WasmEdge is a lightweight, high-performance WebAssembly runtime written in C++, optimized for server-side and edge workloads. It implements the core Wasm spec plus WASI, and layers on extensions for things the base spec does not cover: sockets for networking, a TensorFlow / PyTorch / llama.cpp inference plugin, a KV store interface, and async HTTP.
The runtime can execute Wasm in interpreted mode or ahead-of-time compiled to native via LLVM, which closes most of the performance gap with native binaries. It ships as an OCI-compatible runtime that plugs into containerd and CRI-O via crun or runwasi, so a Pod with a Wasm workload can be scheduled by Kubernetes exactly like a container. It also embeds into Rust, Go, Node.js, and Python processes as a library, which is how it is often used as a sandbox for user-defined functions.
WasmEdge is a CNCF sandbox project maintained by Second State, and is commonly used for serverless functions, edge compute, and running LLM inference at the edge.