G-Micro is a 110-million-parameter language model built from nothing: the tokenizer, the training loop and the Polish corpus are all ours. This is the app you talk to it through.

The page itself computes nothing. It writes a task into a database, a MacBook picks it up, runs the model and streams the answer back token by token. Both sides dial out, so no port is ever opened on the Mac, deliberately chosen over a tunnel.

The same window also edits photos with G-Images, a second model of ours. Ask for black and white or a blur and it makes the change directly, with no service in the middle.