Skip to content

scheduling.md

Latest commit

 

History

History
86 lines (62 loc) · 3.09 KB

scheduling.md

File metadata and controls

86 lines (62 loc) · 3.09 KB
Raw

Scheduling

Run agents on a cron schedule. Schedules are stored in Conductor and survive server restarts — no cron daemon or external scheduler needed. Use the SDK's typed SchedulerClient directly.

Deploy an agent with a schedule

import com.netflix.conductor.common.metadata.workflow.StartWorkflowRequest;

import io.orkes.conductor.client.SchedulerClient;
import io.orkes.conductor.client.model.SaveScheduleRequest;

Agent reportAgent = Agent.builder()
    .name("daily_report")
    .model("anthropic/claude-sonnet-4-6")
    .instructions("Generate a daily sales summary.")
    .build();

try (AgentRuntime runtime = new AgentRuntime()) {
    runtime.deploy(reportAgent);

    StartWorkflowRequest workflow = new StartWorkflowRequest();
    workflow.setName(reportAgent.getName());
    SchedulerClient schedules = runtime.getSchedulerClient();
    schedules.saveSchedule(new SaveScheduleRequest()
        .name("daily_report-daily")
        .cronExpression("0 0 9 * * ?") // 9 AM every day
        .zoneId("America/New_York")
        .startWorkflowRequest(workflow));
}

Schedule with custom input

Pass a fixed prompt or parameters to the scheduled run:

StartWorkflowRequest workflow = new StartWorkflowRequest();
workflow.setName("daily_report");
workflow.setInput(Map.of("report_type", "weekly", "include_charts", true));

SaveScheduleRequest weeklyDigest = new SaveScheduleRequest()
    .name("daily_report-weekly")
    .cronExpression("0 0 8 ? * MON")
    .zoneId("UTC")
    .description("Monday morning executive digest")
    .startWorkflowRequest(workflow);

Manage schedules

SchedulerClient schedules = runtime.getSchedulerClient();

// List all schedules for an agent
List<WorkflowSchedule> all = schedules.getAllSchedules("daily_report");

// Pause and resume (by wire name). A pause reason is retained by Conductor.
schedules.pauseSchedule("daily_report-daily", "quarter-end freeze");
schedules.resumeSchedule("daily_report-daily");

// Delete (by wire name)
schedules.deleteSchedule("daily_report-daily");

// Preview the next N fire times for a cron expression
List<Long> next = schedules.getNextFewSchedules("0 0 9 * * ?", null, null, 5); // epoch millis

Native schedule fields

SaveScheduleRequest accepts the schedule name, cron expression, timezone, pause/catchup flags, optional start/end timestamps, description, and a StartWorkflowRequest. The StartWorkflowRequest identifies the deployed agent workflow and carries its fixed input.

Migration from the AI scheduling facade

AgentRuntime.schedules(), Schedule, and ScheduleInfo are not part of the AI SDK. Obtain the shared client with runtime.getSchedulerClient(), build a SaveScheduleRequest directly, and use WorkflowSchedule for retrieved schedules. Schedule creation, update, and deletion are explicit; deploying an agent does not reconcile schedules.

Cron syntax

Quartz cron: second minute hour day-of-month month day-of-week [year]

0 0 9 * * ?         every day at 9:00 AM
0 0 */6 * * ?       every 6 hours
0 0 8 ? * MON-FRI   weekdays at 8 AM
0 30 17 1 * ?       1st of every month at 5:30 PM