Google's Open Agentic Orchestrator
Google open-sourced AX, a declarative orchestrator that sandboxes and scales agentic tasks to billions of concurrent sessions on its Agent Substrate runtime.
AX is an open, declarative control plane from Google for running agentic workloads, with tasks and workspaces defined in YAML (apiVersion ax.io/v1alpha1) and managed via a CLI (ax apply, ax watch, ax ssh, ax suspend/resume). Each task runs as a lightweight actor in a sandboxed workspace with fenced networking and git-backed environments, scaling to billions of concurrent agent sessions per cluster with sub-second suspend/resume and dense resource multiplexing. The platform adds generative features such as workspaces prepared from plain-English descriptions by an agent at first boot. AX builds on Google's Agent Substrate compute runtime and draws on agentic runtime research from Google DeepMind.
- Declarative YAML primitives define tasks, workspaces, network policies, and models for agent execution.
- Lightweight actors scale to billions of concurrent agent sessions with sub-second suspend, resume, and no cold start.
- Tasks run in sandboxed workspaces with fenced networks, git-backed repos, and interactive ax ssh debugging.
- Generative workspaces accept plain-English environment descriptions fulfilled by an agent on first boot.
- Built on Agent Substrate, targeting RL loops, trajectory collection, and agent evaluation at scale.
- Vendors
- GoogleGoogle DeepMind
- Products
- AXAgent Substrate
- Organizations
- GoogleGoogle DeepMind
From the original article
3 min read at agentexecutor.ioAX sandboxes your task, wires up its workspace, fences its network, and helps you run billions of them per cluster. Either use a single task per agent, or compose as many as your agent needs. goal: "Ensure that Go tool chain is available and is built from source"
By blazarquasar · agentexecutor.io
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.