Mission
Own your intelligence.
We're betting the future of AI is local — running 24/7 on hardware you own, in the rooms where you live and work. We build the hardware, the OS, and the agent runtime so you don't have to choose between capability and ownership.
Why now
Local just crossed the threshold.
A year ago, “local AI” meant a 7B model writing broken haikus.
Today, a 27B model running on consumer hardware can hit frontier-level reasoning on real agentic work. Models keep getting smaller and smarter at the same time. Tool-use reliability and long-horizon agent loops still favor frontier today — but the gap is closing in months, not years.
That milestone matters, but consumer hardware is a compromise. Laptops weren't built to run agents 24/7. Phones throttle. Desktops idle and burn power. The hardware that's technically capable wasn't designed for the job. We are.
Meanwhile, cloud AI is going the other direction. Bans. Silent capacity cuts. Billing changes you can't predict. Tokens only get more expensive. The frontier is becoming local — own it before you can't.
Vision
A mini data center for inference, in every home.
Mau lives where you live — at home, in the office, on the workshop floor. The box on the wall keeps your AI close: running, available, holding state across the day. Your phone is one way to reach it, not the place it lives.
Mau ships as a family of devices that form a private compute mesh. Ambient displays in your rooms. Larger compute bricks for the heavy lifts. Rack nodes for businesses. Every Mau you own shares compute with the others — your inference runs on hardware you bought.
As local capability scales, more of your AI workload moves home. The frontier still has its place — for the work that genuinely needs it. The rest runs on the box you own. Inference stops being something you rent.
The lineup
Early concept sketches. Final hardware will differ.
MauHub H1
Ambient display
The box on the wall. Runs agents 24/7 on a small, silent unit you can glance at.
MauHub C1
Compute brick
Sits beside the H1 for heavier workloads. Pools VRAM with everything else on your network.
MauHub R1
Rack node
Stack a few in a closet, run agents for a whole office. Same OS, same agent runtime.
How we build
Hardware and software, designed for agents.
The way we interface with computers hasn't fundamentally changed in fifty years. Mouse, keyboard, window, app. That works for humans. It doesn't work for agents.
We're building the stack from the ground up so agents are first-class citizens. The OS knows what's running. The hardware exposes the right surfaces. The runtime plans with a frontier model when it has to and does the work locally when it can — automatically, for free.
You shouldn't pay frontier prices for routine work. You shouldn't lose privacy or speed for a task your box can do offline. The system figures out which one to use.
What we believe
A short list.
- 01Local-first. Your model, your memory, your workflow. Always.
- 02Hybrid by default. Plan with the biggest model. Do the work for free, locally.
- 03Built for agents, not browsers. The interface assumes machines, not mice.
- 04Easy like Ubiquiti. Hardware and software shipped together so it just works.
- 05Yours to own. No banned accounts, no silent capacity cuts, no surprise pricing.
The future is local.
First batch ships to people who want to own their intelligence. Join the list.