Deployment, scaling, and graceful shutdown

Run TaskHandler/worker processes and AgentRuntime.serve() as long-lived Node.js services — containers, VMs, or any process manager. Do not construct a new client/runtime per request; reuse them for the application's lifetime.

process.on("SIGTERM", async () => {
  await handler.stopWorkers();
  process.exit(0);
});

Scale by adding worker instances (each polls independently) and use concurrency/domain to isolate or parallelize specific task types rather than the whole process. On shutdown, stop task handlers and runtimes so in-flight tasks are redelivered instead of abandoned — see reliability for timeout/retry policy that governs redelivery.