and GPU, and small quantised models are noticeably worse at nuance. For translation drafts, extraction and classification of confidential documents, keeping data on the machine is often the whole point.

Standout features

• Model browser with quantised builds
• OpenAI-compatible local API
• Works fully offline
• Apple Silicon and CUDA acceleration

Best for