Cron jobs handle work that must recur: backing up a database on a timer, keeping two systems in sync. You declare them in code, then edit the config and redeploy to change them. That model assumes the timing is known ahead of time. Agents interacting with users as work evolves need schedules created and changed in real time — which is the gap Vercel Schedules fills, letting agents and applications create, modify, or cancel one-time and recurring schedules through code without a redeploy.
Schedules versus Cron Jobs
Schedules extends the static model of Vercel Cron Jobs so that users and agents can decide when work happens while the application is running. A schedule only records when Vercel should trigger a task and where the task should be delivered; the application supplies the code that does the work.
Two flavors exist:
- Static schedules live in
vercel.jsonor the Build Output API configuration and sync on deploy. They can invoke private functions or HTTP paths, or publish to Queue topics. - Dynamic schedules are created and managed at runtime through the JS SDK, CLI, or API, covering both one-time and recurring jobs via Queue topics.
Scheduled work can be tried on Preview deployments and in custom environments before it reaches Production.
Replacing the database-and-poller pattern
Consider a user asking an agent with web access to summarize the top five Hacker News stories the next morning. The conventional approach stores the request and its due time in a database, with a cron job polling for work that has come due. With Schedules, the agent writes a schedule holding the request and its execution time; no separate database or polling job is needed. The code below defines an AI SDK tool that gives an agent this ability.
lib/schedule-task-tool.ts
import { randomUUID } from 'node:crypto';
import { tool } from 'ai';
import { z } from 'zod';
import { Schedules } from '@vercel/schedules';
export function scheduleTaskTool(tenantId: string, userId: string) {
return tool({
description: 'Schedule a task to run at a future time.',
inputSchema: z.object({
task: z.string().min(1).max(1000),
at: z.string().describe('Future local time: YYYY-MM-DDTHH:mm'),
timezone: z.string().describe('IANA time zone, e.g. America/New_York'),
}),
execute: ({ task, at, timezone }) =>
Schedules.create({
name: `agent-task-${randomUUID()}`,
namespace: tenantId,
expression: { type: 'single', at },
timezone,
target: { topic: 'agent-work' },
payload: { userId, task },
}),
});
}
When the task comes due, Vercel delivers the stored request to the app via Queues. The app reads the request, rechecks the user's current permissions, and launches the agent, which fetches the posts with a web tool and returns a summary the app stores for the user.
Existing agent stacks can add an equivalent tool: fx's JavaScript library libfx and Pi's agent library both allow a scheduling tool to be attached in the same manner. The eve agent framework uses Vercel Schedules to offer dynamic scheduling tools to production agents hosted on Vercel.
Recurring jobs and private invocation
Recurring work can be created dynamically through the JS SDK, API, and CLI, or declared in project configuration. A daily job that purges expired records, for instance, can invoke a private function in a chosen time zone:
vercel.json
{
"$schema": "https://openapi.vercel.sh/vercel.json",
"schedules": {
"cleanup": {
"schedule": "0 2 * * *",
"function": "api/cleanup",
"timezone": "America/New_York"
}
},
"functions": {
"api/cleanup.ts": {
"experimentalTriggers": [
{ "type": "schedule/v1beta" }
]
}
}
}
Schedules that private cleanup function for 2 a.m. New York time each day. Vercel calls the code directly, with no public URL exposed, and verifies that each call originates from the scheduling service, so no additional check is required in your application.
Apps already running the Nitro 3 server framework can use its experimental integration to execute the recurring jobs in their existing scheduledTasks configuration as private functions.
Trying schedules before production
Both the agent task and the cleanup job can be exercised locally with vercel dev, then validated on a Preview branch or in a custom environment. The Schedules tab in a project offers Run Schedule to trigger either task immediately rather than waiting for its time, and shows when past runs occurred.
Availability
Vercel Schedules is in Public Beta and can be applied to recurring tasks an application already runs. When a user or an agent picks the timing for a piece of work, the application can create and manage a schedule for it. The Vercel Schedules skill and the documentation are the places to start.



