Concepts

Four ideas carry the whole engine. Everything else in these docs is a consequence of one of them.

Durability is replay

A run survives a crash because the engine replays the workflow body from the top and feeds already-completed steps their saved results instead of executing them again. That single mechanism is what buys you crash recovery, and it is also the one rule it imposes: the body must be deterministic — no Date.now(), Math.random(), or direct I/O outside a step. See Durability & replay.

Workflows orchestrate, steps do

A @Workflow decides what happens next; a @Step is the only place a side effect is allowed to happen. ctx.step is the single dispatched primitive — it checkpoints, routes to a worker, retries on its own policy, and returns the recorded result on the next replay. Everything else (fan-out, child workflows, entities) composes on top of it. See Workflows & steps.

Waiting costs nothing

A run parked on ctx.sleep or ctx.waitForSignal holds no process, no thread, and no connection. Its state is a row and a wake time; the engine resumes it when the timer is due or the signal arrives. A month-long wait and a one-second wait cost the same while parked. See Sleep & signals.

Runs belong to someone

Every run is stamped with a namespace, and that stamp decides which pool drives it and which workers execute it. It is what lets a shared broker host a deployed cluster and a developer's laptop at once without either stealing the other's work. See Tenancy, and Topologies for the deployment shapes it rides on.