Schedules are coming soon for UNFI AI Agents. The behavior described below reflects the planned functionality.
Creating a schedule
During a session, ask the agent to schedule a task:“Schedule this sales summary to run every Monday at 8 AM and email it to me.”The scheduling agent creates the schedule, sets the cron expression, and configures delivery. You can also ask it to update an existing schedule’s timing, description, or delivery channel.
Schedule settings
Each schedule has the following properties:Viewing the task prompt
The task prompt is rendered as formatted markdown. Click Preview on any schedule to see the full prompt in a modal.Timezone support
Every schedule stores its own timezone. When you create a schedule, it defaults to your personal timezone (set by instructing the Agent your preferred timezone). You can override the timezone on any individual schedule. The cron expression is always evaluated relative to the schedule’s timezone. For example, a schedule set to0 9 * * * in America/Chicago runs at 9 AM Central regardless of daylight-saving changes.
Editing a schedule
Click any schedule to open the edit modal. You can change the description, cron expression, task prompt, delivery channel, timezone, and status. Changes take effect on the next scheduled run. Switching a schedule between Active and Inactive is done with a toggle — no destructive actions or confirmations required.Viewing run history
Click View runs on any schedule to see a history of past executions, including:- Run status — Pending, Running, Completed, or Failed
- Start and end times
- A link to the full session for each run
Running a schedule manually
Click Run now on any schedule to trigger an immediate execution, regardless of the cron expression. This is useful for testing a schedule or re-running after a failure.One-off runs
Runs that were triggered manually (via Run now or by the scheduling agent without a cron expression) appear in the Ad hoc tab and are labeled One-off run.Reliability
Scheduled runs are designed to recover from temporary failures. If a run cannot complete, it is marked Failed in the run history with an error message you can share with support.Data freshness checks
Before a scheduled run generates a report, a check confirms that the underlying data sources are fresh. If required data is missing or has not yet refreshed for the period the report covers, the run is skipped instead of producing a misleading result. When that happens:- Recipients receive a short notification on the configured delivery channel letting them know data is not yet available.
- The run is retried automatically — there is nothing you need to do to re-queue it.
- Once the data lands, the next retry generates the report normally.
Approvals inbox
Scheduled runs are non-interactive, so the assistant can’t pause mid-run to ask you a question. When a scheduled action needs your sign-off — for example, emailing a recipient who isn’t on your approved list — AI Agents parks that single action instead of failing the whole run. What you’ll see:- The run still completes and delivers everything that didn’t need approval. Parked actions are excluded from delivery and called out in the report.
- You receive an email notification with a Review button linking to the Approvals inbox.
- The Approvals entry in the sidebar shows a badge with the number of pending items.
- Approve an action — AI Agents runs that one action (e.g., sends the parked email) without re-running the entire schedule.
- Reject an action — the action is discarded, and nothing is sent.
