Agent wakeups and schedules

Waking an agent appends a proactive turn to its lifetime conversation. A wake is a normal agent run with the usual model, budget, and permissions.

Agent background work follows the agent creator's no-training preference and any organization requirement. Later work that reuses automation results also follows those automation creators' no-training preferences.

A scheduled wakeup can be a broad routine check-in, a named recurring responsibility, or an exact one-time reminder.

Trigger	Prompt	If the agent is busy	Schedule effect
Manual wake	Routine check-in	No-op: returns already running	Does not move a schedule
Scheduled wakeup	Its own saved prompt	Runs when the agent is free	Completes once or moves to its next time after it starts

Finish agent creation before configuring or testing wakeups. A draft may already have an inbox or connected channels, but it cannot run until activated. Timing or trial requests made during conversational setup are saved for later review; setup does not create a schedule or execute a trial task.

During a wakeup, an agent with approved browser access can start its saved browser under either conversation approval setting. The browser panel can stay closed. Website permissions and private browser control still apply. See Browser access.

Manual wakeup

Use Wake now for an immediate check-in. The agent reviews what changed within its standing instructions and replies briefly when nothing needs attention. It can use connectors enabled in its settings; turn off any that autonomous work should not reach.

A manual wake does not reset or advance any scheduled wakeup. If the session is already active, the wake is a no-op because the agent is already working. If today's daily budget is exhausted, d5s refuses the wake until the next local day.

Scheduled wakeups

Every scheduled wakeup has its own name, instructions, timing, state, and optional delivery destination. Ask the agent to deliver to a supported connected chat channel or a linked Telegram teammate DM when the result should be sent outside d5s. The run's response also remains in the agent conversation.

For example, ask the agent to send your task briefing to your linked Telegram DM each morning at a stated time and timezone. The teammate must remain linked and have permission to view and run the agent. Recipient access and the bot connection are checked again before each Telegram message is sent. If the linked recipient or bot is no longer authorized when a wakeup is due, the schedule pauses. If the existing destination is still connected, restore the recipient’s access and resume the wakeup. After disconnecting and reconnecting the bot or linking again, ask the agent to update the existing wakeup to the newly linked DM, even for the same recipient, then resume it. You can also ask the agent to choose another authorized destination.

Create and manage wakeups under Manage agent → Autonomy → Scheduled wakeups. You need permission to run the agent in its project to create, change, or remove its wakeups. You can also ask the agent to create, move, pause, or remove its own wakeups.

Choose Once for an exact date, time, and timezone. Choose Recurring for every 5 minutes, every 15 minutes, hourly, every 6 hours, daily, or weekly. When you ask the agent, it can also set any whole number of minutes, at least 5. A weekly wakeup can select any combination of Monday through Sunday, so one wakeup can cover weekdays, weekends, or another weekly pattern. A recurring wakeup can start immediately or on a chosen date, and can continue indefinitely, end on a date, or stop after a chosen number of successful runs. The editor previews the next occurrence before you save.

A chosen start date delays the first fire without changing the cadence. The first run is the first aligned occurrence at or after that start. A daily 09:00 wakeup starting on Monday therefore first fires at 09:00 on Monday, not at midnight.

A weekday selection belongs to the weekly cadence. Moving a weekly wakeup to any other cadence drops the selection and keeps the local time of day, so a Monday-to-Friday 08:00 wakeup moved to daily then runs at 08:00 every day.

An agent's Inbox check runs every hour on every day of the week. To limit it to part of the day, open it, select Only between in Active hours, and set a start and an end time. The times use the wakeup's timezone. The check runs from the start time until just before the end time. An end time earlier than the start time continues past midnight, for example 22:00 and 06:00 for an overnight check. The start and end times must differ, and at least one run must fall between them. Clear Only between to run it all day again. Active hours apply only to a cadence shorter than one day; changing the check to daily, weekly, or once removes them.

Selecting an existing wakeup opens the same editor in place. The enable toggle and delete action remain separate from editing. Completed one-time and finite recurring wakeups remain visible with their completion state.

If the agent is busy or a temporary error occurs, the due wakeup runs later. Once it starts, a one-time wakeup completes and a recurring wakeup advances. A recurring wakeup also completes after its final allowed run or when its next occurrence would fall beyond its end date. A budget-skipped recurring occurrence advances without counting toward a run limit. A budget-skipped one-time wakeup stays due and retries when budget becomes available.

If the agent's selected model family is disabled or has no available version under the organization settings, the scheduled wakeup pauses. Choose an available model under Manage agent → Profile, or ask an organization administrator to restore the family, then enable the wakeup again under Autonomy → Scheduled wakeups. The saved family choice stays in place until you change it.

A run can still fail. If a run fails for a temporary reason before the agent acts, d5s retries it automatically a few times within the next hour. Routine check-ins retry the same way. A retry does not count toward a run limit. A later occurrence, a new time, or disabling the wakeup cancels the retry. Other failures are not retried. When a one-time wakeup's run fails without a retry, the wakeup turns off but is not marked complete; set a new time to run it again. A recurring wakeup continues at its next occurrence.

Check a mailbox on a schedule

Under Manage agent → Autonomy → Scheduled wakeups, choose New wakeup, then set Work to Check mailbox. Select an IMAP mailbox connector that this agent can already access. Give the check a name and instructions, choose its schedule, and select Save check.

The mailbox needs Search messages and Read message set to Allow. Ask first cannot authorize an unattended check. The editor shows unavailable connections and their permission reasons. Creating a check grants no new connector access.

New checks use the existing daily default. The existing cadence choices and 5-minute minimum stay the same. Saving checks the mailbox before activation; keep the editor open while Saving... appears. If this check fails, correct the connection before saving again.

Only messages received after successful activation count as new mail. Existing messages stay available for manual review. New messages start agent work with the saved instructions. No new messages means no agent run or chat notification. Outgoing drafts do not trigger incoming work, and reading or marking mail read does not change check progress.

The row shows Last check, its result, and the next scheduled time. Pause and resume keep progress, including messages received while paused. Agent pauses and budget limits still apply. If budget is unavailable, new messages remain eligible for a later check.

Connection failures retry on the existing schedule. Repeated failures pause the check. A permission failure pauses it immediately. Use Review connection to correct the connection or permissions, then enable the check again. Progress is kept.

If the mailbox account, watched folder, or mailbox identity changes, review the connection, delete the affected check, and create a new one. The new check starts from now. An existing mailbox check cannot switch to another mailbox or to Run instructions. Ordinary password rotation keeps progress when the mailbox identity stays the same.

Agent work can still fail after new mail is found. The existing limited retry rules apply; they do not guarantee eventual completion. Failed work remains in conversation history. A connection change during a pending retry can pause the check. Review the connection, then enable the check to keep its progress. If the mailbox identity changed, recreate the check as described above.

Editing a mailbox check’s name or instructions replaces its pending failed-run retry. Later checks keep the same mailbox progress.

How the agent manages its own wakeups

Ask the agent to create, list, change, pause, or cancel its wakeups. Pausing keeps a wakeup visible so it can be enabled again; cancelling removes it.

Asking the agent to create a wakeup with the same name as an active one replaces that wakeup's instructions and timing and restarts its run history. Its delivery destination stays the same unless you request a change. A mailbox check also keeps progress when it keeps the same connection. Retrying the same creation request does not add a duplicate.

Wakeups can deliver only to supported destinations connected to the agent. Changing a wakeup from the app requires permission to manage that agent.

You can ask the agent to manage its wakeups from d5s or a connected channel. Its ordinary schedule limits and permissions still apply.

Cost boundary

Every agent run uses the configured model. A mailbox check with no new messages does not start a model run. Frequency, accumulated context, model choice, reasoning level, sub-agents, research, and connector calls therefore affect cost. Connector availability alone does not start a call.

The agent's daily budget is a hard cap across manual and autonomous work. It resets at midnight in the agent's budget timezone. A budget skip adds at most one notice to the conversation per local day, and scheduled work resumes automatically on the next day unless the agent is separately paused.

Automation findings

When an attached automation finishes, its assigned agent can receive the result without a new human message. In Automation team, Keep informed asks for an acknowledgement unless your standing instructions already call for follow-up. Act on findings asks the agent to assess actionable results under its existing permissions and approval rules. Delivery can wait while the agent is paused, busy, or over budget. The agent can open the result when it has access. Receiving it does not authorize external messages or change ordinary approval requirements.

An agent can choose either mode for itself or stop following an automation. Stopping suppresses linked automation notifications that have not started; changing to Keep informed also suppresses unstarted attention wakeups. Already received results and running work remain in its conversation. Other scheduled wakeups are separate from these subscription settings.

Some older automation reminders were saved as ordinary agent wakeups. Those reminders need to be canceled separately under Manage agent → Autonomy → Scheduled wakeups. Stopping an automation subscription cannot identify or cancel them automatically.

Wake from incoming events

In the agent’s Webhooks settings, event triggers can start work from a merged GitHub PR, a published release, a Grafana alert, or an authenticated JSON webhook from another service. Select Sources → Add source to connect a service. Generic webhook is selected by default. Then select New trigger, choose a filter, and give the agent instructions. Set up event triggers.