Vercel Schedules enters public beta with runtime-managed cron for apps and agents

Vercel has moved Schedules, its service for one-time and recurring task execution, into public beta. Announced in a blog post and changelog entry on November 5, the feature is available on all plans, including the free Hobby tier, and Vercel describes it as the successor to Vercel Cron Jobs — though existing cron jobs keep working, and a migration guide covers the transition.

The central change from Cron Jobs is that scheduling becomes a runtime resource rather than a deploy-time setting. Schedules can be created, modified, enabled, disabled or fired on demand by running code, with no redeploy in between.

Static and dynamic schedules

According to the blog post, Schedules comes in two flavors. Static schedules are declared in vercel.json or through the Build Output API and synchronize when you deploy; they can invoke a private Vercel Function, hit an HTTP path, or publish to a Vercel Queue topic. Dynamic schedules are managed while the application runs, via a JavaScript SDK, the CLI or the API, and cover both single-fire and recurring patterns; these always dispatch to a Queue topic. Both kinds accept an IANA time zone and can carry a payload alongside the trigger.

A scheduling primitive for agents

Much of the announcement targets AI agent workloads. Vercel's example: a user asks an agent with web access to summarize the top five Hacker News stories tomorrow morning. The conventional implementation would store the request and its run time in a database, then run a cron job that polls for due work. With Schedules, the agent calls a tool that invokes Schedules.create with a one-time expression, a time zone and the task payload. When the moment arrives, Vercel delivers the saved request to the application through Queues, and the app's code re-checks the user's permissions before launching the agent.

Vercel says the Eve agent framework already uses Schedules to give production agents dynamic scheduling tools, and that agent builders using fx's JavaScript library, libfx or Pi's agent library can wire in a comparable tool. The Nitro 3 server framework has an experimental integration that runs its existing scheduledTasks configuration as private functions.

Private invocation and queue delivery

A long-standing annoyance with HTTP-triggered crons is securing the endpoint. Vercel's changelog notes that adding a schedule/v1beta trigger to a function makes it private: it gets no public URL, and only Vercel's scheduling infrastructure can invoke it. The blog post adds that Vercel verifies each call originates from the scheduling service, so developers don't have to build that check themselves. The @vercel/schedules package ships a handleSchedule helper for processing dispatches, which arrive with metadata such as the schedule's name and scheduled time.

Testing, pricing and migration

Scheduled work can be exercised locally with vercel dev, then tested on Preview deployments or in custom environments before production. A Run Schedule action in the project's Schedules tab fires a schedule immediately rather than waiting for its time slot, and the tab also shows when past runs happened.

On pricing, the changelog lists $5 per million runs and $10 per million schedule creations for Pro and Enterprise, while Hobby includes 5,000 runs and 500 creations per month.

Why it matters

Scheduled execution is one of the oldest primitives in computing, yet on most platforms it remains frozen at deploy time. By turning schedules into something a running program — or an agent acting on a user's behalf — can create, cancel and reschedule on the fly, Vercel eliminates the database-plus-polling pattern that teams currently hand-roll for exactly this case. The private-invocation model and per-run pricing are the concrete changes for existing users, but the agent story is the reason the product exists: deferred, reschedulable work is becoming a core requirement for agent frameworks, and infrastructure vendors are starting to ship for it directly.