What changed
Vercel introduced Daemons in public beta on 15 October. They run long-lived processes that pull work from queues, databases or third-party systems without exposing an HTTP server. Vercel lists queue consumers for its own Queues, Celery, BullMQ, RabbitMQ and Kafka, as well as Temporal or Cadence workers and self-hosted CI runners.
A daemon is packaged with a Dockerfile or Containerfile and deployed in the same project as the application. Vercel builds the image, keeps the configured number of instances running in each selected region, restarts processes that exit or crash, and exposes their output in project logs. Capacity is set manually or through the API.
The first release has important boundaries: daemons cannot serve HTTP traffic and do not include persistent filesystem storage. Vercel also tells operators to handle termination signals gracefully before a restart.
Our perspective
This fills a genuine gap between request-driven functions and pull-based background work. It is a useful fit for coding agents, indexing pipelines and other jobs that wait on queues, especially when teams want frontend, backend and workers to ship from one project.
But an always-on worker is not a durable workflow. A restarted process restores capacity; it does not prove that the interrupted job was acknowledged, completed once or resumed from the correct step. Reliability still depends on queue visibility timeouts, idempotency keys, checkpointed state and safe retries.
The clean design boundary is to keep job intent and progress in a durable queue or workflow system, keep artifacts in suitable persistent storage, and treat each daemon as replaceable compute. Before adopting the beta, kill workers during real jobs and verify redelivery, duplicate handling, deploy hand-off, logs and regional recovery.
Go to the source
Primary reporting behind this brief. Our perspective is editorial analysis.
- Vercel — Daemons public beta announcement 15 Oct 2026
- Vercel — Daemons product documentation 15 Oct 2026
- Vercel — Backends on Vercel: Python, containers, and Daemons 15 Oct 2026