Every unit of work is a topology — a graph of role-slots bound to real AI agents. Compose and run agentic systems from a single claw up to a whole org, on a durable, crash-resumable runner.
Each rung composes the one below — a company is staffed with teams, an org with companies. Pick a scale; we instantiate the topology and bind it to real, chattable claws.
Classify a task, build the structure that fits it, and switch patterns without rewriting a thing — authority stays invariant across every one.
Executing a node runs its entire sub-topology, all the way down to the leaf claws — on a durable runner that checkpoints every step.
Get a quality/cost Pareto front, then let a MAP-Elites search evolve better team configurations using the comparison harness as fitness.
Mix backends across a single topology — a cheap model for fan-out, a strong one for the judge.
No choice of structure — and no switch between structures — can let an agent exceed its sandbox. The network segmentation is the security model.
A single-node Docker Compose deployment runs everything with the same network-segmented security model as the Kubernetes path. The server self-migrates on boot.
From one agent to a whole org — same model, same guarantees.