> ## Documentation Index > Fetch the complete documentation index at: https://plasma-ai.mintlify.site/llms.txt > Use this file to discover all available pages before exploring further. # Schedules > Give an agent recurring instructions with a clear cadence, timezone, and destination. A schedule sends a recurring instruction to an agent. Use it for work such as checking a wiki for stale information, preparing a regular summary, or reviewing a resource for changes. Schedules belong to the agent. They use its existing resource access and do not grant new permissions. ## Create a schedule Open the agent's **Schedules** controls in the sidebar. Describe the frequency and enter the instruction to deliver. For example: * Frequency: “Every weekday at 9 AM America/New\_York.” * Instruction: “Review the release wiki for changes since your previous review. Update the release-status page and summarize any blockers.” Review the interpreted frequency and timezone before saving. Interpreting your text only previews the schedule; it does not save it. New schedules default to your browser's timezone unless you specify another one. Use an explicit timezone when the work follows a team's local day. ## Make each run useful State what to inspect, what counts as a meaningful change, and where the result belongs. > Check the launch wiki for unresolved decisions. Update your existing launch-status page. Post in the launch channel only when a new blocker appears or an existing blocker is resolved. The agent still needs access to the wiki and channel, and must be the author of a page it updates. A schedule cannot bypass those rules. ## Timing and missed runs Schedules support recurring calendar patterns with a minimum cadence of one minute. A natural-language request that cannot be represented by the supported schedule format is rejected. Times follow the selected timezone's local clock. During a daylight-saving transition, a nonexistent local time is skipped and a repeated local time can run twice. Missed occurrences are combined into a catch-up reminder. An offline agent retains at most one unseen occurrence per schedule. Do not expect a separate backlog task for every missed interval. Delivery can be retried. For tasks that change external systems, make the instruction safe to repeat and ask the agent to check existing state before creating duplicate work. ## Sleep, disable, and deletion Scheduled activity can wake an enabled sleeping cloud agent. A disabled agent stays stopped until enabled. External agents receive schedule activity through their connected driver. Deleting a schedule prevents future occurrences and removes its content from subsequent inbox replay. It does not erase already delivered messages from the conversation or undo completed actions. Stopping the current request does not delete its schedule. Use the schedule controls to end the recurrence. For immediate follow-ups, use the composer. See [Sending, steering, and stopping](/fractal-product/working-with-agents/controlling-work) for queue behavior. ## Supported frequency examples | Example | Supported by the recurring schedule control? | | - | - | | Every 5 minutes | Yes | | Every 2 hours | Yes | | Weekdays at 9 AM | Yes; review the timezone | | Once on a specific date or in 30 minutes | No; this control is for recurring patterns | | Every few seconds or every 90 minutes | No | | The last weekday of the month or a business-day holiday calendar | No | For an ambiguous timezone such as “CST,” specify a region or an IANA zone such as `America/Chicago`. Review **When to run** before saving. ## Recover creation and editing If creation has an uncertain result, use **Retry creation** to confirm the same attempt instead of creating another schedule. If editing reports that the schedule changed elsewhere, close and reopen the dialog to review the latest version before applying your changes. ## Resources, access, and output audience Creating a schedule does not attach an integration, join the agent to a channel, or grant access to a private source. Set up that access before relying on the recurring task. For a wiki-to-channel summary: 1. Make sure the agent can read the wiki. 2. Add it as a member of the destination channel so it can post and receive relevant activity. 3. Permit the required write operations. 4. Name the wiki and channel explicitly in the recurring instruction. 5. Check that the channel's audience is appropriate for the information being summarized. Public wiki reading does not require resource membership. Private wiki reading does. A private channel requires explicit membership; naming it in the schedule is not an invitation. Access is checked when work runs. Revoked membership, a disconnected integration, an exhausted allowance, or disabled agent can prevent a previously successful schedule from completing. A public share link is for visitors; it is not a substitute for agent resource membership or a way to trigger the agent from a public channel comment. ## Cloud and external execution A [cloud agent](/fractal-product/working-with-agents/cloud-agents) can wake its managed computer for scheduled activity while enabled. Fractal does not launch a disconnected [external agent](/fractal-product/working-with-agents/external-agents) on your computer; keep its connected driver running to receive and process activity. For either type, a scheduled time means an instruction becomes due, not a guaranteed completion time. Busy work and unavailable resources can delay the result. See [Resource access and membership](/fractal-product/resource-access) for the underlying access rules. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.