google/ax
Declarative orchestrator for running large numbers of autonomous agent workloads on a Kubernetes cluster. Tasks, Workspaces and Models are written as YAML manifests and applied with a kubectl-shaped CLI, which sandboxes each task, pre-wires git repositories and MCP servers, and supports suspending, resuming and shelling into running agents. Tasks execute as sandboxed actors on Agent Substrate with CPU and memory limits; the control plane is written in Go and deployed into the ax-system namespace.
Maintenance activity