LM Studio

LM Studio solved the problem that kept local AI a hobbyist niche: making it approachable. Where the llama.cpp ecosystem assumes you enjoy compiling flags, LM Studio wraps model discovery, download, quantization choices and chat into a desktop app your non-terminal colleagues can actually use — then goes deeper than most GUIs dare, with an OpenAI-compatible local server, the lms CLI, SDKs, local RAG over your own documents, and an MCP client that gained OAuth in April 2026. The dual-engine design is quietly its best trick: GGUF via llama.cpp on PCs, Apple MLX on Macs, so an M-series MacBook performs like the first-class citizen it should be. Since July 2025 the whole thing is free even for commercial use, which removed the last awkward asterisk. The honest limits are physics and philosophy. Physics: local inference lives and dies by your VRAM — a 70B model on a thin laptop is a slideshow, and no software fixes that. Philosophy: the GUI is closed-source, which rubs part of the local-AI crowd the wrong way, and the new Bionic companion app — helpful as its zero-data-retention cloud inference may be — reintroduces exactly the cloud dependency many users came here to escape. Treat those as footnotes, not dealbreakers. If you want a terminal-native, scriptable runner, Ollama remains the purist's pick; for everyone else who wants private, capable AI running on their own metal by this afternoon, LM Studio is the recommendation we give first.