Answer
Every model that wavecat uses runs locally, on your own computer, through llama.cpp. There is no cloud inference anywhere in the loop. From the moment you launch wavecat, the models that read your screen and power the chat all execute on-device.
Because nothing is sent to a remote server for inference, wavecat keeps working even with your Wi-Fi turned off. Your screen — and everything wavecat learns from it — stays on your machine.
If you’d prefer, you can also connect your own model for the heavier interactive work. But that’s an advanced and in-developemtn option — the default experience is fully local and needs no setup.