A small studio
in Prague that refused
to ship slow software.

Why we built this

We kept running into the same wall. Medical PACS choking on 4K studies. VR teams shipping pyramids of pre-baked tiles because real-time JPEG 2000 was "impossible". Broadcast engineers paying a fortune for AJA cards to do what a Mac Studio should do in software.

The codec wasn't broken. The implementations were. So we sat down and rebuilt the whole pipeline for one architecture — Apple Silicon — and stopped pretending we cared about anything else.

The bet we made

One platform. No portability tax. Every line of MetaWave assumes unified memory, Metal, NEON, and the Neural Engine. If you run anything other than an M-series Mac, we are politely the wrong product.

• Metal compute kernels — wavelet transforms written for tile memory, not retrofitted from CUDA.
• Zero-copy unified memory — the GPU reads the same bytes the CPU just wrote. No round-trips.
• NEON entropy coder — hand-tuned per chip generation, M1 through M5.
• Neural Engine off-load — color transforms and resampling, where it makes sense and not where it doesn't.

Who works on it

A few of us. Codec people, Apple platform people, one stubborn radiology consultant who keeps us honest about clinical workflows. We don't have a sales team. The person who replies to your email is the person writing the code.

What we won't do

• Port to Windows or Linux. Not now, not later.
• Build a "cloud" version. The whole point is local performance.
• Ship features we haven't measured ourselves on real hardware.
• Pretend a benchmark we can't reproduce.

Where we are

Prague, Czech Republic. We meet around a kitchen table more often than around a conference table. The closest thing we have to a marketing department is the GitHub commit log.