Overview

AX is Google's innovative open-source agentic orchestration runtime, designed to manage and scale autonomous agent workloads with unparalleled efficiency. Built on top of Agent Substrate for secure, sandboxed execution, AX allows you to declaratively define agent tasks, complete with workspace and gateway specifications. It handles the complexities of sandboxing, network fencing, and scaling, enabling the deployment of billions of agent tasks within a cluster. If you're familiar with Kubernetes, AX will feel intuitively similar, offering a powerful new paradigm for agent-based applications.

Key Features

• Declarative Task Definition: Define agent tasks, workspaces, gateways, and models using simple ax.io/v1alpha1 manifests.
• Sandboxed Execution: Run untrusted agent code in isolated environments with configurable CPU and memory limits.
• Pre-wired Workspaces: Seamlessly integrate Git repositories, MCP servers, and skill packages, ensuring agents start with all necessary dependencies.
• Network Fencing: Securely control outbound network traffic by defining explicit host allowlists through Gateway specifications.
• Model Configuration: Easily configure the LLM platform used by AX, with secure credential management via Kubernetes secrets.
• Lifecycle Management: Pause and resume agent execution precisely where it left off using ax suspend and ax resume.
• Interactive Debugging: Gain deep visibility into running agents with ax ssh for interactive shells and command execution.
• Scalability: Designed to orchestrate billions of agent tasks within a cluster.
• Kubernetes Integration: Operates on Kubernetes, offering a familiar experience for users of container orchestration platforms.
• Developer-Friendly CLI: A kubectl-shaped CLI (ax) for intuitive management of tasks, workspaces, gateways, and models.

Typical Use Cases

• AI Agent Development & Deployment: Orchestrate and scale complex AI agents that require isolated environments, external tool access, and state management.
• Autonomous Workflows: Build and run sophisticated, multi-step autonomous workflows that can interact with external services and adapt based on accumulated state.
• Secure Code Execution: Safely run potentially untrusted code in a sandboxed environment with strict network controls, ideal for code generation or analysis tasks.
• LLM-Powered Applications: Deploy and manage applications that leverage Large Language Models, ensuring efficient resource utilization and controlled interactions.
• Large-Scale Agent Operations: Manage and monitor a vast number of concurrent agent instances for tasks such as data processing, automation, or simulation.