🧩 SMG: Engine-Agnostic LLM Gateway

SMG is an engine-agnostic LLM gateway written in Rust. It provides full API compatibility for OpenAI and Anthropic, while connecting to various underlying inference frameworks like SGLang, vLLM, and TRT-LLM. The project features a gRPC pipeline, KV cache-aware routing, WASM plugins, and MCP support. The hard part is not building a simple API proxy, but managing efficient traffic routing and state across a complex, multi-model, multi-engine environment. SMG attempts to squeeze maximum performance out of inference clusters through low-level optimizations like KV cache-aware routing and tokenization caching. For teams managing large-scale LLM inference services, this type of high-performance middleware is becoming increasingly critical. It handles the friction of multi-model integration while offering a technically rigorous approach to performance optimization.