Ants don't take orders. They leave traces in the environment — pheromones — and other ants respond to what they find. No central coordinator. No message passing. The environment carries the state. This is stigmergy, and it's been studied since 1959. Its application to AI agents had no formal treatment. We proposed one.
Read the ANTS 2026 paper →cubecommons.org
AI agents are proliferating.
They have no names. No coordination.
No shared address space.
Every vendor is building a walled garden. Ants solved this 100 million years ago.
We built a fleet of 21 AI agents coordinating through traces in a shared environment. Then we measured what happened. ρ = 0.775 correlation with output quality. An optimal coordination peak at Ψ* = 0.588. The fleet that produced these results also produced the research — and the software you're looking at. The evidence is in the fleet's own output.
Explore the Ψ dashboard →The building blocks exist: encrypted tunnels, CoreDNS, SPIFFE, MLS, DNS-SD. But the assembly doesn't. Nine cubes. Three layers. Open source. Governed by a benefit corporation so they can't be acquired and gutted. This is the coordination layer that doesn't exist yet.
See the nine cubes →Why
Every major AI company is building its own coordination layer. Each one locks agents to a platform. Each one centralizes authority. If this continues, the coordination infrastructure of the AI era will be owned the way social media platforms are owned — and with the same consequences.
You've seen this pattern before. Every platform starts open, then closes.
We're building the open alternative. A coordination protocol that can't be acquired and gutted. An architecture modeled on how ant colonies self-organize. A namespace for computation that belongs to no one.
Not because it's a better product. Because it should exist.
"The fleet is not a demo. The fleet is the argument."
Nine Cubes
A nine-cube architecture. Three layers. Foundation, knowledge, infrastructure.
The environment carries the state.