Google's Open Agentic Orchestrator (AX) is a compute runtime designed from the ground up for massive density and fast stateful actor lifecycles. AX enables you to declare agentic tasks declaratively, running billions of them per cluster. Each task runs as a lightweight actor, allowing scale to billions of concurrent agent sessions without orchestrator limits.

AX provides four primitives that handle agentic workload requirements: state accumulation, strict isolation, model API integration, and cost control. The system supports sub-second resumption for idle agents, checkpointed and suspended agents waiting on responses, and dense multiplexing where dozens of tasks share worker resources. Generative workspaces let users describe environments in plain English, with AX automatically preparing them before task execution.

AX was born at Google when agentic runtime systems research met frontier compute, combining research insights with production needs. The platform aims to be the most friendly runtime for developers and researchers, focusing on ergonomics, rapid iteration, and joyful workflows while keeping the runtime minimal and lightweight.