Compile plain JS to optimized WASM.
For DSP, math, visuals and more.
Demo compiling in your browser…
JZ compiles a distilled JavaScript subset, Crockford’s “good parts” – into native-grade WASM.
Valid JZ is valid JS. Same source runs as plain JS or compiles without rewrite to .wasm – sandboxed, portable, memory-safe. No runtime, no GC, no type annotations, auto-SIMD, deterministic output.
| Good for | Not for |
|---|---|
| DSP, audio, synthesis | UI, DOM, frontend state |
| Images, video, pixels | Network, hot I/O, serving HTTP |
| Simulation, physics, games | Dynamic object models and monkey-patching |
| Parsers, codecs, compression | Allocation-heavy, long-lived object graphs |
| Scientific, numeric, edge ML | Security-sensitive cryptography and arbitrary-precision integers |
| Hashing, checksums, RNG | Tiny calls where the JS/WASM boundary dominates |
A minimal modern functional JS subset – no dynamic constructs or reflection. The jzify layer lowers legacy patterns, generators, and async/await.
let x = 0.5, Float32Array, an array index, a loop counter. JZ infers it instead of turning the file into another language. Ambiguous values take a slower, always-correct dynamic path.Porffor aims to cover the full JS spec but emits no WASM (its 2026 rewrite compiles through C to native) and runs slower than V8 on the cases it completes; scriptc makes native CLI binaries from TypeScript, no WASM target; AssemblyScript compiles a typed TypeScript dialect to WASM.
JZ compiles plain JS that still runs and tests as JS, and the release gate keeps its WASM fastest per case; the price is the overdynamic constructs it refuses.
Rust, Go, and C are excellent when a rewrite and second toolchain make sense. JZ is for the hot path you want to keep in JavaScript: one source, one test suite, a JS fallback, and comparable WASM or native performance.
JZ leads V8 and AssemblyScript by geomean on the covered corpus and targets near-native speed. The release gate is stricter than an average: JZ must be the fastest WASM on every case. Per-case numbers, missing target coverage, and every measured loss stay visible on the bench page; a rival win remains a bug to close, not an exception to hide.
If you hit a slow case, report it →
WASM isn't magically faster – V8 also JIT-compiles hot JS to machine code. But a JIT pays for staying ready for any JS: shape checks, deopt guards, GC pauses in the hot loop. JZ's subset is closed, so it commits ahead of time to what a JIT only gambles on – static layout, monomorphic calls, no GC – and auto-vectorizes loops to SIMD, which V8 doesn't attempt on plain JS.
jz → wasm2c → clang -O3 produces a standalone native binary. See full pipeline →.
jz.wasm. CI checks the wasm-hosted compiler against the test suite. 
It’s experimental (pre-1.0) – ABI can still shift.
But real code already ships on it: color-space v3 builds its precompiled 27-space WASM backend with JZ.
CI gates the full suite, test262, the bench, and the self-compile. A fuzzer diffs every kernel against the same source in V8.
And the exit is free: valid JZ is valid JS – drop the compile step and the same file keeps running.
jz kernel.js) or the API (compile()) produces a plain .wasm you load like any other module. Setup & options →Find it useful?
Sponsor the dev
