Why LLMX exists.

Modern networks demand line-rate forwarding, predictable control, and allocation-free parsing — all written in code that's safe to maintain. LLMX is a family of small, focused C++ projects that take those constraints seriously.

LLMX/router is the flagship: a line-rate L3 multipath eXchange for NVIDIA ConnectX NICs, built on DOCA Flow and DPDK, driven by a netlink control plane and exposed through an asynchronous gRPC API.

LLMX/rtaco and LLMX/vbvx are the sub-projects it stands on — coroutine-first RTNETLINK and zero-copy packet header views, respectively. Both are independently useful, and both are designed to combine into something larger.

The family is intentionally open-ended. New sub-projects slot into the same control-plane / data-path / parsing fabric, sharing conventions, tooling, and a single design language across every site under llmx.io.