AX, short for Agent Executor, is Google’s open-source orchestration runtime for running agentic workloads at scale. The project has more than 14,000 stars on GitHub and was under active development in 2026.
AX treats agents as a new kind of workload: not stateless microservices and not batch jobs, but stateful tasks that accumulate context and need strict isolation. You declare a task and its workspace in YAML — Kubernetes users will feel at home — including the git repositories to mount and the goal the agent should pursue. AX then sandboxes the task on top of the Agent Substrate project, wires up its workspace and runs it in a cluster, with a command-line workflow to apply definitions, watch progress and even SSH into a running sandbox to look over the agent’s shoulder.
The “Why” section of the project explains the motivation: agents accumulate state, run for unpredictable durations, need strict sandboxing and benefit from human oversight, none of which map cleanly onto microservices or batch jobs. AX answers with workspaces that pre-provision repositories and toolchains, tasks that declare a goal and a model specification, and an operator experience modeled on kubectl — apply a definition, watch the task, and SSH in when you want to debug or intervene. It is built to run agent workloads at very large scale per cluster, reflecting Google’s internal experience with fleets of coding agents.
AX is free and open source under the Apache-2.0 license. Note that Google flags the project as under heavy development, so breaking changes are likely before a stable release.
Platform engineers and agent infrastructure teams, especially Kubernetes shops that want a declarative way to run fleets of coding and research agents.