Skip to content

Schedules

Schedules let the harness run work without an interactive chat session. They are the recurring half of autonomous operation, alongside agents.

What a schedule is

A schedule is a harness job that fires once or periodically:

  • Schedule expression — cron-like string or a delay
  • Timezone — when the expression is interpreted (default UTC)
  • Instructions — what the agent should do when the job fires
  • Conversation — the job can be tied to a conversation the harness creates or continues
  • Status — scheduled, running, finished, or error, with run count and last/next run times

Jobs are created by operators in the UI or by the agent itself through the schedule tool during a conversation.

Typical uses

  • Periodic health or coverage checks (“every Monday, summarize new failed log sources”)
  • Recurring hunts or report drafts
  • Follow-ups after a delay (“in 24 hours, re-check this IOC”)
  • Maintenance tasks that should not depend on someone being in the chat

Autonomy and control

Schedules run under the same harness rules as chat:

  • Tools still go through the sandbox
  • Vetting can still pause sensitive calls
  • The audit log records the conversation items and tool executions the job produces

You can leave a schedule fully autonomous, or combine it with vetting and later review in Conversations / Schedules.

In the UI

  • Schedules — list jobs, open detail (expression, status, linked conversation, errors)
  • From a live conversation, the agent may create a job with the schedule tool; it then appears in this list