#Events and Recovery

#Raw event listener

createInMemoryOrchestratorStore can receive each raw workflow event as it is appended:

const store = createInMemoryOrchestratorStore({
  onEvent: async (event) => {
    console.log(event);
  },
});

appendEvent invokes the callback. If the callback throws, the append rejects and the run follows core's event-hook failure behavior.

The in-memory store keeps at most 100 snapshots and 5,000 events by default. Pass maxSnapshots or maxEvents to tune retention, or Infinity to disable a cap.

#Recovery

createOrchestrator can recover a managed run from a stored snapshot when waitForRun, cancelRun, or cancelTask receives a run ID that is no longer active in memory:

const orchestrator = createOrchestrator({
  store,
  workerGateway: gateway,
  recovery: {
    interruptedTasks: "retry",
    loadSnapshot: async (runId) => store.getSnapshot(runId),
  },
});

const completed = await orchestrator.waitForRun("run_123");

Recovery uses the core runner's snapshot recovery. If you omit interruptedTasks, core uses the workflow's recovery config and falls back to "manual".

#Limitations

• Workers execute one task job, not a full workflow.
• Workers use core's TaskRegistry for real task handlers.
• The host orchestrator uses core's createRunner for DAG state.
• worker/thread provides local process isolation, not a hostile-code security sandbox.
• Persistence is interface-ready; alpha includes only the in-memory store.