Vercel's Daemons beta moves queue-consuming workers onto the platform that hosts the frontend
Vercel Daemons, now in public beta, keep a customer's container running and hold the instance count set for each region. At launch a Daemon only pulls work from a queue or database and is never invoked by internet requests.
The Engineer · Build desk
Vercel says 5,000 teams made 150,000 container deployments since container support entered public beta. Vercel runs control plane API traffic of about a billion requests a month on Vercel. Delphi moved its backend from AWS ECS and ships more than 100 times a day.
| Figure | Value | Note | Claim |
|---|---|---|---|
| Container deployments by 5,000 teams since beta | 150,000 deployments | 11 | |
| Vercel control plane requests a month, about | 1,000,000,000 requests | a good chunk of control plane API | 18 |
| Delphi production deploys a day, more than | 100 deploys | backend moved from AWS ECS to Vercel | 22 |
What happened
- Vercel restarts a Daemons process if it exits or crashes, so the instance count holds without the customer running a supervisor.
- Vercel says 5,000 teams have made 150,000 container deployments since container support entered public beta.
- Vercel runs a good chunk of its control plane API on its own platform, about a billion requests a month, and has started moving asynchronous workloads to Daemons.
- Python support and Drives, Vercel's persistent storage for AI agents, went generally available at the same event.
Why it matters
- Who has to decide decision A team running a queue consumer on another provider now has a concrete test: does the worker only pull jobs, and does it fit in a container? If both answers are yes, the move is on the table.
- What it limits constraint Capacity is a number you set per region. The sources describe a fixed count that Vercel maintains and do not mention autoscaling for Daemons.
- What it costs cost Daemons billing is not described in the coverage. Containers deployed as Functions skip idle waits under Active CPU pricing, but Ubl's transcoding example wants CPUs pinned near 100%, which leaves little idle time to waive.
A Daemon is a container with a supervisor attached. You deploy through Git or the Vercel CLI, and Vercel builds the image and stores it in its Container Registry [8]. The worker runs on Fluid compute [6]. Vercel's example config is three lines under a `[daemons.videos]` table in vercel.toml, and the one that sets capacity is `instances.iad1 = 1`, which keeps a single video processor running in iad1, Washington, D.C. [7]
That count is the control Vercel says was missing. Vercel CTO Malte Ubl used the company's own deployment pipeline as the example. "When you call our API to make a deployment, we don't do that deployment in that process that handles your request, because we can't actually do an infinite number of concurrent deployments," Ubl said [12]. GitHub rate limits are one reason to back off, and Vercel's existing compute primitives did not give developers "good control over the amount of concurrency" for that kind of worker [13]. In our view the instance count is a coarse version of that control. N workers in a region means at most N pullers. How many jobs each one takes at once is decided in the worker's own code.
The pull model is the design. A Daemon is not invoked by requests from the internet. It reaches out for work [5]. Ubl put it opposite Vercel's sandboxes: "They're almost the opposite of temporary," he said of Daemons. "For temporary, you have sandboxes." [14] He also said Vercel did not invent the model: "we ship it because people very much want this." [15]
The container costs you patching. With a managed runtime for Node, Bun or Python, Vercel applies a security release without the customer redeploying, Ubl said [16]. A container is different: "you do have to redeploy. It's ultimately your responsibility," he said. "From our point of view, it's just bytes." [17] Vercel's example Dockerfile starts from python:3.12-slim and installs ffmpeg with apt-get at build time [9]. Both stay as built until someone redeploys.
Vercel's post offers one migration story, and it predates Daemons. Delphi moved its Python backend from AWS ECS to Vercel. It has ten engineers, no dedicated infrastructure role, and handles background jobs with Queues [22]. Vercel already sells queues, durable workflows and Services for deploying frontend and backend together. Ubl said those cover request-driven work but not everything a sophisticated backend does [23]. Daemons are aimed at what is left. Ubl described the push this way: "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," he said [21].
What to watch
- How Vercel prices Daemons, and whether Active CPU billing applies to always-on workers.
- Whether Daemons support anything beyond pull-based jobs once the beta ends.
- How much of Vercel's own asynchronous load moves off the control plane's roughly billion monthly requests and onto Daemons.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence50
- Adoption
- Insufficient
- Hype gap+5
- Incentives70
- Confidence58
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
At its Ship conference in San Francisco, Vercel introduced Daemons, a service for continuously running background workers, which is in public beta.
- [2]
Daemons lets developers deploy a container and specify how many instances should run in each region.
- [3]
Vercel makes sure Daemons processes keep running and restarts them if they exit or crash.
- [4]
At launch Daemons supports pull-based workloads, where workers fetch jobs from a queue or database and process them as they arrive.
- [5]
Daemons stay on but are not invoked by requests from the internet; they reach out for work instead.
- [6]
Vercel Daemons deploys containers as continuously running background workers on Fluid compute.
- [7]
In Vercel's example vercel.toml, a [daemons.videos] table sets root = ".", entrypoint = "Dockerfile" and instances.iad1 = 1, which keeps one video processor instance running in iad1 (Washington, D.C.).
- [8]
When deploying through Git or the Vercel CLI, Vercel builds the container image and stores it in the Vercel Container Registry.
- [9]
Vercel's example Dockerfile for the video processor starts FROM python:3.12-slim and installs ffmpeg with apt-get during the image build.
- [10]
Alongside Daemons, Vercel made Python support generally available and made Drives, its persistent storage for AI agents, generally available.
- [11]
Vercel says 5,000 teams have made 150,000 container deployments since container support entered public beta.
- [12]
"When you call our API to make a deployment, we don't do that deployment in that process that handles your request, because we can't actually do an infinite number of concurrent deployments," Ubl said.
- [13]
GitHub rate limits are one reason Vercel's deployment pipeline backs off, and Vercel's existing compute primitives did not give developers "good control over the amount of concurrency" for that kind of worker.
- [14]
"They're almost the opposite of temporary," Ubl said of Daemons. "For temporary, you have sandboxes."
- [15]
Ubl acknowledged Vercel did not invent the model: "we ship it because people very much want this."
- [16]
With a managed runtime for Node, Bun or Python, Vercel applies a security release without the customer redeploying.
- [17]
Of containers, Ubl said: "you do have to redeploy. It's ultimately your responsibility," and "From our point of view, it's just bytes."
- [18]
Vercel runs a good chunk of its control plane API on Vercel, about a billion requests a month, and is starting to move many of its asynchronous workloads onto Daemons.
- [19]
Containers deployed as Vercel Functions on Fluid compute use Active CPU pricing, which does not charge for idle time spent waiting for API responses, database queries or LLM inferencing.
- [20]
Ubl called transcoding a typical Daemons case, with "extremely CPU-bound workloads" where "I just need computers that I want to always be 100% CPU."
- [21]
"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," Ubl said.
- [22]
Delphi moved its Python backend from AWS ECS to Vercel; with ten engineers and no dedicated infrastructure role, it ships to production more than 100 times a day and uses Queues for background jobs.
- [23]
Vercel already offers queues and durable workflows, and its Services feature deploys frontends and backends together; Ubl said these cover request-driven work but not everything a sophisticated backend does.
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
2 independent publishers whose own reporting we read for this story.
- blog.vercel.comBackends on Vercel: Python, containers, and Daemons
1 article · October 15, 2026
- thenewstack.ioVercel adds always-on workers to run more of your backend
1 article · October 15, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.