Workers are the runtime processes that actually execute your agents, workflows, and functions. When you deploy, AGNT5 rolls your code out to Workers; when a workflow is invoked, a Worker picks it up, runs it against the durable runtime, and reports every step back to the journal.

You don’t provision Workers directly. The runtime manages their lifecycle, scaling, and recovery — if a Worker crashes mid-execution, another Worker picks up where it left off by replaying the journal to the last checkpoint. This is what makes durable execution possible without application-level retry code.

Workers are stateless. All durable state lives in the runtime journal and in State & Memory, so scaling a fleet of Workers up or down is a runtime concern, not a data-consistency one.

A deeper guide is in progress. For how Workers compose with the SDK today, see the Python SDK Worker docs.