LM Studio

LM Studio has become the default on-ramp for running open-weight LLMs locally, pairing a genuinely friendly desktop GUI with serious developer infrastructure — an OpenAI-compatible API, SDKs, a CLI, and now a headless runtime (llmster) for server and CI use. It handles the tricky parts of local inference (quantization, GPU offload, model discovery) well enough that both hobbyists and engineers building privacy-sensitive apps can get a model running in minutes, and it costs nothing. The tradeoffs are inherent to local inference rather than to the app itself: you're capped by your own hardware, and there's no training or fine-tuning story here, just fast, private inference. For anyone who wants Llama, Gemma, Qwen, or DeepSeek running on their own machine without wrestling with raw llama.cpp builds, LM Studio is the easiest path there.