AX

AX is a self-hosted AI agent orchestrator for teams running autonomous tasks on Kubernetes. It manages agents that retain state and need access to models, repositories and tools while they work.

• Sandboxed tasks: Run agent code in isolated environments with CPU and memory limits through Agent Substrate.
• Prepared workspaces: Give agents access to Git repositories, MCP servers and skill packages as part of their working environment.
• Pause and resume: Suspend an idle task and continue from its saved state.
• Task inspection: Watch task status and access a running agent's sandbox to inspect its work.

AX requires a Kubernetes cluster with Agent Substrate. Its command-line interface follows Kubernetes conventions and works with the active Kubernetes context, including cluster switching through kubectx. Declarative task, workspace and model definitions let teams describe workloads and their environments together.

The orchestration and sandboxes run in your cluster. Agents can send requests to external model APIs and tool servers; model credentials come from Kubernetes secrets. AX is open source under Apache 2.0. The project is under heavy development, and its core concepts and interfaces may undergo breaking changes.