AI that understands how your business works.

We are an emerging AI startup building enterprise ontology world models, Large Graphical Models and agent infrastructure. We also provide a tool ecosystem for independent developers worldwide and are building a global FDE service network, to close the last mile of AI delivery.

Explore projects Visit GitHub

01Real deliveries

18sectors

Customer deliveries in China, Singapore and the United States since 2024

Medical Care and Compassion in a Hospital Bed provided by Nova Ukraine — Healthcare

BRS2 Amazon fulfilment centre in Swindon — Robotics

Flo EV fast charging station at Hingham Depot — EV charging

UPS worker unloads boxes from semi trailer truck onto conveyor belt — Logistics

Johor Bahru–Singapore Rapid Transit System construction site — Construction

Switchyard of electric substation 750kV — Power grid

Tesla Motors Assembly Line — Manufacturing

Xujiahui Shanghai September 2023 — Property

02Environments

20scenarios

130+ scripted people, each with a role and a position

03Expert variants

10,000+expert hours

95problems

The same world, different facts behind the problem

04Review and anchors

reference ≥ 0.95 literal ask ≤ 0.2 shortcuts ≤ 0.8

0 · untouched1

270+strategies

Run on every grader, through rounds of expert review

05The benchmark

FDE-Gym

The last mile of AI delivery

13 models evaluated
4 brief levels, L0 to L3
For people and agents

Latest research

FDE-Gym, a benchmark of whether AI can deliver in real customer environments

Projects

Utopia

World's first enterprise world model

Turn business facts, relationships and rules into knowledge AI can reason with. Utopia tracks change through a bitemporal graph, detects conflicts and preserves provenance. Deploy fully offline on your own hardware.

Explore Utopia View source

IHigh-performance bitemporal knowledge graph
IIOntology reasoning and conflict detection
IIIOffline deployment with traceable sources

forkd

101 ms

High-performance sandbox

Fork 100 microVMs

Rapidly create isolated environments for parallel AI agents. Fork microVMs from a warm parent snapshot, sharing memory through copy-on-write while preserving KVM isolation for every instance.