AX
![]()
AX, which its site says was born at Google, is an open-source runtime for running agent workloads at scale on your own cluster. You declare a Task (an isolated sandbox with CPU and memory limits), a Workspace (Git repositories, MCP servers and skill packages, or a plain-English goal that AX sets up before the task starts) and a Model in YAML manifests that will feel familiar to anyone who has used Kubernetes, and the ax CLI can suspend, resume and shell into a running agent. It runs on top of Agent Substrate. Its README warns that AX is in heavy development, with major breaking changes likely before a stable release, so "billions of tasks" per cluster is a design aim, not a measurement.
Was this useful?