← Back to Discover

AX

Open-source runtime from Google that runs autonomous agent workloads on a cluster from declarative YAML manifests, with sandboxed tasks, pre-wired workspaces, suspend and resume, and SSH access.

Agent frameworksAutonomous agents
Visit website ↗

What it does

AX is a declarative orchestrator for running autonomous agent workloads in a cluster. You describe the work as YAML manifests and apply them with a single command-line tool, in a style that will feel familiar if you have used Kubernetes. Three kinds of manifest do the work. A Task runs agent code in an isolated sandbox with CPU and memory limits. A Workspace pre-wires Git repositories, MCP servers and skill packages so agents start ready. A Model sets which LLM the platform itself uses, with credentials taken from a Kubernetes secret. Commands let you apply, list and watch tasks, suspend and resume them, and open a shell inside a running sandbox when debugging is enabled. AX schedules every task on Agent Substrate, which must already be running in your Kubernetes cluster. Deploying it also needs Go, kubectl, ko and a container registry. The README warns that AX is in heavy development and that major breaking changes are likely before a stable release.

Capabilities

  • Declarative YAML manifests for Task, Workspace and Model
  • Sandboxed task execution with CPU and memory limits
  • Workspaces that pre-wire Git repos, MCP servers and skill packages
  • Suspend a task and resume it where it left off
  • Shell into a running sandbox for debugging

When to use it

  • You run many autonomous agents and need isolation and resource limits
  • You already operate Kubernetes and want agents managed with familiar manifests
  • You want to pause idle agents and pick their work up later
  • You are comfortable adopting an early-stage project that may change