google/ax
Agent Executor (AX) is Google's open agentic orchestration runtime
What it solves
Agents are complex workloads that require more than standard microservices; they need isolated environments, persistent state, and the ability to pause and resume without losing progress. AX provides a declarative way to manage these agentic workloads at scale, handling the underlying infrastructure so you can focus on the agent's behavior.
How it works
AX uses a Kubernetes-like approach to manage agents. You define your requirements—such as Git repositories, model configurations, and task goals—in YAML manifests. AX then uses these manifests to provision sandboxed environments, wire up necessary tools, and manage the agent's lifecycle. You interact with these agents using a command-line interface that allows you to start, pause, resume, and even "ssh" into the agent's sandbox to inspect its activity.
Who it’s for
This project is for developers and engineers building autonomous agent systems who need a reliable, scalable way to deploy and manage millions of agent tasks across a cluster while maintaining strict isolation and observability.
Highlights
- Declarative management of agent tasks using YAML manifests.
- Built-in support for pausing and resuming agents to save resources.
- Secure, isolated sandboxing for untrusted agent code.
- Familiar command-line interface similar to kubectl for managing tasks, workspaces, and models.
- Ability to "ssh" into running agent environments for real-time debugging.
Related
- Dispatch
- Project
- Project
- Project
- Project