A Node.js HTTP framework that answers requests in C++ sounds contradictory. Velociradix does exactly that — and the design decisions behind it are the interesting part.
The Problem With Express
JavaScript frameworks are elegant but slow. Every request in Express flows through JavaScript: object allocation, callback dispatch, property lookups. For a JSON API with a database behind it, nobody notices. For a static asset server or a health-check endpoint, you are spending microseconds on JSON dispatch you never asked for.
The Hybrid Architecture
Velociradix splits its response path in two:
fastGet/fastPost— static routes answered entirely in C++. The bytes are written by the native addon; V8 never runs.app.get/ middleware /ctx.json— dynamic routes with JavaScript business logic.
The native core uses kqueue (macOS) or epoll (Linux), spawns workers with SO_REUSEPORT, and routes requests through a radix-trie router compiled into the addon.
The Custom HTTP Parser
The most controversial decision: a hand-written HTTP/1.1 parser in C++ instead of llhttp.
This is not performance theater. Writing your own parser is a trust decision. llhttp is a C library that has to translate into V8's world anyway. By keeping parsing in the C++ layer, Velociradix validates and routes everything before a single object enters JavaScript — and you can read the entire parsing logic instead of trusting black-box bindings.
What Fast Actually Means
On static routes, the C++ engine writes the response envelope — status line, headers, body — in one pass. The JavaScript runtime is bypassed for the hot path, which is where "fast" stops being a benchmark and starts being infrastructure: an uptime probe that responds in microseconds instead of milliseconds.
The Lesson
The biggest win was architectural: you do not have to choose between ergonomics and speed. A thin JavaScript API over a compiled core gives you the best of both. JPROT, AeroCI, and Velociforge all follow this same pattern — the JS side is the ergonomics, the C++ or compact format side is the performance.
It proves that the most interesting engineering lives at the boundaries between layers.