Schedules

Agents often need to run without an incoming request, such as for daily summaries, recurring reports, data synchronization, or cleanup.

Flue does not prescribe a scheduling library; nor do we build scheduling into the framework itself. Instead, use the scheduling tools provided by your deployment environment and send the resulting input to an agent with dispatch(...). This guide uses Cloudflare Cron Triggers and Croner for Node.js.

Scheduling on Cloudflare

Add a Cron Trigger to your project’s wrangler.jsonc:

{
  "triggers": {
    "crons": ["0 9 * * *"],
  },
}

Then define the handler in src/cloudflare.ts and dispatch the scheduled input to an agent:

import { dispatch } from '@flue/runtime';
import dailySummary from './agents/daily-summary.ts';

export default {
  async scheduled(controller: ScheduledController) {
    await dispatch(dailySummary, {
      id: 'daily-summary',
      input: {
        type: 'schedule',
        prompt: 'Review recent activity and prepare the daily summary.',
        scheduledAt: new Date(controller.scheduledTime).toISOString(),
      },
    });
  },
};

Cron Triggers use UTC. See Cloudflare’s scheduled() handler documentation for the complete API.

Scheduling on Node.js

Node.js does not include a built-in cron scheduler, so you will need to choose an ecosystem option that fits how your application is deployed. This example uses Croner, a popular lightweight scheduler with async callbacks, overlap protection, and timezone support.

import { dispatch } from '@flue/runtime';
import { Cron } from 'croner';
import dailySummary from './agents/daily-summary.ts';

new Cron(
  '0 9 * * *',
  {
    protect: true,
    timezone: 'UTC',
    catch: (error) => console.error('Scheduled agent admission failed', error),
  },
  async () => {
    await dispatch(dailySummary, {
      id: 'daily-summary',
      input: {
        type: 'schedule',
        prompt: 'Review recent activity and prepare the daily summary.',
        scheduledAt: new Date().toISOString(),
      },
    });
  },
);

For production schedules that must survive restarts or coordinate across replicas, we suggest a more persistent scheduler, such as BullMQ. An in-process scheduler like Croner only runs while that Node process is alive.

Next steps

• Agents — create the agent that receives scheduled input.
• Workflows — create and expose finite scheduled operations.
• Cloudflare — configure the Cloudflare target and cloudflare.ts entrypoint.
• Node.js — build and operate the generated Node.js server.