| LM Studio | Windows / macOS / Linux | Graphical local model management and inference service, recommended first for desktop or workstation users. |
| vLLM | Linux / Docker | Aimed at high-throughput GPU inference, suitable for professional nodes that stay online long-term. |
| SGLang | Linux / Docker | Aimed at low-latency, high-concurrency inference, suitable for professional nodes that stay online long-term. |
| oMLX | macOS Apple Silicon | Specifically optimized for macOS MLX models, suitable for Apple Silicon Mac supply nodes. |
| llama.cpp | Windows / macOS / Linux | Lightweight GGUF inference service, suitable for local nodes running quantized models directly. |
| Ollama | Windows / macOS / Linux | The native /api compatible path is still usable; recommended when you already have an Ollama environment. |