Vercel adds Daemons, always-on container workers for backends

A frontend host that also runs the queue consumer and the Python API is a different product.

At its Ship conference in San Francisco, Vercel has announced Daemons in public beta, a service for running long-lived container workers next to a Vercel app, and made Python support and the Drives storage product generally available. Chief technology officer Malte Ubl told The New Stack the company's plan is "let's look at what people are doing, and let's make sure that a larger chunk of them could just only use Vercel as their infrastructure provider".

What a Daemon is

Daemons run from a container image, declared in vercel.json under a daemons key, and Vercel keeps the number of instances the config asks for in each region. The service does not accept inbound HTTP traffic. A Daemon reaches out to a queue, a workflow engine or an external API for its work and processes jobs as they arrive. Vercel builds the Dockerfile at deploy time, pushes the image to its own Container Registry and rolls out instances with the production deployment. If an instance crashes, Vercel restarts it. A rollout sends SIGTERM and waits ten seconds before termination.

The company's own example combines a Next.js front end and a Python API with a video processor in a container that pulls jobs from Vercel Queues to generate playback files and thumbnails using FFmpeg. Ubl called transcoding a typical case, "extremely CPU-bound workloads" where "I just need computers that I want to always be 100% CPU". Ubl said Vercel's own control-plane API handles "about a billion requests a month" on Vercel, and the company is moving many of its asynchronous workloads onto Daemons.

Three limits of the beta that will matter during an evaluation: Daemons run only in production for now, not in preview deployments; there is no autoscaling and no persistent storage attached to a Daemon; and vercel dev prints a warning and leaves the Daemon out of local runs. Daemons are available on all plans, including the Hobby tier, within each plan's quotas.

Python went generally available

Vercel's Python support left beta on the same day. The platform now runs FastAPI, Flask and Django on CPython with PyPI dependencies, and Vercel takes care of framework detection, builds and scaling. The company says 150,000 teams already run Python backends on its platform, including engineers at OpenAI, Polymarket and Exa, and that 5,000 teams have made 150,000 container deployments since container support entered public beta. These are Vercel's own figures.

Drives for agent files

Vercel Drives, a storage product designed for AI agents, also went generally available. Writes land on local NVMe disk and stream asynchronously to object storage. An agent can mount a drive in a new sandbox, change its runtime and keep the files from the previous session. "I wouldn't claim it's an industry first, but it's certainly an instance of a new type of storage product," Ubl said. Default capacity is 1 TiB on paid plans, 1 GiB on Hobby, and up to 16 TiB on configuration. Agents can also fork a drive to try different approaches in separate writable workspaces, with no merge back into the original.

A team already deploying its frontend on Vercel can now run a queue consumer, a Python API and the storage its agents write to on the same control plane. The trade is the usual one: a managed runtime like Node, Bun or Python gets security updates without a redeploy, while a container is the customer's job to keep current.

Source

Daemons documentation, Vercel, and Vercel adds always-on workers to run more of your backend, The New Stack.