Building the Real World Web: Auki’s Long-Term Plan

Why We’re Obsessed With the Physical World

If you’re new: our mission is simple to say, hard to execute: We work on making the physical world accessible to AI.

More than two-thirds of the world economy is still tied to physical locations and physical labor. LLMs mostly serve white-collar work on the internet. To really grow the total addressable market of AI, we need physical AI.

We see three civilization-scale opportunities opening up this decade:

1. The real world web – a decentralized machine perception network / nervous system for AI
2. The app store for the real world – apps for robots, glasses and embodied AI
3. The OS for embodied AI – the operating system that runs on robots, glasses, drones, cars

Our strategy is to build these in the right order so we get a real shot at all three.

Step 1: The Real World Web (What the Auki Network Is)

Digital systems can’t “see” the physical world by default. When a robot visits a venue, it needs a way to browse that place, the way a browser hits a website.

We don’t believe all world data will sit in one company’s cloud. Just as all of the world’s websites are not stored on a single server, it’s very unlikely all physical data will live in one cloud.

Instead, each venue should self-host its own digital representation, and robots/glasses/phones discover and connect to those hyperlocal maps and compute.

That’s what we started building around 2021:

• The Auki network / posemesh
• A decentralized machine perception network
• An external sense of space for robots and AI

This is our real world web, and it’s the base layer for everything else.

Step 2: The App Store for the Real World

Robots won’t jump straight to Artificial General Physical Intelligence. Even with VLAs (Vision-Language-Action models):

• They can recognize a cup or a bottle cap
• But they still need policies and low-level control to actually move hands and bodies correctly
• Every robot body is different, so action data doesn’t generalize easily

So instead of one omnipotent robot brain, we think we’ll see an app store for robots, where you buy the ability to fold laundry, walk the dog, empty trash cans, etc.

To support that, general-purpose robots need six layers:

1. Locomotion
2. Manipulation
3. Spatial semantic perception
4. Mapping (object permanence in space)
5. Positioning (GPS doesn’t work indoors)
6. Applications to tie it all together

Most of the world is grinding on 1 and 2. We chose to start with 3–5 plus applications.

That’s why we can already ship AI copilots on phones and glasses: perception, mapping and positioning running on top of the real world web.

Our first big proof: Cactus, our spatial AI for retail, is already bringing in millions per year in ARR even just in handheld form. And with our next multi-million-dollar deal, a customer will run two apps (Cactus and Gotu) in the same domain on the Auki network – an early taste of the app store for physical spaces.

Step 3: Copilots Before Full Robots

We don’t want to wait until humanoids are perfect to create value. Glasses and phones are like robots that aren’t bothered with having arms and legs yet.

So the sequence is:

• Build positioning, mapping, perception first
• Ship AI co-pilots for physical labor on phones and glasses
• Prove value, generate revenue, and accumulate data and experience

Just like LLMs became co-pilots for engineers, doctors and lawyers, we expect physical co-pilots to be the dominant form factor for a long time. There are trillions to be earned in that era alone.

Steps 4–6: Robots, OS, and 100 Billion Devices

Once you have:

1. The real world web
2. An app store for the real world
3. Proven AI copilots at scale

…you’re in a strong position to:

4. Become the world’s largest robot distributor
   • Spaces are already prepared (mapped and positioned)
   • Robot OEMs plug into the same co-pilot logic
   • We help move 100–1000x more robots than any single hardware company
5. Build the OS for embodied AI
   • The OS that runs across robots, glasses, drones, cars
   • All speaking the same real-world web protocol
6. Become the first software layer on 100 billion devices
   • Each one burning tokens daily as it uses the network

We already partner with major Chinese robotics companies; starting next week, we’re kicking off partnership talks with leading US robotics companies too.

As Nils puts it, “If you just build the App Store for the real world, you’re huge. If you just build the OS for embodied AI, you’re huge. Our bet is that by doing things in the right order, we get a stab at all three.”

Reconstruction Node Open-Sourced

We’ve now open-sourced the reconstruction node on our GitHub.

What it does:

• Ingests captures from our DMT app (you walk a space with your phone for up to ~1 minute per clip)
• Runs heavy compute to turn those captures into accurate 3D models of the environment
• Will soon run on community nodes, not just our infrastructure

The flow:

1. Capture with DMT
2. Hit “refine” → jobs get distributed to community-run reconstruction nodes
3. Nodes output a 3D model that apps like Cactus (and others) can use

Right now:

• The code + deployment docs are public
• We’re onboarding a small beta set of node operators and builders
• You can already play with it if you’re comfortable running servers and want to earn bounties / help find bugs

Auki Data → NVIDIA Robot Navigation

We also showed a nice integration example: using NVIDIA’s Swagger tooling to generate a navigation mesh from Auki domain data.

• DMT + the reconstruction pipeline produced our demo store’s 3D map
• Our spatial semantic domain data was fed into Swagger
• Swagger produced a nav mesh robots can use to plan paths through the store

We’re now very close to an end-to-end pipeline: “Film a space with your phone, and any NVIDIA-enabled robot knows how to navigate it.”

Next Killer Demo: “Where’s My Yellow Rubber Duck?”

The next big demo we want to ship this year is one we’ve dreamed about for years: making a physical space queryable.

The idea:

1. Fill our office with cameras
2. Cameras stream images into a new vision node on the Auki network
3. The vision node detects objects (e.g. a yellow rubber duck, your hat, your keys)
4. A spatial compute node converts 2D detections into precise 3D coordinates
5. Your phone or glasses guide you in AR to the object

We want you to be able to ask, "Where did I leave my yellow rubber duck?" and get guided to it in AR.

Once that works, you can imagine the next step: asking Terri to go fetch it.

We think we’re only days of focused engineering away from a first internal version; the main constraint is time, not tech.

If you want the unfiltered version of these updates (and the bits we don’t put on X), join us in Discord.

‍

Watch the whole update on X.

About Auki

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