https://catmouse.ai

We were building something ordinary — a service that needed to read a handful of public product pages on a schedule. It worked for about a week. Then a request came back 429 with a wall of minified JavaScript attached, and everything stopped.

The received wisdom was to run a headless browser. So we did, and it was miserable: a fleet of Chromium processes eating memory, each one a slow, detectable, expensive way to answer a question that was ultimately just maths. Every rotation broke the automation harness. Every fix was a workaround stacked on a workaround.

So we did the other thing. We sat down with the challenge script and read it. It took a long time. Underneath the obfuscation was a small register virtual machine, a bytecode program, a permutation, and a set of probes measuring the browser it was running inside. None of it needed a browser to reproduce — it needed to be understood. Once it was understood, the answer took milliseconds to compute in native code and never opened a socket to a rendering engine.

That's the whole company. We do the reading once, carefully, and hand back the answer as a cookie jar. You keep the HTTP client you already have.