
Auki Labs
Auki Labs builds the posemesh, a decentralized spatial network whose open source nodes reconstruct physical spaces as Gaussian splats for robots and AI.
Location
Hong Kong
Hiring?
Platform
Decentralized network with self-hosted Docker nodes on Windows and Ubuntu, an iOS capture app and a browser based domain viewer
GPU
NVIDIA GPU with 8 GiB or more of VRAM to run a reconstruction or splatter node, tested on RTX 3090, RTX 4060 and T4. None stated for the browser viewer, which renders splats through Spark on WebGL2
Pricing
Node and viewer code is free under MIT. Network jobs are paid in credits, and Auki priced a splat at 0.2 credits, about 20 cents, per scan when the splatter node launched in March 2026. Cactus pricing is available on request
License
MIT for the reconstruction node, splatter node and domain viewer. Auki's own services and apps are proprietary
Compare licence terms ↗
Best for
Shared spatial maps of physical spaces for robots, wearables and phones
Updated
July 2026
Auki Labs is the company behind the posemesh, also called the Auki network, which it describes as a decentralized machine perception network and collaborative spatial computing protocol. Founder and CEO Nils Pihl leads the company, which runs a lab in Hong Kong. The network lets robots, wearables and phones exchange spatial data inside a domain, a persistent 3D coordinate system anchored to a physical space by portal QR codes. Auki’s flagship product on top of it is Cactus, a spatial AI platform for retail, alongside Gotu for indoor navigation at events and properties.
Gaussian splatting enters through the reconstruction pipeline. Scans recorded with the Domain Management Tool app go to a reconstruction node, where a custom structure from motion pipeline refines the camera poses and creates point clouds, bounding boxes and Gaussian splats. A separate splatter node trains the 3D Gaussian splat and starts from the reconstruction node’s refined poses, which Auki says makes training more robust to challenging indoor environments and noisy captures. The whitepaper describes the splat as serving both photorealistic rendering and a spatial representation compatible with AI and robotics.
Both nodes are MIT licensed and run in Docker on Windows 10 or 11 or Ubuntu 22.04 or 24.04 with an NVIDIA GPU carrying at least 8 GiB of VRAM. Auki open sourced the splatter node in March 2026 and launched it on the network later that month. Operators register a compute node in the Posemesh Console and stake AUKI tokens for it. At launch Auki put the cost of turning a scan into a splat at 0.2 credits, about 20 cents, and the cost of reconstructing a supermarket at around $10.
Results are viewed in the Auki Domain Viewer, an MIT licensed web application built on Three.js and React Three Fiber that shows point clouds, navigation and occlusion meshes, portal locations and splats. Splat rendering goes through Spark with dynamic loading and partitioned level of detail. Auki reported in February 2026 that loading low and high detail layers separately lets it render entire supermarkets in a browser in real time. In May 2026 it reported a depth regularization feature that makes the Gaussians adhere more closely to the geometry of the environment.
The capture flow and reconstruction in the Domain Management Tool are supported on iOS only, and Auki lists the Android version of the app as returning soon. Developers and domain owners pay for network services by burning the AUKI utility token, while infrastructure operators receive token rewards. Cactus has no published price and is sold through a pilot request.
Frequently Asked Questions
How does Auki use Gaussian splatting?
Auki uses Gaussian splatting to build photorealistic 3D representations of the physical spaces mapped on its network. A reconstruction node refines the camera poses of Domain Management Tool scans, and a splatter node then trains a 3D Gaussian splat that the web based Domain Viewer can display.
Is Auki’s splatting software open source?
Auki publishes its reconstruction node, splatter node and Domain Viewer on GitHub under the MIT License. The splatter node credits nerfstudio among the projects it builds on, and the viewer renders splats with Spark.
What hardware does an Auki splatter node need?
An Auki splatter node needs a 64 bit Windows 10 or 11 or Ubuntu 22.04 or 24.04 machine with Docker, 4 CPU cores, 12 GiB of RAM and an NVIDIA GPU with at least 8 GiB of VRAM. Auki has tested it on the RTX 3090, RTX 4060 and T4 and lists the RTX 50 series as not currently supported.
How much does a splat cost on the Auki network?
Auki put the price of turning one scan into a splat at 0.2 credits, about 20 cents, when it launched the splatter node in March 2026. The company put the cost of reconstructing a supermarket at around $10.
What is the posemesh?
The posemesh is the decentralized machine perception network that Auki Labs is building, also called the Auki network. It lets devices privately exchange spatial data and computing power with the domain they are visiting instead of sharing a camera feed with a central provider.
Alternatives
Recent Headlines

11:38
Watch · New every Monday
Splats Come to Fortnite, Nike Shoots in 4D, Splats on a Nintendo 3DS
Episode 12 · Nov 9, 2026
Recent Job Openings


