1) Wallie can now HEAR.

It already saw your screen and reacted. Now it hears everything on your system music, videos, voices and reacts live, fused with what it sees. Two senses, one reaction.

And it actually understands music: mood, tempo, major/minor key, even the lyrics. Play a sad song and it feels the melancholy. A beat drops and it reacts to the energy. It'll call a track a banger or roast a muddy mix.

2) It's now way easier to try.

You voted. I ran it. Here's the clip

The winning comment:

"Make the AI a wildly overconfident conspiracy theorist playing Minecraft every normal thing gets an increasingly ridiculous theory. No breaking character."

So I typed that personality in and Wallie became it. Fully committed, zero breaks, the whole run:

Quick demo of Wallie actually doing its thing

It watches the screen and reacts live like a streamer gets jumped by a

skeleton, dreads creepers in the dark, swears it's "close to diamonds."

First-person personality, real-time voice, fully local.

Here's the deal: you decide what my AI does next, and I'll actually run it.

Pick anything:

- a game it should play Minecraft? a horror game? Elden Ring?

- a video it should react to

Thanks again to everyone who backed the launch you got us to #26!

Since then I shipped a performance update snappier, lower-latency reactions

and a smoother real-time pipeline, so Wallie feels more alive when it reacts

to your screen. Tagged it v1.0.0.

Human streamers sleep. Go offline. Have bad days. Rage quit.

An AI streamer doesn't.

I've been thinking about this a lot while building Wallie an open-source AI VTuber that runs entirely on your local machine. No cloud, no subscription, just your GPU keeping it alive 24/7 if you want.

What does streaming look like when the "person" behind the camera never burns out?