PAT. Your pocket audio tech.
Talk to it like a sound engineer. “Make the reverb stronger.” It listens, then dials it in.
Concept render
Just say it.
No menus, no presets to dig through. Hold the button, describe the sound in your own words, and PAT makes the change while you keep playing.
- I want to sound like John Mayer.
- Make the reverb stronger.
- It's too bright on the high notes.
- It squeals when I turn up.
- Actually, undo that.
Concept render. Screen content is illustrative.
Try PAT.
Press play, then ask for a change. You'll see which skill PAT picks, the exact tool calls it writes, and the dials move as the sound changes.
Simulated in your browser. On PAT, an on-device LLM makes these calls.
The guitar is a short clean loop, synthesised in your browser. Headphones help.
Pick a request above. PAT chooses a skill, writes tool calls, and the dials move as each one is applied.
On-device. Offline. Private.
The language model, the speech recognition and the audio engine all run inside PAT. No cloud round trip, so nothing waits on a network and nothing you play or say is uploaded.
Round trip from guitar in to amp out, with every effect running.
Of signal needed. It works in a basement, on a plane or in the backyard.
Your playing and your voice stay on the device.
Target specifications for the first hardware. Not yet measured on a prototype.
Built in Rust, from the kernel up.
PAT OS: a real-time Linux kernel and an all-Rust userland, so the AI can never make your sound stutter.
How PAT worksDesigned to disappear into your pocket.
One knob, one button, one voice. Everything else is a conversation.
See the designHelp us build it.
PAT is at the concept stage. We're looking for investors who want to back on-device AI hardware for musicians.