Join the movement and participate on Uniswap.
Participate
Last week Nils laid out the open Auki SDK and Cactus as a harness for physical AI. Today we showed it on the floor.
A BracketBot localized in a remodeled lab it had never seen, on a map the phones had already built. Enrolling the next robot is a QR code. Arshak put store cameras into that same twin — pose, gaze, and a heatmap of what people actually look at.
If you follow us on X, you've already seen the clip: a BracketBot navigating a map it did not create.
Two densities of point cloud. Close to the white ball, thick — the robot's own sensors filling in what it can see. Further out, thinner — the map the phones already walked. The robot uses that phone map to know where it is and to go places it has never been, and it adds to the map with better sensors.
We pulled a longer take from internal demo day. The robot starts in a previous version of the lab. We've remodeled the space and remapped it with phones. As far as the BracketBot is concerned, this is a different building. It connects, pulls the new map, and looks for its bearings.
Phil put it in manual drive and turned it so it was looking into the map. That was enough to lock position — the white dot. It fills in what it can see, drives the space, and still navigates to a click on a part of the map it has never visited.
Retailers already run Cactus as a copilot on phones. They pay as much as $500 a month per location. That business is already a couple of million dollars a year.
If a site is on Cactus, getting a robot onto that same network is showing it a QR code. Customer-specific interface, devices, servers, components — the robot joins the domain the phones already made.
Getting a robot enrolled is as easy as showing it a QR code.
We think that's the fastest way to scale robotics. Nils's number off the existing customer base: 1,000–3,000 robots next year. First pilots early next year. Units ship from China and the US in November, arriving late December / early January.
We also showed the armed BracketBot scanning a shelf — price tags, whether the bay matches the plan. We got the armed version last week. Each arm rides its own elevator, not a shared torso lift.
Arshak used the Auki SDK to talk to security cameras and run pose estimation on humans and humanoids in the store.
A small robot in frame gets detected. In the map the phones and robots already maintain, you see where people are and where they're looking — a flash on the shelf when a head turns. Then a heatmap of dwell: this hour, this day, this month. The product map is live, so you can tie that attention to SKUs, and then to sales.
Same twin the staff use in AR and the robots use to move. If someone stands in front of a bay for a long time, a robot can drive over and ask what they're looking for.
The merchandising version: put the things people will hunt for in the parts of the store nobody looks at, and those empty stretches shrink.
We started designing this years ago. The SDK is far enough along, and the AI tooling is good enough, that it came together quickly.
Arshak on the stack:
Most of what this used on our side is already open source. The pose-detection piece — same shape as QR Lab — lands soon as Pose Lab on GitHub. We'll tell you when the repo is up.
We've tried to make the components quite understandable to AI, so it's easy to vibe code with these components.
Auki is making the physical world accessible to AI by building the real world web: a way for robots and digital devices like smart glasses and phones to browse, navigate, and search physical locations.
70% of the world economy is still tied to physical locations and labor, so making the physical world accessible to AI represents a 3X increase in the TAM of AI in general. Auki's goal is to become the decentralized nervous system of AI in the physical world, providing collaborative spatial reasoning for the next 100bn devices on Earth and beyond.
X | Discord | LinkedIn | YouTube | Whitepaper | auki.com
